Loading...
1/*
2 * Copyright (c) 2007-2012 Nicira Networks.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21#include <linux/init.h>
22#include <linux/module.h>
23#include <linux/if_arp.h>
24#include <linux/if_vlan.h>
25#include <linux/in.h>
26#include <linux/ip.h>
27#include <linux/jhash.h>
28#include <linux/delay.h>
29#include <linux/time.h>
30#include <linux/etherdevice.h>
31#include <linux/genetlink.h>
32#include <linux/kernel.h>
33#include <linux/kthread.h>
34#include <linux/mutex.h>
35#include <linux/percpu.h>
36#include <linux/rcupdate.h>
37#include <linux/tcp.h>
38#include <linux/udp.h>
39#include <linux/ethtool.h>
40#include <linux/wait.h>
41#include <asm/div64.h>
42#include <linux/highmem.h>
43#include <linux/netfilter_bridge.h>
44#include <linux/netfilter_ipv4.h>
45#include <linux/inetdevice.h>
46#include <linux/list.h>
47#include <linux/openvswitch.h>
48#include <linux/rculist.h>
49#include <linux/dmi.h>
50#include <linux/workqueue.h>
51#include <net/genetlink.h>
52
53#include "datapath.h"
54#include "flow.h"
55#include "vport-internal_dev.h"
56
57/**
58 * DOC: Locking:
59 *
60 * Writes to device state (add/remove datapath, port, set operations on vports,
61 * etc.) are protected by RTNL.
62 *
63 * Writes to other state (flow table modifications, set miscellaneous datapath
64 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
65 * genl_mutex.
66 *
67 * Reads are protected by RCU.
68 *
69 * There are a few special cases (mostly stats) that have their own
70 * synchronization but they nest under all of above and don't interact with
71 * each other.
72 */
73
74/* Global list of datapaths to enable dumping them all out.
75 * Protected by genl_mutex.
76 */
77static LIST_HEAD(dps);
78
79#define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
80static void rehash_flow_table(struct work_struct *work);
81static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
82
83static struct vport *new_vport(const struct vport_parms *);
84static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
85 const struct dp_upcall_info *);
86static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
87 const struct dp_upcall_info *);
88
89/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
90static struct datapath *get_dp(int dp_ifindex)
91{
92 struct datapath *dp = NULL;
93 struct net_device *dev;
94
95 rcu_read_lock();
96 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
97 if (dev) {
98 struct vport *vport = ovs_internal_dev_get_vport(dev);
99 if (vport)
100 dp = vport->dp;
101 }
102 rcu_read_unlock();
103
104 return dp;
105}
106
107/* Must be called with rcu_read_lock or RTNL lock. */
108const char *ovs_dp_name(const struct datapath *dp)
109{
110 struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
111 return vport->ops->get_name(vport);
112}
113
114static int get_dpifindex(struct datapath *dp)
115{
116 struct vport *local;
117 int ifindex;
118
119 rcu_read_lock();
120
121 local = rcu_dereference(dp->ports[OVSP_LOCAL]);
122 if (local)
123 ifindex = local->ops->get_ifindex(local);
124 else
125 ifindex = 0;
126
127 rcu_read_unlock();
128
129 return ifindex;
130}
131
132static void destroy_dp_rcu(struct rcu_head *rcu)
133{
134 struct datapath *dp = container_of(rcu, struct datapath, rcu);
135
136 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
137 free_percpu(dp->stats_percpu);
138 kfree(dp);
139}
140
141/* Called with RTNL lock and genl_lock. */
142static struct vport *new_vport(const struct vport_parms *parms)
143{
144 struct vport *vport;
145
146 vport = ovs_vport_add(parms);
147 if (!IS_ERR(vport)) {
148 struct datapath *dp = parms->dp;
149
150 rcu_assign_pointer(dp->ports[parms->port_no], vport);
151 list_add(&vport->node, &dp->port_list);
152 }
153
154 return vport;
155}
156
157/* Called with RTNL lock. */
158void ovs_dp_detach_port(struct vport *p)
159{
160 ASSERT_RTNL();
161
162 /* First drop references to device. */
163 list_del(&p->node);
164 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
165
166 /* Then destroy it. */
167 ovs_vport_del(p);
168}
169
170/* Must be called with rcu_read_lock. */
171void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
172{
173 struct datapath *dp = p->dp;
174 struct sw_flow *flow;
175 struct dp_stats_percpu *stats;
176 struct sw_flow_key key;
177 u64 *stats_counter;
178 int error;
179 int key_len;
180
181 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
182
183 /* Extract flow from 'skb' into 'key'. */
184 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
185 if (unlikely(error)) {
186 kfree_skb(skb);
187 return;
188 }
189
190 /* Look up flow. */
191 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
192 if (unlikely(!flow)) {
193 struct dp_upcall_info upcall;
194
195 upcall.cmd = OVS_PACKET_CMD_MISS;
196 upcall.key = &key;
197 upcall.userdata = NULL;
198 upcall.pid = p->upcall_pid;
199 ovs_dp_upcall(dp, skb, &upcall);
200 consume_skb(skb);
201 stats_counter = &stats->n_missed;
202 goto out;
203 }
204
205 OVS_CB(skb)->flow = flow;
206
207 stats_counter = &stats->n_hit;
208 ovs_flow_used(OVS_CB(skb)->flow, skb);
209 ovs_execute_actions(dp, skb);
210
211out:
212 /* Update datapath statistics. */
213 u64_stats_update_begin(&stats->sync);
214 (*stats_counter)++;
215 u64_stats_update_end(&stats->sync);
216}
217
218static struct genl_family dp_packet_genl_family = {
219 .id = GENL_ID_GENERATE,
220 .hdrsize = sizeof(struct ovs_header),
221 .name = OVS_PACKET_FAMILY,
222 .version = OVS_PACKET_VERSION,
223 .maxattr = OVS_PACKET_ATTR_MAX
224};
225
226int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
227 const struct dp_upcall_info *upcall_info)
228{
229 struct dp_stats_percpu *stats;
230 int dp_ifindex;
231 int err;
232
233 if (upcall_info->pid == 0) {
234 err = -ENOTCONN;
235 goto err;
236 }
237
238 dp_ifindex = get_dpifindex(dp);
239 if (!dp_ifindex) {
240 err = -ENODEV;
241 goto err;
242 }
243
244 if (!skb_is_gso(skb))
245 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
246 else
247 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
248 if (err)
249 goto err;
250
251 return 0;
252
253err:
254 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
255
256 u64_stats_update_begin(&stats->sync);
257 stats->n_lost++;
258 u64_stats_update_end(&stats->sync);
259
260 return err;
261}
262
263static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
264 const struct dp_upcall_info *upcall_info)
265{
266 struct dp_upcall_info later_info;
267 struct sw_flow_key later_key;
268 struct sk_buff *segs, *nskb;
269 int err;
270
271 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
272 if (IS_ERR(skb))
273 return PTR_ERR(skb);
274
275 /* Queue all of the segments. */
276 skb = segs;
277 do {
278 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
279 if (err)
280 break;
281
282 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
283 /* The initial flow key extracted by ovs_flow_extract()
284 * in this case is for a first fragment, so we need to
285 * properly mark later fragments.
286 */
287 later_key = *upcall_info->key;
288 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
289
290 later_info = *upcall_info;
291 later_info.key = &later_key;
292 upcall_info = &later_info;
293 }
294 } while ((skb = skb->next));
295
296 /* Free all of the segments. */
297 skb = segs;
298 do {
299 nskb = skb->next;
300 if (err)
301 kfree_skb(skb);
302 else
303 consume_skb(skb);
304 } while ((skb = nskb));
305 return err;
306}
307
308static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
309 const struct dp_upcall_info *upcall_info)
310{
311 struct ovs_header *upcall;
312 struct sk_buff *nskb = NULL;
313 struct sk_buff *user_skb; /* to be queued to userspace */
314 struct nlattr *nla;
315 unsigned int len;
316 int err;
317
318 if (vlan_tx_tag_present(skb)) {
319 nskb = skb_clone(skb, GFP_ATOMIC);
320 if (!nskb)
321 return -ENOMEM;
322
323 nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
324 if (!nskb)
325 return -ENOMEM;
326
327 nskb->vlan_tci = 0;
328 skb = nskb;
329 }
330
331 if (nla_attr_size(skb->len) > USHRT_MAX) {
332 err = -EFBIG;
333 goto out;
334 }
335
336 len = sizeof(struct ovs_header);
337 len += nla_total_size(skb->len);
338 len += nla_total_size(FLOW_BUFSIZE);
339 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
340 len += nla_total_size(8);
341
342 user_skb = genlmsg_new(len, GFP_ATOMIC);
343 if (!user_skb) {
344 err = -ENOMEM;
345 goto out;
346 }
347
348 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
349 0, upcall_info->cmd);
350 upcall->dp_ifindex = dp_ifindex;
351
352 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
353 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
354 nla_nest_end(user_skb, nla);
355
356 if (upcall_info->userdata)
357 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
358 nla_get_u64(upcall_info->userdata));
359
360 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
361
362 skb_copy_and_csum_dev(skb, nla_data(nla));
363
364 err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
365
366out:
367 kfree_skb(nskb);
368 return err;
369}
370
371/* Called with genl_mutex. */
372static int flush_flows(int dp_ifindex)
373{
374 struct flow_table *old_table;
375 struct flow_table *new_table;
376 struct datapath *dp;
377
378 dp = get_dp(dp_ifindex);
379 if (!dp)
380 return -ENODEV;
381
382 old_table = genl_dereference(dp->table);
383 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
384 if (!new_table)
385 return -ENOMEM;
386
387 rcu_assign_pointer(dp->table, new_table);
388
389 ovs_flow_tbl_deferred_destroy(old_table);
390 return 0;
391}
392
393static int validate_actions(const struct nlattr *attr,
394 const struct sw_flow_key *key, int depth);
395
396static int validate_sample(const struct nlattr *attr,
397 const struct sw_flow_key *key, int depth)
398{
399 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
400 const struct nlattr *probability, *actions;
401 const struct nlattr *a;
402 int rem;
403
404 memset(attrs, 0, sizeof(attrs));
405 nla_for_each_nested(a, attr, rem) {
406 int type = nla_type(a);
407 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
408 return -EINVAL;
409 attrs[type] = a;
410 }
411 if (rem)
412 return -EINVAL;
413
414 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
415 if (!probability || nla_len(probability) != sizeof(u32))
416 return -EINVAL;
417
418 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
419 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
420 return -EINVAL;
421 return validate_actions(actions, key, depth + 1);
422}
423
424static int validate_tp_port(const struct sw_flow_key *flow_key)
425{
426 if (flow_key->eth.type == htons(ETH_P_IP)) {
427 if (flow_key->ipv4.tp.src && flow_key->ipv4.tp.dst)
428 return 0;
429 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
430 if (flow_key->ipv6.tp.src && flow_key->ipv6.tp.dst)
431 return 0;
432 }
433
434 return -EINVAL;
435}
436
437static int validate_set(const struct nlattr *a,
438 const struct sw_flow_key *flow_key)
439{
440 const struct nlattr *ovs_key = nla_data(a);
441 int key_type = nla_type(ovs_key);
442
443 /* There can be only one key in a action */
444 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
445 return -EINVAL;
446
447 if (key_type > OVS_KEY_ATTR_MAX ||
448 nla_len(ovs_key) != ovs_key_lens[key_type])
449 return -EINVAL;
450
451 switch (key_type) {
452 const struct ovs_key_ipv4 *ipv4_key;
453
454 case OVS_KEY_ATTR_PRIORITY:
455 case OVS_KEY_ATTR_ETHERNET:
456 break;
457
458 case OVS_KEY_ATTR_IPV4:
459 if (flow_key->eth.type != htons(ETH_P_IP))
460 return -EINVAL;
461
462 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
463 return -EINVAL;
464
465 ipv4_key = nla_data(ovs_key);
466 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
467 return -EINVAL;
468
469 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
470 return -EINVAL;
471
472 break;
473
474 case OVS_KEY_ATTR_TCP:
475 if (flow_key->ip.proto != IPPROTO_TCP)
476 return -EINVAL;
477
478 return validate_tp_port(flow_key);
479
480 case OVS_KEY_ATTR_UDP:
481 if (flow_key->ip.proto != IPPROTO_UDP)
482 return -EINVAL;
483
484 return validate_tp_port(flow_key);
485
486 default:
487 return -EINVAL;
488 }
489
490 return 0;
491}
492
493static int validate_userspace(const struct nlattr *attr)
494{
495 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
496 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
497 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
498 };
499 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
500 int error;
501
502 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
503 attr, userspace_policy);
504 if (error)
505 return error;
506
507 if (!a[OVS_USERSPACE_ATTR_PID] ||
508 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
509 return -EINVAL;
510
511 return 0;
512}
513
514static int validate_actions(const struct nlattr *attr,
515 const struct sw_flow_key *key, int depth)
516{
517 const struct nlattr *a;
518 int rem, err;
519
520 if (depth >= SAMPLE_ACTION_DEPTH)
521 return -EOVERFLOW;
522
523 nla_for_each_nested(a, attr, rem) {
524 /* Expected argument lengths, (u32)-1 for variable length. */
525 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
526 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
527 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
528 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
529 [OVS_ACTION_ATTR_POP_VLAN] = 0,
530 [OVS_ACTION_ATTR_SET] = (u32)-1,
531 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
532 };
533 const struct ovs_action_push_vlan *vlan;
534 int type = nla_type(a);
535
536 if (type > OVS_ACTION_ATTR_MAX ||
537 (action_lens[type] != nla_len(a) &&
538 action_lens[type] != (u32)-1))
539 return -EINVAL;
540
541 switch (type) {
542 case OVS_ACTION_ATTR_UNSPEC:
543 return -EINVAL;
544
545 case OVS_ACTION_ATTR_USERSPACE:
546 err = validate_userspace(a);
547 if (err)
548 return err;
549 break;
550
551 case OVS_ACTION_ATTR_OUTPUT:
552 if (nla_get_u32(a) >= DP_MAX_PORTS)
553 return -EINVAL;
554 break;
555
556
557 case OVS_ACTION_ATTR_POP_VLAN:
558 break;
559
560 case OVS_ACTION_ATTR_PUSH_VLAN:
561 vlan = nla_data(a);
562 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
563 return -EINVAL;
564 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
565 return -EINVAL;
566 break;
567
568 case OVS_ACTION_ATTR_SET:
569 err = validate_set(a, key);
570 if (err)
571 return err;
572 break;
573
574 case OVS_ACTION_ATTR_SAMPLE:
575 err = validate_sample(a, key, depth);
576 if (err)
577 return err;
578 break;
579
580 default:
581 return -EINVAL;
582 }
583 }
584
585 if (rem > 0)
586 return -EINVAL;
587
588 return 0;
589}
590
591static void clear_stats(struct sw_flow *flow)
592{
593 flow->used = 0;
594 flow->tcp_flags = 0;
595 flow->packet_count = 0;
596 flow->byte_count = 0;
597}
598
599static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
600{
601 struct ovs_header *ovs_header = info->userhdr;
602 struct nlattr **a = info->attrs;
603 struct sw_flow_actions *acts;
604 struct sk_buff *packet;
605 struct sw_flow *flow;
606 struct datapath *dp;
607 struct ethhdr *eth;
608 int len;
609 int err;
610 int key_len;
611
612 err = -EINVAL;
613 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
614 !a[OVS_PACKET_ATTR_ACTIONS] ||
615 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
616 goto err;
617
618 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
619 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
620 err = -ENOMEM;
621 if (!packet)
622 goto err;
623 skb_reserve(packet, NET_IP_ALIGN);
624
625 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
626
627 skb_reset_mac_header(packet);
628 eth = eth_hdr(packet);
629
630 /* Normally, setting the skb 'protocol' field would be handled by a
631 * call to eth_type_trans(), but it assumes there's a sending
632 * device, which we may not have. */
633 if (ntohs(eth->h_proto) >= 1536)
634 packet->protocol = eth->h_proto;
635 else
636 packet->protocol = htons(ETH_P_802_2);
637
638 /* Build an sw_flow for sending this packet. */
639 flow = ovs_flow_alloc();
640 err = PTR_ERR(flow);
641 if (IS_ERR(flow))
642 goto err_kfree_skb;
643
644 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
645 if (err)
646 goto err_flow_free;
647
648 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
649 &flow->key.phy.in_port,
650 a[OVS_PACKET_ATTR_KEY]);
651 if (err)
652 goto err_flow_free;
653
654 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
655 if (err)
656 goto err_flow_free;
657
658 flow->hash = ovs_flow_hash(&flow->key, key_len);
659
660 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
661 err = PTR_ERR(acts);
662 if (IS_ERR(acts))
663 goto err_flow_free;
664 rcu_assign_pointer(flow->sf_acts, acts);
665
666 OVS_CB(packet)->flow = flow;
667 packet->priority = flow->key.phy.priority;
668
669 rcu_read_lock();
670 dp = get_dp(ovs_header->dp_ifindex);
671 err = -ENODEV;
672 if (!dp)
673 goto err_unlock;
674
675 local_bh_disable();
676 err = ovs_execute_actions(dp, packet);
677 local_bh_enable();
678 rcu_read_unlock();
679
680 ovs_flow_free(flow);
681 return err;
682
683err_unlock:
684 rcu_read_unlock();
685err_flow_free:
686 ovs_flow_free(flow);
687err_kfree_skb:
688 kfree_skb(packet);
689err:
690 return err;
691}
692
693static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
694 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
695 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
696 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
697};
698
699static struct genl_ops dp_packet_genl_ops[] = {
700 { .cmd = OVS_PACKET_CMD_EXECUTE,
701 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
702 .policy = packet_policy,
703 .doit = ovs_packet_cmd_execute
704 }
705};
706
707static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
708{
709 int i;
710 struct flow_table *table = genl_dereference(dp->table);
711
712 stats->n_flows = ovs_flow_tbl_count(table);
713
714 stats->n_hit = stats->n_missed = stats->n_lost = 0;
715 for_each_possible_cpu(i) {
716 const struct dp_stats_percpu *percpu_stats;
717 struct dp_stats_percpu local_stats;
718 unsigned int start;
719
720 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
721
722 do {
723 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
724 local_stats = *percpu_stats;
725 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
726
727 stats->n_hit += local_stats.n_hit;
728 stats->n_missed += local_stats.n_missed;
729 stats->n_lost += local_stats.n_lost;
730 }
731}
732
733static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
734 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
735 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
736 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
737};
738
739static struct genl_family dp_flow_genl_family = {
740 .id = GENL_ID_GENERATE,
741 .hdrsize = sizeof(struct ovs_header),
742 .name = OVS_FLOW_FAMILY,
743 .version = OVS_FLOW_VERSION,
744 .maxattr = OVS_FLOW_ATTR_MAX
745};
746
747static struct genl_multicast_group ovs_dp_flow_multicast_group = {
748 .name = OVS_FLOW_MCGROUP
749};
750
751/* Called with genl_lock. */
752static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
753 struct sk_buff *skb, u32 pid,
754 u32 seq, u32 flags, u8 cmd)
755{
756 const int skb_orig_len = skb->len;
757 const struct sw_flow_actions *sf_acts;
758 struct ovs_flow_stats stats;
759 struct ovs_header *ovs_header;
760 struct nlattr *nla;
761 unsigned long used;
762 u8 tcp_flags;
763 int err;
764
765 sf_acts = rcu_dereference_protected(flow->sf_acts,
766 lockdep_genl_is_held());
767
768 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
769 if (!ovs_header)
770 return -EMSGSIZE;
771
772 ovs_header->dp_ifindex = get_dpifindex(dp);
773
774 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
775 if (!nla)
776 goto nla_put_failure;
777 err = ovs_flow_to_nlattrs(&flow->key, skb);
778 if (err)
779 goto error;
780 nla_nest_end(skb, nla);
781
782 spin_lock_bh(&flow->lock);
783 used = flow->used;
784 stats.n_packets = flow->packet_count;
785 stats.n_bytes = flow->byte_count;
786 tcp_flags = flow->tcp_flags;
787 spin_unlock_bh(&flow->lock);
788
789 if (used &&
790 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
791 goto nla_put_failure;
792
793 if (stats.n_packets &&
794 nla_put(skb, OVS_FLOW_ATTR_STATS,
795 sizeof(struct ovs_flow_stats), &stats))
796 goto nla_put_failure;
797
798 if (tcp_flags &&
799 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
800 goto nla_put_failure;
801
802 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
803 * this is the first flow to be dumped into 'skb'. This is unusual for
804 * Netlink but individual action lists can be longer than
805 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
806 * The userspace caller can always fetch the actions separately if it
807 * really wants them. (Most userspace callers in fact don't care.)
808 *
809 * This can only fail for dump operations because the skb is always
810 * properly sized for single flows.
811 */
812 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
813 sf_acts->actions);
814 if (err < 0 && skb_orig_len)
815 goto error;
816
817 return genlmsg_end(skb, ovs_header);
818
819nla_put_failure:
820 err = -EMSGSIZE;
821error:
822 genlmsg_cancel(skb, ovs_header);
823 return err;
824}
825
826static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
827{
828 const struct sw_flow_actions *sf_acts;
829 int len;
830
831 sf_acts = rcu_dereference_protected(flow->sf_acts,
832 lockdep_genl_is_held());
833
834 /* OVS_FLOW_ATTR_KEY */
835 len = nla_total_size(FLOW_BUFSIZE);
836 /* OVS_FLOW_ATTR_ACTIONS */
837 len += nla_total_size(sf_acts->actions_len);
838 /* OVS_FLOW_ATTR_STATS */
839 len += nla_total_size(sizeof(struct ovs_flow_stats));
840 /* OVS_FLOW_ATTR_TCP_FLAGS */
841 len += nla_total_size(1);
842 /* OVS_FLOW_ATTR_USED */
843 len += nla_total_size(8);
844
845 len += NLMSG_ALIGN(sizeof(struct ovs_header));
846
847 return genlmsg_new(len, GFP_KERNEL);
848}
849
850static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
851 struct datapath *dp,
852 u32 pid, u32 seq, u8 cmd)
853{
854 struct sk_buff *skb;
855 int retval;
856
857 skb = ovs_flow_cmd_alloc_info(flow);
858 if (!skb)
859 return ERR_PTR(-ENOMEM);
860
861 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
862 BUG_ON(retval < 0);
863 return skb;
864}
865
866static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
867{
868 struct nlattr **a = info->attrs;
869 struct ovs_header *ovs_header = info->userhdr;
870 struct sw_flow_key key;
871 struct sw_flow *flow;
872 struct sk_buff *reply;
873 struct datapath *dp;
874 struct flow_table *table;
875 int error;
876 int key_len;
877
878 /* Extract key. */
879 error = -EINVAL;
880 if (!a[OVS_FLOW_ATTR_KEY])
881 goto error;
882 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
883 if (error)
884 goto error;
885
886 /* Validate actions. */
887 if (a[OVS_FLOW_ATTR_ACTIONS]) {
888 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
889 if (error)
890 goto error;
891 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
892 error = -EINVAL;
893 goto error;
894 }
895
896 dp = get_dp(ovs_header->dp_ifindex);
897 error = -ENODEV;
898 if (!dp)
899 goto error;
900
901 table = genl_dereference(dp->table);
902 flow = ovs_flow_tbl_lookup(table, &key, key_len);
903 if (!flow) {
904 struct sw_flow_actions *acts;
905
906 /* Bail out if we're not allowed to create a new flow. */
907 error = -ENOENT;
908 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
909 goto error;
910
911 /* Expand table, if necessary, to make room. */
912 if (ovs_flow_tbl_need_to_expand(table)) {
913 struct flow_table *new_table;
914
915 new_table = ovs_flow_tbl_expand(table);
916 if (!IS_ERR(new_table)) {
917 rcu_assign_pointer(dp->table, new_table);
918 ovs_flow_tbl_deferred_destroy(table);
919 table = genl_dereference(dp->table);
920 }
921 }
922
923 /* Allocate flow. */
924 flow = ovs_flow_alloc();
925 if (IS_ERR(flow)) {
926 error = PTR_ERR(flow);
927 goto error;
928 }
929 flow->key = key;
930 clear_stats(flow);
931
932 /* Obtain actions. */
933 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
934 error = PTR_ERR(acts);
935 if (IS_ERR(acts))
936 goto error_free_flow;
937 rcu_assign_pointer(flow->sf_acts, acts);
938
939 /* Put flow in bucket. */
940 flow->hash = ovs_flow_hash(&key, key_len);
941 ovs_flow_tbl_insert(table, flow);
942
943 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
944 info->snd_seq,
945 OVS_FLOW_CMD_NEW);
946 } else {
947 /* We found a matching flow. */
948 struct sw_flow_actions *old_acts;
949 struct nlattr *acts_attrs;
950
951 /* Bail out if we're not allowed to modify an existing flow.
952 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
953 * because Generic Netlink treats the latter as a dump
954 * request. We also accept NLM_F_EXCL in case that bug ever
955 * gets fixed.
956 */
957 error = -EEXIST;
958 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
959 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
960 goto error;
961
962 /* Update actions. */
963 old_acts = rcu_dereference_protected(flow->sf_acts,
964 lockdep_genl_is_held());
965 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
966 if (acts_attrs &&
967 (old_acts->actions_len != nla_len(acts_attrs) ||
968 memcmp(old_acts->actions, nla_data(acts_attrs),
969 old_acts->actions_len))) {
970 struct sw_flow_actions *new_acts;
971
972 new_acts = ovs_flow_actions_alloc(acts_attrs);
973 error = PTR_ERR(new_acts);
974 if (IS_ERR(new_acts))
975 goto error;
976
977 rcu_assign_pointer(flow->sf_acts, new_acts);
978 ovs_flow_deferred_free_acts(old_acts);
979 }
980
981 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
982 info->snd_seq, OVS_FLOW_CMD_NEW);
983
984 /* Clear stats. */
985 if (a[OVS_FLOW_ATTR_CLEAR]) {
986 spin_lock_bh(&flow->lock);
987 clear_stats(flow);
988 spin_unlock_bh(&flow->lock);
989 }
990 }
991
992 if (!IS_ERR(reply))
993 genl_notify(reply, genl_info_net(info), info->snd_pid,
994 ovs_dp_flow_multicast_group.id, info->nlhdr,
995 GFP_KERNEL);
996 else
997 netlink_set_err(init_net.genl_sock, 0,
998 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
999 return 0;
1000
1001error_free_flow:
1002 ovs_flow_free(flow);
1003error:
1004 return error;
1005}
1006
1007static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1008{
1009 struct nlattr **a = info->attrs;
1010 struct ovs_header *ovs_header = info->userhdr;
1011 struct sw_flow_key key;
1012 struct sk_buff *reply;
1013 struct sw_flow *flow;
1014 struct datapath *dp;
1015 struct flow_table *table;
1016 int err;
1017 int key_len;
1018
1019 if (!a[OVS_FLOW_ATTR_KEY])
1020 return -EINVAL;
1021 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1022 if (err)
1023 return err;
1024
1025 dp = get_dp(ovs_header->dp_ifindex);
1026 if (!dp)
1027 return -ENODEV;
1028
1029 table = genl_dereference(dp->table);
1030 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1031 if (!flow)
1032 return -ENOENT;
1033
1034 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1035 info->snd_seq, OVS_FLOW_CMD_NEW);
1036 if (IS_ERR(reply))
1037 return PTR_ERR(reply);
1038
1039 return genlmsg_reply(reply, info);
1040}
1041
1042static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1043{
1044 struct nlattr **a = info->attrs;
1045 struct ovs_header *ovs_header = info->userhdr;
1046 struct sw_flow_key key;
1047 struct sk_buff *reply;
1048 struct sw_flow *flow;
1049 struct datapath *dp;
1050 struct flow_table *table;
1051 int err;
1052 int key_len;
1053
1054 if (!a[OVS_FLOW_ATTR_KEY])
1055 return flush_flows(ovs_header->dp_ifindex);
1056 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1057 if (err)
1058 return err;
1059
1060 dp = get_dp(ovs_header->dp_ifindex);
1061 if (!dp)
1062 return -ENODEV;
1063
1064 table = genl_dereference(dp->table);
1065 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1066 if (!flow)
1067 return -ENOENT;
1068
1069 reply = ovs_flow_cmd_alloc_info(flow);
1070 if (!reply)
1071 return -ENOMEM;
1072
1073 ovs_flow_tbl_remove(table, flow);
1074
1075 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1076 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1077 BUG_ON(err < 0);
1078
1079 ovs_flow_deferred_free(flow);
1080
1081 genl_notify(reply, genl_info_net(info), info->snd_pid,
1082 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1083 return 0;
1084}
1085
1086static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1087{
1088 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1089 struct datapath *dp;
1090 struct flow_table *table;
1091
1092 dp = get_dp(ovs_header->dp_ifindex);
1093 if (!dp)
1094 return -ENODEV;
1095
1096 table = genl_dereference(dp->table);
1097
1098 for (;;) {
1099 struct sw_flow *flow;
1100 u32 bucket, obj;
1101
1102 bucket = cb->args[0];
1103 obj = cb->args[1];
1104 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1105 if (!flow)
1106 break;
1107
1108 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1109 NETLINK_CB(cb->skb).pid,
1110 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1111 OVS_FLOW_CMD_NEW) < 0)
1112 break;
1113
1114 cb->args[0] = bucket;
1115 cb->args[1] = obj;
1116 }
1117 return skb->len;
1118}
1119
1120static struct genl_ops dp_flow_genl_ops[] = {
1121 { .cmd = OVS_FLOW_CMD_NEW,
1122 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1123 .policy = flow_policy,
1124 .doit = ovs_flow_cmd_new_or_set
1125 },
1126 { .cmd = OVS_FLOW_CMD_DEL,
1127 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1128 .policy = flow_policy,
1129 .doit = ovs_flow_cmd_del
1130 },
1131 { .cmd = OVS_FLOW_CMD_GET,
1132 .flags = 0, /* OK for unprivileged users. */
1133 .policy = flow_policy,
1134 .doit = ovs_flow_cmd_get,
1135 .dumpit = ovs_flow_cmd_dump
1136 },
1137 { .cmd = OVS_FLOW_CMD_SET,
1138 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1139 .policy = flow_policy,
1140 .doit = ovs_flow_cmd_new_or_set,
1141 },
1142};
1143
1144static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1145 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1146 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1147};
1148
1149static struct genl_family dp_datapath_genl_family = {
1150 .id = GENL_ID_GENERATE,
1151 .hdrsize = sizeof(struct ovs_header),
1152 .name = OVS_DATAPATH_FAMILY,
1153 .version = OVS_DATAPATH_VERSION,
1154 .maxattr = OVS_DP_ATTR_MAX
1155};
1156
1157static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1158 .name = OVS_DATAPATH_MCGROUP
1159};
1160
1161static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1162 u32 pid, u32 seq, u32 flags, u8 cmd)
1163{
1164 struct ovs_header *ovs_header;
1165 struct ovs_dp_stats dp_stats;
1166 int err;
1167
1168 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1169 flags, cmd);
1170 if (!ovs_header)
1171 goto error;
1172
1173 ovs_header->dp_ifindex = get_dpifindex(dp);
1174
1175 rcu_read_lock();
1176 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1177 rcu_read_unlock();
1178 if (err)
1179 goto nla_put_failure;
1180
1181 get_dp_stats(dp, &dp_stats);
1182 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1183 goto nla_put_failure;
1184
1185 return genlmsg_end(skb, ovs_header);
1186
1187nla_put_failure:
1188 genlmsg_cancel(skb, ovs_header);
1189error:
1190 return -EMSGSIZE;
1191}
1192
1193static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1194 u32 seq, u8 cmd)
1195{
1196 struct sk_buff *skb;
1197 int retval;
1198
1199 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1200 if (!skb)
1201 return ERR_PTR(-ENOMEM);
1202
1203 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1204 if (retval < 0) {
1205 kfree_skb(skb);
1206 return ERR_PTR(retval);
1207 }
1208 return skb;
1209}
1210
1211/* Called with genl_mutex and optionally with RTNL lock also. */
1212static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1213 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1214{
1215 struct datapath *dp;
1216
1217 if (!a[OVS_DP_ATTR_NAME])
1218 dp = get_dp(ovs_header->dp_ifindex);
1219 else {
1220 struct vport *vport;
1221
1222 rcu_read_lock();
1223 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1224 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1225 rcu_read_unlock();
1226 }
1227 return dp ? dp : ERR_PTR(-ENODEV);
1228}
1229
1230static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1231{
1232 struct nlattr **a = info->attrs;
1233 struct vport_parms parms;
1234 struct sk_buff *reply;
1235 struct datapath *dp;
1236 struct vport *vport;
1237 int err;
1238
1239 err = -EINVAL;
1240 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1241 goto err;
1242
1243 rtnl_lock();
1244 err = -ENODEV;
1245 if (!try_module_get(THIS_MODULE))
1246 goto err_unlock_rtnl;
1247
1248 err = -ENOMEM;
1249 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1250 if (dp == NULL)
1251 goto err_put_module;
1252 INIT_LIST_HEAD(&dp->port_list);
1253
1254 /* Allocate table. */
1255 err = -ENOMEM;
1256 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1257 if (!dp->table)
1258 goto err_free_dp;
1259
1260 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1261 if (!dp->stats_percpu) {
1262 err = -ENOMEM;
1263 goto err_destroy_table;
1264 }
1265
1266 /* Set up our datapath device. */
1267 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1268 parms.type = OVS_VPORT_TYPE_INTERNAL;
1269 parms.options = NULL;
1270 parms.dp = dp;
1271 parms.port_no = OVSP_LOCAL;
1272 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1273
1274 vport = new_vport(&parms);
1275 if (IS_ERR(vport)) {
1276 err = PTR_ERR(vport);
1277 if (err == -EBUSY)
1278 err = -EEXIST;
1279
1280 goto err_destroy_percpu;
1281 }
1282
1283 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1284 info->snd_seq, OVS_DP_CMD_NEW);
1285 err = PTR_ERR(reply);
1286 if (IS_ERR(reply))
1287 goto err_destroy_local_port;
1288
1289 list_add_tail(&dp->list_node, &dps);
1290 rtnl_unlock();
1291
1292 genl_notify(reply, genl_info_net(info), info->snd_pid,
1293 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1294 GFP_KERNEL);
1295 return 0;
1296
1297err_destroy_local_port:
1298 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1299err_destroy_percpu:
1300 free_percpu(dp->stats_percpu);
1301err_destroy_table:
1302 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1303err_free_dp:
1304 kfree(dp);
1305err_put_module:
1306 module_put(THIS_MODULE);
1307err_unlock_rtnl:
1308 rtnl_unlock();
1309err:
1310 return err;
1311}
1312
1313static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1314{
1315 struct vport *vport, *next_vport;
1316 struct sk_buff *reply;
1317 struct datapath *dp;
1318 int err;
1319
1320 rtnl_lock();
1321 dp = lookup_datapath(info->userhdr, info->attrs);
1322 err = PTR_ERR(dp);
1323 if (IS_ERR(dp))
1324 goto exit_unlock;
1325
1326 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1327 info->snd_seq, OVS_DP_CMD_DEL);
1328 err = PTR_ERR(reply);
1329 if (IS_ERR(reply))
1330 goto exit_unlock;
1331
1332 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1333 if (vport->port_no != OVSP_LOCAL)
1334 ovs_dp_detach_port(vport);
1335
1336 list_del(&dp->list_node);
1337 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1338
1339 /* rtnl_unlock() will wait until all the references to devices that
1340 * are pending unregistration have been dropped. We do it here to
1341 * ensure that any internal devices (which contain DP pointers) are
1342 * fully destroyed before freeing the datapath.
1343 */
1344 rtnl_unlock();
1345
1346 call_rcu(&dp->rcu, destroy_dp_rcu);
1347 module_put(THIS_MODULE);
1348
1349 genl_notify(reply, genl_info_net(info), info->snd_pid,
1350 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1351 GFP_KERNEL);
1352
1353 return 0;
1354
1355exit_unlock:
1356 rtnl_unlock();
1357 return err;
1358}
1359
1360static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1361{
1362 struct sk_buff *reply;
1363 struct datapath *dp;
1364 int err;
1365
1366 dp = lookup_datapath(info->userhdr, info->attrs);
1367 if (IS_ERR(dp))
1368 return PTR_ERR(dp);
1369
1370 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1371 info->snd_seq, OVS_DP_CMD_NEW);
1372 if (IS_ERR(reply)) {
1373 err = PTR_ERR(reply);
1374 netlink_set_err(init_net.genl_sock, 0,
1375 ovs_dp_datapath_multicast_group.id, err);
1376 return 0;
1377 }
1378
1379 genl_notify(reply, genl_info_net(info), info->snd_pid,
1380 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1381 GFP_KERNEL);
1382
1383 return 0;
1384}
1385
1386static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1387{
1388 struct sk_buff *reply;
1389 struct datapath *dp;
1390
1391 dp = lookup_datapath(info->userhdr, info->attrs);
1392 if (IS_ERR(dp))
1393 return PTR_ERR(dp);
1394
1395 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1396 info->snd_seq, OVS_DP_CMD_NEW);
1397 if (IS_ERR(reply))
1398 return PTR_ERR(reply);
1399
1400 return genlmsg_reply(reply, info);
1401}
1402
1403static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1404{
1405 struct datapath *dp;
1406 int skip = cb->args[0];
1407 int i = 0;
1408
1409 list_for_each_entry(dp, &dps, list_node) {
1410 if (i >= skip &&
1411 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1412 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1413 OVS_DP_CMD_NEW) < 0)
1414 break;
1415 i++;
1416 }
1417
1418 cb->args[0] = i;
1419
1420 return skb->len;
1421}
1422
1423static struct genl_ops dp_datapath_genl_ops[] = {
1424 { .cmd = OVS_DP_CMD_NEW,
1425 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1426 .policy = datapath_policy,
1427 .doit = ovs_dp_cmd_new
1428 },
1429 { .cmd = OVS_DP_CMD_DEL,
1430 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1431 .policy = datapath_policy,
1432 .doit = ovs_dp_cmd_del
1433 },
1434 { .cmd = OVS_DP_CMD_GET,
1435 .flags = 0, /* OK for unprivileged users. */
1436 .policy = datapath_policy,
1437 .doit = ovs_dp_cmd_get,
1438 .dumpit = ovs_dp_cmd_dump
1439 },
1440 { .cmd = OVS_DP_CMD_SET,
1441 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1442 .policy = datapath_policy,
1443 .doit = ovs_dp_cmd_set,
1444 },
1445};
1446
1447static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1448 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1449 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1450 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1451 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1452 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1453 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1454};
1455
1456static struct genl_family dp_vport_genl_family = {
1457 .id = GENL_ID_GENERATE,
1458 .hdrsize = sizeof(struct ovs_header),
1459 .name = OVS_VPORT_FAMILY,
1460 .version = OVS_VPORT_VERSION,
1461 .maxattr = OVS_VPORT_ATTR_MAX
1462};
1463
1464struct genl_multicast_group ovs_dp_vport_multicast_group = {
1465 .name = OVS_VPORT_MCGROUP
1466};
1467
1468/* Called with RTNL lock or RCU read lock. */
1469static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1470 u32 pid, u32 seq, u32 flags, u8 cmd)
1471{
1472 struct ovs_header *ovs_header;
1473 struct ovs_vport_stats vport_stats;
1474 int err;
1475
1476 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1477 flags, cmd);
1478 if (!ovs_header)
1479 return -EMSGSIZE;
1480
1481 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1482
1483 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1484 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1485 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1486 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1487 goto nla_put_failure;
1488
1489 ovs_vport_get_stats(vport, &vport_stats);
1490 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1491 &vport_stats))
1492 goto nla_put_failure;
1493
1494 err = ovs_vport_get_options(vport, skb);
1495 if (err == -EMSGSIZE)
1496 goto error;
1497
1498 return genlmsg_end(skb, ovs_header);
1499
1500nla_put_failure:
1501 err = -EMSGSIZE;
1502error:
1503 genlmsg_cancel(skb, ovs_header);
1504 return err;
1505}
1506
1507/* Called with RTNL lock or RCU read lock. */
1508struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1509 u32 seq, u8 cmd)
1510{
1511 struct sk_buff *skb;
1512 int retval;
1513
1514 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1515 if (!skb)
1516 return ERR_PTR(-ENOMEM);
1517
1518 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1519 if (retval < 0) {
1520 kfree_skb(skb);
1521 return ERR_PTR(retval);
1522 }
1523 return skb;
1524}
1525
1526/* Called with RTNL lock or RCU read lock. */
1527static struct vport *lookup_vport(struct ovs_header *ovs_header,
1528 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1529{
1530 struct datapath *dp;
1531 struct vport *vport;
1532
1533 if (a[OVS_VPORT_ATTR_NAME]) {
1534 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1535 if (!vport)
1536 return ERR_PTR(-ENODEV);
1537 if (ovs_header->dp_ifindex &&
1538 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1539 return ERR_PTR(-ENODEV);
1540 return vport;
1541 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1542 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1543
1544 if (port_no >= DP_MAX_PORTS)
1545 return ERR_PTR(-EFBIG);
1546
1547 dp = get_dp(ovs_header->dp_ifindex);
1548 if (!dp)
1549 return ERR_PTR(-ENODEV);
1550
1551 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1552 if (!vport)
1553 return ERR_PTR(-ENOENT);
1554 return vport;
1555 } else
1556 return ERR_PTR(-EINVAL);
1557}
1558
1559static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1560{
1561 struct nlattr **a = info->attrs;
1562 struct ovs_header *ovs_header = info->userhdr;
1563 struct vport_parms parms;
1564 struct sk_buff *reply;
1565 struct vport *vport;
1566 struct datapath *dp;
1567 u32 port_no;
1568 int err;
1569
1570 err = -EINVAL;
1571 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1572 !a[OVS_VPORT_ATTR_UPCALL_PID])
1573 goto exit;
1574
1575 rtnl_lock();
1576 dp = get_dp(ovs_header->dp_ifindex);
1577 err = -ENODEV;
1578 if (!dp)
1579 goto exit_unlock;
1580
1581 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1582 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1583
1584 err = -EFBIG;
1585 if (port_no >= DP_MAX_PORTS)
1586 goto exit_unlock;
1587
1588 vport = rtnl_dereference(dp->ports[port_no]);
1589 err = -EBUSY;
1590 if (vport)
1591 goto exit_unlock;
1592 } else {
1593 for (port_no = 1; ; port_no++) {
1594 if (port_no >= DP_MAX_PORTS) {
1595 err = -EFBIG;
1596 goto exit_unlock;
1597 }
1598 vport = rtnl_dereference(dp->ports[port_no]);
1599 if (!vport)
1600 break;
1601 }
1602 }
1603
1604 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1605 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1606 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1607 parms.dp = dp;
1608 parms.port_no = port_no;
1609 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1610
1611 vport = new_vport(&parms);
1612 err = PTR_ERR(vport);
1613 if (IS_ERR(vport))
1614 goto exit_unlock;
1615
1616 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1617 OVS_VPORT_CMD_NEW);
1618 if (IS_ERR(reply)) {
1619 err = PTR_ERR(reply);
1620 ovs_dp_detach_port(vport);
1621 goto exit_unlock;
1622 }
1623 genl_notify(reply, genl_info_net(info), info->snd_pid,
1624 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1625
1626exit_unlock:
1627 rtnl_unlock();
1628exit:
1629 return err;
1630}
1631
1632static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1633{
1634 struct nlattr **a = info->attrs;
1635 struct sk_buff *reply;
1636 struct vport *vport;
1637 int err;
1638
1639 rtnl_lock();
1640 vport = lookup_vport(info->userhdr, a);
1641 err = PTR_ERR(vport);
1642 if (IS_ERR(vport))
1643 goto exit_unlock;
1644
1645 err = 0;
1646 if (a[OVS_VPORT_ATTR_TYPE] &&
1647 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1648 err = -EINVAL;
1649
1650 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1651 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1652 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1653 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1654
1655 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1656 OVS_VPORT_CMD_NEW);
1657 if (IS_ERR(reply)) {
1658 netlink_set_err(init_net.genl_sock, 0,
1659 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1660 goto exit_unlock;
1661 }
1662
1663 genl_notify(reply, genl_info_net(info), info->snd_pid,
1664 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1665
1666exit_unlock:
1667 rtnl_unlock();
1668 return err;
1669}
1670
1671static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1672{
1673 struct nlattr **a = info->attrs;
1674 struct sk_buff *reply;
1675 struct vport *vport;
1676 int err;
1677
1678 rtnl_lock();
1679 vport = lookup_vport(info->userhdr, a);
1680 err = PTR_ERR(vport);
1681 if (IS_ERR(vport))
1682 goto exit_unlock;
1683
1684 if (vport->port_no == OVSP_LOCAL) {
1685 err = -EINVAL;
1686 goto exit_unlock;
1687 }
1688
1689 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1690 OVS_VPORT_CMD_DEL);
1691 err = PTR_ERR(reply);
1692 if (IS_ERR(reply))
1693 goto exit_unlock;
1694
1695 ovs_dp_detach_port(vport);
1696
1697 genl_notify(reply, genl_info_net(info), info->snd_pid,
1698 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1699
1700exit_unlock:
1701 rtnl_unlock();
1702 return err;
1703}
1704
1705static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1706{
1707 struct nlattr **a = info->attrs;
1708 struct ovs_header *ovs_header = info->userhdr;
1709 struct sk_buff *reply;
1710 struct vport *vport;
1711 int err;
1712
1713 rcu_read_lock();
1714 vport = lookup_vport(ovs_header, a);
1715 err = PTR_ERR(vport);
1716 if (IS_ERR(vport))
1717 goto exit_unlock;
1718
1719 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1720 OVS_VPORT_CMD_NEW);
1721 err = PTR_ERR(reply);
1722 if (IS_ERR(reply))
1723 goto exit_unlock;
1724
1725 rcu_read_unlock();
1726
1727 return genlmsg_reply(reply, info);
1728
1729exit_unlock:
1730 rcu_read_unlock();
1731 return err;
1732}
1733
1734static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1735{
1736 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1737 struct datapath *dp;
1738 u32 port_no;
1739 int retval;
1740
1741 dp = get_dp(ovs_header->dp_ifindex);
1742 if (!dp)
1743 return -ENODEV;
1744
1745 rcu_read_lock();
1746 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1747 struct vport *vport;
1748
1749 vport = rcu_dereference(dp->ports[port_no]);
1750 if (!vport)
1751 continue;
1752
1753 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1754 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1755 OVS_VPORT_CMD_NEW) < 0)
1756 break;
1757 }
1758 rcu_read_unlock();
1759
1760 cb->args[0] = port_no;
1761 retval = skb->len;
1762
1763 return retval;
1764}
1765
1766static void rehash_flow_table(struct work_struct *work)
1767{
1768 struct datapath *dp;
1769
1770 genl_lock();
1771
1772 list_for_each_entry(dp, &dps, list_node) {
1773 struct flow_table *old_table = genl_dereference(dp->table);
1774 struct flow_table *new_table;
1775
1776 new_table = ovs_flow_tbl_rehash(old_table);
1777 if (!IS_ERR(new_table)) {
1778 rcu_assign_pointer(dp->table, new_table);
1779 ovs_flow_tbl_deferred_destroy(old_table);
1780 }
1781 }
1782
1783 genl_unlock();
1784
1785 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1786}
1787
1788static struct genl_ops dp_vport_genl_ops[] = {
1789 { .cmd = OVS_VPORT_CMD_NEW,
1790 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1791 .policy = vport_policy,
1792 .doit = ovs_vport_cmd_new
1793 },
1794 { .cmd = OVS_VPORT_CMD_DEL,
1795 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1796 .policy = vport_policy,
1797 .doit = ovs_vport_cmd_del
1798 },
1799 { .cmd = OVS_VPORT_CMD_GET,
1800 .flags = 0, /* OK for unprivileged users. */
1801 .policy = vport_policy,
1802 .doit = ovs_vport_cmd_get,
1803 .dumpit = ovs_vport_cmd_dump
1804 },
1805 { .cmd = OVS_VPORT_CMD_SET,
1806 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1807 .policy = vport_policy,
1808 .doit = ovs_vport_cmd_set,
1809 },
1810};
1811
1812struct genl_family_and_ops {
1813 struct genl_family *family;
1814 struct genl_ops *ops;
1815 int n_ops;
1816 struct genl_multicast_group *group;
1817};
1818
1819static const struct genl_family_and_ops dp_genl_families[] = {
1820 { &dp_datapath_genl_family,
1821 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1822 &ovs_dp_datapath_multicast_group },
1823 { &dp_vport_genl_family,
1824 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1825 &ovs_dp_vport_multicast_group },
1826 { &dp_flow_genl_family,
1827 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1828 &ovs_dp_flow_multicast_group },
1829 { &dp_packet_genl_family,
1830 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1831 NULL },
1832};
1833
1834static void dp_unregister_genl(int n_families)
1835{
1836 int i;
1837
1838 for (i = 0; i < n_families; i++)
1839 genl_unregister_family(dp_genl_families[i].family);
1840}
1841
1842static int dp_register_genl(void)
1843{
1844 int n_registered;
1845 int err;
1846 int i;
1847
1848 n_registered = 0;
1849 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1850 const struct genl_family_and_ops *f = &dp_genl_families[i];
1851
1852 err = genl_register_family_with_ops(f->family, f->ops,
1853 f->n_ops);
1854 if (err)
1855 goto error;
1856 n_registered++;
1857
1858 if (f->group) {
1859 err = genl_register_mc_group(f->family, f->group);
1860 if (err)
1861 goto error;
1862 }
1863 }
1864
1865 return 0;
1866
1867error:
1868 dp_unregister_genl(n_registered);
1869 return err;
1870}
1871
1872static int __init dp_init(void)
1873{
1874 struct sk_buff *dummy_skb;
1875 int err;
1876
1877 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1878
1879 pr_info("Open vSwitch switching datapath\n");
1880
1881 err = ovs_flow_init();
1882 if (err)
1883 goto error;
1884
1885 err = ovs_vport_init();
1886 if (err)
1887 goto error_flow_exit;
1888
1889 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1890 if (err)
1891 goto error_vport_exit;
1892
1893 err = dp_register_genl();
1894 if (err < 0)
1895 goto error_unreg_notifier;
1896
1897 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1898
1899 return 0;
1900
1901error_unreg_notifier:
1902 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1903error_vport_exit:
1904 ovs_vport_exit();
1905error_flow_exit:
1906 ovs_flow_exit();
1907error:
1908 return err;
1909}
1910
1911static void dp_cleanup(void)
1912{
1913 cancel_delayed_work_sync(&rehash_flow_wq);
1914 rcu_barrier();
1915 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1916 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1917 ovs_vport_exit();
1918 ovs_flow_exit();
1919}
1920
1921module_init(dp_init);
1922module_exit(dp_cleanup);
1923
1924MODULE_DESCRIPTION("Open vSwitch switching datapath");
1925MODULE_LICENSE("GPL");
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
52unsigned int ovs_net_id __read_mostly;
53
54static struct genl_family dp_packet_genl_family;
55static struct genl_family dp_flow_genl_family;
56static struct genl_family dp_datapath_genl_family;
57
58static const struct nla_policy flow_policy[];
59
60static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
61 .name = OVS_FLOW_MCGROUP,
62};
63
64static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
65 .name = OVS_DATAPATH_MCGROUP,
66};
67
68static 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. */
74static 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
81static 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
104static DEFINE_MUTEX(ovs_mutex);
105
106void ovs_lock(void)
107{
108 mutex_lock(&ovs_mutex);
109}
110
111void ovs_unlock(void)
112{
113 mutex_unlock(&ovs_mutex);
114}
115
116#ifdef CONFIG_LOCKDEP
117int 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
126static struct vport *new_vport(const struct vport_parms *);
127static 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);
131static 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
136static void ovs_dp_masks_rebalance(struct work_struct *work);
137
138static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
139
140/* Must be called with rcu_read_lock or ovs_mutex. */
141const 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
147static 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
165static 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
177static 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. */
184struct 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. */
199static 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
213static 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
233void 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. */
245void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
246{
247 const struct vport *p = OVS_CB(skb)->input_vport;
248 struct datapath *dp = p->dp;
249 struct sw_flow *flow;
250 struct sw_flow_actions *sf_acts;
251 struct dp_stats_percpu *stats;
252 u64 *stats_counter;
253 u32 n_mask_hit;
254 u32 n_cache_hit;
255 int error;
256
257 stats = this_cpu_ptr(dp->stats_percpu);
258
259 /* Look up flow. */
260 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
261 &n_mask_hit, &n_cache_hit);
262 if (unlikely(!flow)) {
263 struct dp_upcall_info upcall;
264
265 memset(&upcall, 0, sizeof(upcall));
266 upcall.cmd = OVS_PACKET_CMD_MISS;
267
268 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
269 upcall.portid =
270 ovs_dp_get_upcall_portid(dp, smp_processor_id());
271 else
272 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
273
274 upcall.mru = OVS_CB(skb)->mru;
275 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
276 switch (error) {
277 case 0:
278 case -EAGAIN:
279 case -ERESTARTSYS:
280 case -EINTR:
281 consume_skb(skb);
282 break;
283 default:
284 kfree_skb(skb);
285 break;
286 }
287 stats_counter = &stats->n_missed;
288 goto out;
289 }
290
291 ovs_flow_stats_update(flow, key->tp.flags, skb);
292 sf_acts = rcu_dereference(flow->sf_acts);
293 error = ovs_execute_actions(dp, skb, sf_acts, key);
294 if (unlikely(error))
295 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
296 ovs_dp_name(dp), error);
297
298 stats_counter = &stats->n_hit;
299
300out:
301 /* Update datapath statistics. */
302 u64_stats_update_begin(&stats->syncp);
303 (*stats_counter)++;
304 stats->n_mask_hit += n_mask_hit;
305 stats->n_cache_hit += n_cache_hit;
306 u64_stats_update_end(&stats->syncp);
307}
308
309int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
310 const struct sw_flow_key *key,
311 const struct dp_upcall_info *upcall_info,
312 uint32_t cutlen)
313{
314 struct dp_stats_percpu *stats;
315 int err;
316
317 if (trace_ovs_dp_upcall_enabled())
318 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
319
320 if (upcall_info->portid == 0) {
321 err = -ENOTCONN;
322 goto err;
323 }
324
325 if (!skb_is_gso(skb))
326 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
327 else
328 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
329
330 ovs_vport_update_upcall_stats(skb, upcall_info, !err);
331 if (err)
332 goto err;
333
334 return 0;
335
336err:
337 stats = this_cpu_ptr(dp->stats_percpu);
338
339 u64_stats_update_begin(&stats->syncp);
340 stats->n_lost++;
341 u64_stats_update_end(&stats->syncp);
342
343 return err;
344}
345
346static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
347 const struct sw_flow_key *key,
348 const struct dp_upcall_info *upcall_info,
349 uint32_t cutlen)
350{
351 unsigned int gso_type = skb_shinfo(skb)->gso_type;
352 struct sw_flow_key later_key;
353 struct sk_buff *segs, *nskb;
354 int err;
355
356 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
357 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
358 if (IS_ERR(segs))
359 return PTR_ERR(segs);
360 if (segs == NULL)
361 return -EINVAL;
362
363 if (gso_type & SKB_GSO_UDP) {
364 /* The initial flow key extracted by ovs_flow_key_extract()
365 * in this case is for a first fragment, so we need to
366 * properly mark later fragments.
367 */
368 later_key = *key;
369 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
370 }
371
372 /* Queue all of the segments. */
373 skb_list_walk_safe(segs, skb, nskb) {
374 if (gso_type & SKB_GSO_UDP && skb != segs)
375 key = &later_key;
376
377 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
378 if (err)
379 break;
380
381 }
382
383 /* Free all of the segments. */
384 skb_list_walk_safe(segs, skb, nskb) {
385 if (err)
386 kfree_skb(skb);
387 else
388 consume_skb(skb);
389 }
390 return err;
391}
392
393static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
394 unsigned int hdrlen, int actions_attrlen)
395{
396 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
397 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
398 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
399 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
400 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
401
402 /* OVS_PACKET_ATTR_USERDATA */
403 if (upcall_info->userdata)
404 size += NLA_ALIGN(upcall_info->userdata->nla_len);
405
406 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
407 if (upcall_info->egress_tun_info)
408 size += nla_total_size(ovs_tun_key_attr_size());
409
410 /* OVS_PACKET_ATTR_ACTIONS */
411 if (upcall_info->actions_len)
412 size += nla_total_size(actions_attrlen);
413
414 /* OVS_PACKET_ATTR_MRU */
415 if (upcall_info->mru)
416 size += nla_total_size(sizeof(upcall_info->mru));
417
418 return size;
419}
420
421static void pad_packet(struct datapath *dp, struct sk_buff *skb)
422{
423 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
424 size_t plen = NLA_ALIGN(skb->len) - skb->len;
425
426 if (plen > 0)
427 skb_put_zero(skb, plen);
428 }
429}
430
431static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
432 const struct sw_flow_key *key,
433 const struct dp_upcall_info *upcall_info,
434 uint32_t cutlen)
435{
436 struct ovs_header *upcall;
437 struct sk_buff *nskb = NULL;
438 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
439 struct nlattr *nla;
440 size_t len;
441 unsigned int hlen;
442 int err, dp_ifindex;
443 u64 hash;
444
445 dp_ifindex = get_dpifindex(dp);
446 if (!dp_ifindex)
447 return -ENODEV;
448
449 if (skb_vlan_tag_present(skb)) {
450 nskb = skb_clone(skb, GFP_ATOMIC);
451 if (!nskb)
452 return -ENOMEM;
453
454 nskb = __vlan_hwaccel_push_inside(nskb);
455 if (!nskb)
456 return -ENOMEM;
457
458 skb = nskb;
459 }
460
461 if (nla_attr_size(skb->len) > USHRT_MAX) {
462 err = -EFBIG;
463 goto out;
464 }
465
466 /* Complete checksum if needed */
467 if (skb->ip_summed == CHECKSUM_PARTIAL &&
468 (err = skb_csum_hwoffload_help(skb, 0)))
469 goto out;
470
471 /* Older versions of OVS user space enforce alignment of the last
472 * Netlink attribute to NLA_ALIGNTO which would require extensive
473 * padding logic. Only perform zerocopy if padding is not required.
474 */
475 if (dp->user_features & OVS_DP_F_UNALIGNED)
476 hlen = skb_zerocopy_headlen(skb);
477 else
478 hlen = skb->len;
479
480 len = upcall_msg_size(upcall_info, hlen - cutlen,
481 OVS_CB(skb)->acts_origlen);
482 user_skb = genlmsg_new(len, GFP_ATOMIC);
483 if (!user_skb) {
484 err = -ENOMEM;
485 goto out;
486 }
487
488 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
489 0, upcall_info->cmd);
490 if (!upcall) {
491 err = -EINVAL;
492 goto out;
493 }
494 upcall->dp_ifindex = dp_ifindex;
495
496 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
497 if (err)
498 goto out;
499
500 if (upcall_info->userdata)
501 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
502 nla_len(upcall_info->userdata),
503 nla_data(upcall_info->userdata));
504
505 if (upcall_info->egress_tun_info) {
506 nla = nla_nest_start_noflag(user_skb,
507 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
508 if (!nla) {
509 err = -EMSGSIZE;
510 goto out;
511 }
512 err = ovs_nla_put_tunnel_info(user_skb,
513 upcall_info->egress_tun_info);
514 if (err)
515 goto out;
516
517 nla_nest_end(user_skb, nla);
518 }
519
520 if (upcall_info->actions_len) {
521 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
522 if (!nla) {
523 err = -EMSGSIZE;
524 goto out;
525 }
526 err = ovs_nla_put_actions(upcall_info->actions,
527 upcall_info->actions_len,
528 user_skb);
529 if (!err)
530 nla_nest_end(user_skb, nla);
531 else
532 nla_nest_cancel(user_skb, nla);
533 }
534
535 /* Add OVS_PACKET_ATTR_MRU */
536 if (upcall_info->mru &&
537 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
538 err = -ENOBUFS;
539 goto out;
540 }
541
542 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
543 if (cutlen > 0 &&
544 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
545 err = -ENOBUFS;
546 goto out;
547 }
548
549 /* Add OVS_PACKET_ATTR_HASH */
550 hash = skb_get_hash_raw(skb);
551 if (skb->sw_hash)
552 hash |= OVS_PACKET_HASH_SW_BIT;
553
554 if (skb->l4_hash)
555 hash |= OVS_PACKET_HASH_L4_BIT;
556
557 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
558 err = -ENOBUFS;
559 goto out;
560 }
561
562 /* Only reserve room for attribute header, packet data is added
563 * in skb_zerocopy() */
564 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
565 err = -ENOBUFS;
566 goto out;
567 }
568 nla->nla_len = nla_attr_size(skb->len - cutlen);
569
570 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
571 if (err)
572 goto out;
573
574 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
575 pad_packet(dp, user_skb);
576
577 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
578
579 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
580 user_skb = NULL;
581out:
582 if (err)
583 skb_tx_error(skb);
584 consume_skb(user_skb);
585 consume_skb(nskb);
586
587 return err;
588}
589
590static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
591{
592 struct ovs_header *ovs_header = genl_info_userhdr(info);
593 struct net *net = sock_net(skb->sk);
594 struct nlattr **a = info->attrs;
595 struct sw_flow_actions *acts;
596 struct sk_buff *packet;
597 struct sw_flow *flow;
598 struct sw_flow_actions *sf_acts;
599 struct datapath *dp;
600 struct vport *input_vport;
601 u16 mru = 0;
602 u64 hash;
603 int len;
604 int err;
605 bool log = !a[OVS_PACKET_ATTR_PROBE];
606
607 err = -EINVAL;
608 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
609 !a[OVS_PACKET_ATTR_ACTIONS])
610 goto err;
611
612 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
613 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
614 err = -ENOMEM;
615 if (!packet)
616 goto err;
617 skb_reserve(packet, NET_IP_ALIGN);
618
619 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
620
621 /* Set packet's mru */
622 if (a[OVS_PACKET_ATTR_MRU]) {
623 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
624 packet->ignore_df = 1;
625 }
626 OVS_CB(packet)->mru = mru;
627
628 if (a[OVS_PACKET_ATTR_HASH]) {
629 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
630
631 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
632 !!(hash & OVS_PACKET_HASH_SW_BIT),
633 !!(hash & OVS_PACKET_HASH_L4_BIT));
634 }
635
636 /* Build an sw_flow for sending this packet. */
637 flow = ovs_flow_alloc();
638 err = PTR_ERR(flow);
639 if (IS_ERR(flow))
640 goto err_kfree_skb;
641
642 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
643 packet, &flow->key, log);
644 if (err)
645 goto err_flow_free;
646
647 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
648 &flow->key, &acts, log);
649 if (err)
650 goto err_flow_free;
651
652 rcu_assign_pointer(flow->sf_acts, acts);
653 packet->priority = flow->key.phy.priority;
654 packet->mark = flow->key.phy.skb_mark;
655
656 rcu_read_lock();
657 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
658 err = -ENODEV;
659 if (!dp)
660 goto err_unlock;
661
662 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
663 if (!input_vport)
664 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
665
666 if (!input_vport)
667 goto err_unlock;
668
669 packet->dev = input_vport->dev;
670 OVS_CB(packet)->input_vport = input_vport;
671 sf_acts = rcu_dereference(flow->sf_acts);
672
673 local_bh_disable();
674 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
675 local_bh_enable();
676 rcu_read_unlock();
677
678 ovs_flow_free(flow, false);
679 return err;
680
681err_unlock:
682 rcu_read_unlock();
683err_flow_free:
684 ovs_flow_free(flow, false);
685err_kfree_skb:
686 kfree_skb(packet);
687err:
688 return err;
689}
690
691static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
692 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
693 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
694 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
695 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
696 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
697 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
698};
699
700static const struct genl_small_ops dp_packet_genl_ops[] = {
701 { .cmd = OVS_PACKET_CMD_EXECUTE,
702 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
703 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
704 .doit = ovs_packet_cmd_execute
705 }
706};
707
708static struct genl_family dp_packet_genl_family __ro_after_init = {
709 .hdrsize = sizeof(struct ovs_header),
710 .name = OVS_PACKET_FAMILY,
711 .version = OVS_PACKET_VERSION,
712 .maxattr = OVS_PACKET_ATTR_MAX,
713 .policy = packet_policy,
714 .netnsok = true,
715 .parallel_ops = true,
716 .small_ops = dp_packet_genl_ops,
717 .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
718 .resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
719 .module = THIS_MODULE,
720};
721
722static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
723 struct ovs_dp_megaflow_stats *mega_stats)
724{
725 int i;
726
727 memset(mega_stats, 0, sizeof(*mega_stats));
728
729 stats->n_flows = ovs_flow_tbl_count(&dp->table);
730 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
731
732 stats->n_hit = stats->n_missed = stats->n_lost = 0;
733
734 for_each_possible_cpu(i) {
735 const struct dp_stats_percpu *percpu_stats;
736 struct dp_stats_percpu local_stats;
737 unsigned int start;
738
739 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
740
741 do {
742 start = u64_stats_fetch_begin(&percpu_stats->syncp);
743 local_stats = *percpu_stats;
744 } while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
745
746 stats->n_hit += local_stats.n_hit;
747 stats->n_missed += local_stats.n_missed;
748 stats->n_lost += local_stats.n_lost;
749 mega_stats->n_mask_hit += local_stats.n_mask_hit;
750 mega_stats->n_cache_hit += local_stats.n_cache_hit;
751 }
752}
753
754static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
755{
756 return ovs_identifier_is_ufid(sfid) &&
757 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
758}
759
760static bool should_fill_mask(uint32_t ufid_flags)
761{
762 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
763}
764
765static bool should_fill_actions(uint32_t ufid_flags)
766{
767 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
768}
769
770static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
771 const struct sw_flow_id *sfid,
772 uint32_t ufid_flags)
773{
774 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
775
776 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
777 * see ovs_nla_put_identifier()
778 */
779 if (sfid && ovs_identifier_is_ufid(sfid))
780 len += nla_total_size(sfid->ufid_len);
781 else
782 len += nla_total_size(ovs_key_attr_size());
783
784 /* OVS_FLOW_ATTR_KEY */
785 if (!sfid || should_fill_key(sfid, ufid_flags))
786 len += nla_total_size(ovs_key_attr_size());
787
788 /* OVS_FLOW_ATTR_MASK */
789 if (should_fill_mask(ufid_flags))
790 len += nla_total_size(ovs_key_attr_size());
791
792 /* OVS_FLOW_ATTR_ACTIONS */
793 if (should_fill_actions(ufid_flags))
794 len += nla_total_size(acts->orig_len);
795
796 return len
797 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
798 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
799 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
800}
801
802/* Called with ovs_mutex or RCU read lock. */
803static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
804 struct sk_buff *skb)
805{
806 struct ovs_flow_stats stats;
807 __be16 tcp_flags;
808 unsigned long used;
809
810 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
811
812 if (used &&
813 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
814 OVS_FLOW_ATTR_PAD))
815 return -EMSGSIZE;
816
817 if (stats.n_packets &&
818 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
819 sizeof(struct ovs_flow_stats), &stats,
820 OVS_FLOW_ATTR_PAD))
821 return -EMSGSIZE;
822
823 if ((u8)ntohs(tcp_flags) &&
824 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
825 return -EMSGSIZE;
826
827 return 0;
828}
829
830/* Called with ovs_mutex or RCU read lock. */
831static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
832 struct sk_buff *skb, int skb_orig_len)
833{
834 struct nlattr *start;
835 int err;
836
837 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
838 * this is the first flow to be dumped into 'skb'. This is unusual for
839 * Netlink but individual action lists can be longer than
840 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
841 * The userspace caller can always fetch the actions separately if it
842 * really wants them. (Most userspace callers in fact don't care.)
843 *
844 * This can only fail for dump operations because the skb is always
845 * properly sized for single flows.
846 */
847 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
848 if (start) {
849 const struct sw_flow_actions *sf_acts;
850
851 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
852 err = ovs_nla_put_actions(sf_acts->actions,
853 sf_acts->actions_len, skb);
854
855 if (!err)
856 nla_nest_end(skb, start);
857 else {
858 if (skb_orig_len)
859 return err;
860
861 nla_nest_cancel(skb, start);
862 }
863 } else if (skb_orig_len) {
864 return -EMSGSIZE;
865 }
866
867 return 0;
868}
869
870/* Called with ovs_mutex or RCU read lock. */
871static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
872 struct sk_buff *skb, u32 portid,
873 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
874{
875 const int skb_orig_len = skb->len;
876 struct ovs_header *ovs_header;
877 int err;
878
879 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
880 flags, cmd);
881 if (!ovs_header)
882 return -EMSGSIZE;
883
884 ovs_header->dp_ifindex = dp_ifindex;
885
886 err = ovs_nla_put_identifier(flow, skb);
887 if (err)
888 goto error;
889
890 if (should_fill_key(&flow->id, ufid_flags)) {
891 err = ovs_nla_put_masked_key(flow, skb);
892 if (err)
893 goto error;
894 }
895
896 if (should_fill_mask(ufid_flags)) {
897 err = ovs_nla_put_mask(flow, skb);
898 if (err)
899 goto error;
900 }
901
902 err = ovs_flow_cmd_fill_stats(flow, skb);
903 if (err)
904 goto error;
905
906 if (should_fill_actions(ufid_flags)) {
907 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
908 if (err)
909 goto error;
910 }
911
912 genlmsg_end(skb, ovs_header);
913 return 0;
914
915error:
916 genlmsg_cancel(skb, ovs_header);
917 return err;
918}
919
920/* May not be called with RCU read lock. */
921static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
922 const struct sw_flow_id *sfid,
923 struct genl_info *info,
924 bool always,
925 uint32_t ufid_flags)
926{
927 struct sk_buff *skb;
928 size_t len;
929
930 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
931 return NULL;
932
933 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
934 skb = genlmsg_new(len, GFP_KERNEL);
935 if (!skb)
936 return ERR_PTR(-ENOMEM);
937
938 return skb;
939}
940
941/* Called with ovs_mutex. */
942static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
943 int dp_ifindex,
944 struct genl_info *info, u8 cmd,
945 bool always, u32 ufid_flags)
946{
947 struct sk_buff *skb;
948 int retval;
949
950 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
951 &flow->id, info, always, ufid_flags);
952 if (IS_ERR_OR_NULL(skb))
953 return skb;
954
955 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
956 info->snd_portid, info->snd_seq, 0,
957 cmd, ufid_flags);
958 if (WARN_ON_ONCE(retval < 0)) {
959 kfree_skb(skb);
960 skb = ERR_PTR(retval);
961 }
962 return skb;
963}
964
965static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
966{
967 struct net *net = sock_net(skb->sk);
968 struct nlattr **a = info->attrs;
969 struct ovs_header *ovs_header = genl_info_userhdr(info);
970 struct sw_flow *flow = NULL, *new_flow;
971 struct sw_flow_mask mask;
972 struct sk_buff *reply;
973 struct datapath *dp;
974 struct sw_flow_key *key;
975 struct sw_flow_actions *acts;
976 struct sw_flow_match match;
977 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
978 int error;
979 bool log = !a[OVS_FLOW_ATTR_PROBE];
980
981 /* Must have key and actions. */
982 error = -EINVAL;
983 if (!a[OVS_FLOW_ATTR_KEY]) {
984 OVS_NLERR(log, "Flow key attr not present in new flow.");
985 goto error;
986 }
987 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
988 OVS_NLERR(log, "Flow actions attr not present in new flow.");
989 goto error;
990 }
991
992 /* Most of the time we need to allocate a new flow, do it before
993 * locking.
994 */
995 new_flow = ovs_flow_alloc();
996 if (IS_ERR(new_flow)) {
997 error = PTR_ERR(new_flow);
998 goto error;
999 }
1000
1001 /* Extract key. */
1002 key = kzalloc(sizeof(*key), GFP_KERNEL);
1003 if (!key) {
1004 error = -ENOMEM;
1005 goto err_kfree_flow;
1006 }
1007
1008 ovs_match_init(&match, key, false, &mask);
1009 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1010 a[OVS_FLOW_ATTR_MASK], log);
1011 if (error)
1012 goto err_kfree_key;
1013
1014 ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1015
1016 /* Extract flow identifier. */
1017 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
1018 key, log);
1019 if (error)
1020 goto err_kfree_key;
1021
1022 /* Validate actions. */
1023 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1024 &new_flow->key, &acts, log);
1025 if (error) {
1026 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1027 goto err_kfree_key;
1028 }
1029
1030 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1031 ufid_flags);
1032 if (IS_ERR(reply)) {
1033 error = PTR_ERR(reply);
1034 goto err_kfree_acts;
1035 }
1036
1037 ovs_lock();
1038 dp = get_dp(net, ovs_header->dp_ifindex);
1039 if (unlikely(!dp)) {
1040 error = -ENODEV;
1041 goto err_unlock_ovs;
1042 }
1043
1044 /* Check if this is a duplicate flow */
1045 if (ovs_identifier_is_ufid(&new_flow->id))
1046 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1047 if (!flow)
1048 flow = ovs_flow_tbl_lookup(&dp->table, key);
1049 if (likely(!flow)) {
1050 rcu_assign_pointer(new_flow->sf_acts, acts);
1051
1052 /* Put flow in bucket. */
1053 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1054 if (unlikely(error)) {
1055 acts = NULL;
1056 goto err_unlock_ovs;
1057 }
1058
1059 if (unlikely(reply)) {
1060 error = ovs_flow_cmd_fill_info(new_flow,
1061 ovs_header->dp_ifindex,
1062 reply, info->snd_portid,
1063 info->snd_seq, 0,
1064 OVS_FLOW_CMD_NEW,
1065 ufid_flags);
1066 BUG_ON(error < 0);
1067 }
1068 ovs_unlock();
1069 } else {
1070 struct sw_flow_actions *old_acts;
1071
1072 /* Bail out if we're not allowed to modify an existing flow.
1073 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1074 * because Generic Netlink treats the latter as a dump
1075 * request. We also accept NLM_F_EXCL in case that bug ever
1076 * gets fixed.
1077 */
1078 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1079 | NLM_F_EXCL))) {
1080 error = -EEXIST;
1081 goto err_unlock_ovs;
1082 }
1083 /* The flow identifier has to be the same for flow updates.
1084 * Look for any overlapping flow.
1085 */
1086 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1087 if (ovs_identifier_is_key(&flow->id))
1088 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1089 &match);
1090 else /* UFID matches but key is different */
1091 flow = NULL;
1092 if (!flow) {
1093 error = -ENOENT;
1094 goto err_unlock_ovs;
1095 }
1096 }
1097 /* Update actions. */
1098 old_acts = ovsl_dereference(flow->sf_acts);
1099 rcu_assign_pointer(flow->sf_acts, acts);
1100
1101 if (unlikely(reply)) {
1102 error = ovs_flow_cmd_fill_info(flow,
1103 ovs_header->dp_ifindex,
1104 reply, info->snd_portid,
1105 info->snd_seq, 0,
1106 OVS_FLOW_CMD_NEW,
1107 ufid_flags);
1108 BUG_ON(error < 0);
1109 }
1110 ovs_unlock();
1111
1112 ovs_nla_free_flow_actions_rcu(old_acts);
1113 ovs_flow_free(new_flow, false);
1114 }
1115
1116 if (reply)
1117 ovs_notify(&dp_flow_genl_family, reply, info);
1118
1119 kfree(key);
1120 return 0;
1121
1122err_unlock_ovs:
1123 ovs_unlock();
1124 kfree_skb(reply);
1125err_kfree_acts:
1126 ovs_nla_free_flow_actions(acts);
1127err_kfree_key:
1128 kfree(key);
1129err_kfree_flow:
1130 ovs_flow_free(new_flow, false);
1131error:
1132 return error;
1133}
1134
1135/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1136static noinline_for_stack
1137struct sw_flow_actions *get_flow_actions(struct net *net,
1138 const struct nlattr *a,
1139 const struct sw_flow_key *key,
1140 const struct sw_flow_mask *mask,
1141 bool log)
1142{
1143 struct sw_flow_actions *acts;
1144 struct sw_flow_key masked_key;
1145 int error;
1146
1147 ovs_flow_mask_key(&masked_key, key, true, mask);
1148 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1149 if (error) {
1150 OVS_NLERR(log,
1151 "Actions may not be safe on all matching packets");
1152 return ERR_PTR(error);
1153 }
1154
1155 return acts;
1156}
1157
1158/* Factor out match-init and action-copy to avoid
1159 * "Wframe-larger-than=1024" warning. Because mask is only
1160 * used to get actions, we new a function to save some
1161 * stack space.
1162 *
1163 * If there are not key and action attrs, we return 0
1164 * directly. In the case, the caller will also not use the
1165 * match as before. If there is action attr, we try to get
1166 * actions and save them to *acts. Before returning from
1167 * the function, we reset the match->mask pointer. Because
1168 * we should not to return match object with dangling reference
1169 * to mask.
1170 * */
1171static noinline_for_stack int
1172ovs_nla_init_match_and_action(struct net *net,
1173 struct sw_flow_match *match,
1174 struct sw_flow_key *key,
1175 struct nlattr **a,
1176 struct sw_flow_actions **acts,
1177 bool log)
1178{
1179 struct sw_flow_mask mask;
1180 int error = 0;
1181
1182 if (a[OVS_FLOW_ATTR_KEY]) {
1183 ovs_match_init(match, key, true, &mask);
1184 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1185 a[OVS_FLOW_ATTR_MASK], log);
1186 if (error)
1187 goto error;
1188 }
1189
1190 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1191 if (!a[OVS_FLOW_ATTR_KEY]) {
1192 OVS_NLERR(log,
1193 "Flow key attribute not present in set flow.");
1194 error = -EINVAL;
1195 goto error;
1196 }
1197
1198 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1199 &mask, log);
1200 if (IS_ERR(*acts)) {
1201 error = PTR_ERR(*acts);
1202 goto error;
1203 }
1204 }
1205
1206 /* On success, error is 0. */
1207error:
1208 match->mask = NULL;
1209 return error;
1210}
1211
1212static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1213{
1214 struct net *net = sock_net(skb->sk);
1215 struct nlattr **a = info->attrs;
1216 struct ovs_header *ovs_header = genl_info_userhdr(info);
1217 struct sw_flow_key key;
1218 struct sw_flow *flow;
1219 struct sk_buff *reply = NULL;
1220 struct datapath *dp;
1221 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1222 struct sw_flow_match match;
1223 struct sw_flow_id sfid;
1224 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1225 int error = 0;
1226 bool log = !a[OVS_FLOW_ATTR_PROBE];
1227 bool ufid_present;
1228
1229 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1230 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1231 OVS_NLERR(log,
1232 "Flow set message rejected, Key attribute missing.");
1233 return -EINVAL;
1234 }
1235
1236 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1237 &acts, log);
1238 if (error)
1239 goto error;
1240
1241 if (acts) {
1242 /* Can allocate before locking if have acts. */
1243 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1244 ufid_flags);
1245 if (IS_ERR(reply)) {
1246 error = PTR_ERR(reply);
1247 goto err_kfree_acts;
1248 }
1249 }
1250
1251 ovs_lock();
1252 dp = get_dp(net, ovs_header->dp_ifindex);
1253 if (unlikely(!dp)) {
1254 error = -ENODEV;
1255 goto err_unlock_ovs;
1256 }
1257 /* Check that the flow exists. */
1258 if (ufid_present)
1259 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1260 else
1261 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1262 if (unlikely(!flow)) {
1263 error = -ENOENT;
1264 goto err_unlock_ovs;
1265 }
1266
1267 /* Update actions, if present. */
1268 if (likely(acts)) {
1269 old_acts = ovsl_dereference(flow->sf_acts);
1270 rcu_assign_pointer(flow->sf_acts, acts);
1271
1272 if (unlikely(reply)) {
1273 error = ovs_flow_cmd_fill_info(flow,
1274 ovs_header->dp_ifindex,
1275 reply, info->snd_portid,
1276 info->snd_seq, 0,
1277 OVS_FLOW_CMD_SET,
1278 ufid_flags);
1279 BUG_ON(error < 0);
1280 }
1281 } else {
1282 /* Could not alloc without acts before locking. */
1283 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1284 info, OVS_FLOW_CMD_SET, false,
1285 ufid_flags);
1286
1287 if (IS_ERR(reply)) {
1288 error = PTR_ERR(reply);
1289 goto err_unlock_ovs;
1290 }
1291 }
1292
1293 /* Clear stats. */
1294 if (a[OVS_FLOW_ATTR_CLEAR])
1295 ovs_flow_stats_clear(flow);
1296 ovs_unlock();
1297
1298 if (reply)
1299 ovs_notify(&dp_flow_genl_family, reply, info);
1300 if (old_acts)
1301 ovs_nla_free_flow_actions_rcu(old_acts);
1302
1303 return 0;
1304
1305err_unlock_ovs:
1306 ovs_unlock();
1307 kfree_skb(reply);
1308err_kfree_acts:
1309 ovs_nla_free_flow_actions(acts);
1310error:
1311 return error;
1312}
1313
1314static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1315{
1316 struct nlattr **a = info->attrs;
1317 struct ovs_header *ovs_header = genl_info_userhdr(info);
1318 struct net *net = sock_net(skb->sk);
1319 struct sw_flow_key key;
1320 struct sk_buff *reply;
1321 struct sw_flow *flow;
1322 struct datapath *dp;
1323 struct sw_flow_match match;
1324 struct sw_flow_id ufid;
1325 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1326 int err = 0;
1327 bool log = !a[OVS_FLOW_ATTR_PROBE];
1328 bool ufid_present;
1329
1330 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1331 if (a[OVS_FLOW_ATTR_KEY]) {
1332 ovs_match_init(&match, &key, true, NULL);
1333 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1334 log);
1335 } else if (!ufid_present) {
1336 OVS_NLERR(log,
1337 "Flow get message rejected, Key attribute missing.");
1338 err = -EINVAL;
1339 }
1340 if (err)
1341 return err;
1342
1343 ovs_lock();
1344 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1345 if (!dp) {
1346 err = -ENODEV;
1347 goto unlock;
1348 }
1349
1350 if (ufid_present)
1351 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1352 else
1353 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1354 if (!flow) {
1355 err = -ENOENT;
1356 goto unlock;
1357 }
1358
1359 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1360 OVS_FLOW_CMD_GET, true, ufid_flags);
1361 if (IS_ERR(reply)) {
1362 err = PTR_ERR(reply);
1363 goto unlock;
1364 }
1365
1366 ovs_unlock();
1367 return genlmsg_reply(reply, info);
1368unlock:
1369 ovs_unlock();
1370 return err;
1371}
1372
1373static int ovs_flow_cmd_del(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 = NULL;
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;
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],
1393 NULL, log);
1394 if (unlikely(err))
1395 return err;
1396 }
1397
1398 ovs_lock();
1399 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1400 if (unlikely(!dp)) {
1401 err = -ENODEV;
1402 goto unlock;
1403 }
1404
1405 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1406 err = ovs_flow_tbl_flush(&dp->table);
1407 goto unlock;
1408 }
1409
1410 if (ufid_present)
1411 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1412 else
1413 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1414 if (unlikely(!flow)) {
1415 err = -ENOENT;
1416 goto unlock;
1417 }
1418
1419 ovs_flow_tbl_remove(&dp->table, flow);
1420 ovs_unlock();
1421
1422 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1423 &flow->id, info, false, ufid_flags);
1424 if (likely(reply)) {
1425 if (!IS_ERR(reply)) {
1426 rcu_read_lock(); /*To keep RCU checker happy. */
1427 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1428 reply, info->snd_portid,
1429 info->snd_seq, 0,
1430 OVS_FLOW_CMD_DEL,
1431 ufid_flags);
1432 rcu_read_unlock();
1433 if (WARN_ON_ONCE(err < 0)) {
1434 kfree_skb(reply);
1435 goto out_free;
1436 }
1437
1438 ovs_notify(&dp_flow_genl_family, reply, info);
1439 } else {
1440 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1441 PTR_ERR(reply));
1442 }
1443 }
1444
1445out_free:
1446 ovs_flow_free(flow, true);
1447 return 0;
1448unlock:
1449 ovs_unlock();
1450 return err;
1451}
1452
1453static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1454{
1455 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1456 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1457 struct table_instance *ti;
1458 struct datapath *dp;
1459 u32 ufid_flags;
1460 int err;
1461
1462 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1463 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1464 if (err)
1465 return err;
1466 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1467
1468 rcu_read_lock();
1469 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1470 if (!dp) {
1471 rcu_read_unlock();
1472 return -ENODEV;
1473 }
1474
1475 ti = rcu_dereference(dp->table.ti);
1476 for (;;) {
1477 struct sw_flow *flow;
1478 u32 bucket, obj;
1479
1480 bucket = cb->args[0];
1481 obj = cb->args[1];
1482 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1483 if (!flow)
1484 break;
1485
1486 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1487 NETLINK_CB(cb->skb).portid,
1488 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1489 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1490 break;
1491
1492 cb->args[0] = bucket;
1493 cb->args[1] = obj;
1494 }
1495 rcu_read_unlock();
1496 return skb->len;
1497}
1498
1499static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1500 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1501 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1502 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1503 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1504 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1505 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1506 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1507};
1508
1509static const struct genl_small_ops dp_flow_genl_ops[] = {
1510 { .cmd = OVS_FLOW_CMD_NEW,
1511 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1512 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1513 .doit = ovs_flow_cmd_new
1514 },
1515 { .cmd = OVS_FLOW_CMD_DEL,
1516 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1517 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1518 .doit = ovs_flow_cmd_del
1519 },
1520 { .cmd = OVS_FLOW_CMD_GET,
1521 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1522 .flags = 0, /* OK for unprivileged users. */
1523 .doit = ovs_flow_cmd_get,
1524 .dumpit = ovs_flow_cmd_dump
1525 },
1526 { .cmd = OVS_FLOW_CMD_SET,
1527 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1528 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1529 .doit = ovs_flow_cmd_set,
1530 },
1531};
1532
1533static struct genl_family dp_flow_genl_family __ro_after_init = {
1534 .hdrsize = sizeof(struct ovs_header),
1535 .name = OVS_FLOW_FAMILY,
1536 .version = OVS_FLOW_VERSION,
1537 .maxattr = OVS_FLOW_ATTR_MAX,
1538 .policy = flow_policy,
1539 .netnsok = true,
1540 .parallel_ops = true,
1541 .small_ops = dp_flow_genl_ops,
1542 .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1543 .resv_start_op = OVS_FLOW_CMD_SET + 1,
1544 .mcgrps = &ovs_dp_flow_multicast_group,
1545 .n_mcgrps = 1,
1546 .module = THIS_MODULE,
1547};
1548
1549static size_t ovs_dp_cmd_msg_size(void)
1550{
1551 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1552
1553 msgsize += nla_total_size(IFNAMSIZ);
1554 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1555 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1556 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1557 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1558 msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1559
1560 return msgsize;
1561}
1562
1563/* Called with ovs_mutex. */
1564static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1565 u32 portid, u32 seq, u32 flags, u8 cmd)
1566{
1567 struct ovs_header *ovs_header;
1568 struct ovs_dp_stats dp_stats;
1569 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1570 struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1571 int err, pids_len;
1572
1573 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1574 flags, cmd);
1575 if (!ovs_header)
1576 goto error;
1577
1578 ovs_header->dp_ifindex = get_dpifindex(dp);
1579
1580 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1581 if (err)
1582 goto nla_put_failure;
1583
1584 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1585 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1586 &dp_stats, OVS_DP_ATTR_PAD))
1587 goto nla_put_failure;
1588
1589 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1590 sizeof(struct ovs_dp_megaflow_stats),
1591 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1592 goto nla_put_failure;
1593
1594 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1595 goto nla_put_failure;
1596
1597 if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1598 ovs_flow_tbl_masks_cache_size(&dp->table)))
1599 goto nla_put_failure;
1600
1601 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1602 pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1603 if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1604 goto nla_put_failure;
1605 }
1606
1607 genlmsg_end(skb, ovs_header);
1608 return 0;
1609
1610nla_put_failure:
1611 genlmsg_cancel(skb, ovs_header);
1612error:
1613 return -EMSGSIZE;
1614}
1615
1616static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1617{
1618 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1619}
1620
1621/* Called with rcu_read_lock or ovs_mutex. */
1622static struct datapath *lookup_datapath(struct net *net,
1623 const struct ovs_header *ovs_header,
1624 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1625{
1626 struct datapath *dp;
1627
1628 if (!a[OVS_DP_ATTR_NAME])
1629 dp = get_dp(net, ovs_header->dp_ifindex);
1630 else {
1631 struct vport *vport;
1632
1633 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1634 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1635 }
1636 return dp ? dp : ERR_PTR(-ENODEV);
1637}
1638
1639static void ovs_dp_reset_user_features(struct sk_buff *skb,
1640 struct genl_info *info)
1641{
1642 struct datapath *dp;
1643
1644 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1645 info->attrs);
1646 if (IS_ERR(dp))
1647 return;
1648
1649 pr_warn("%s: Dropping previously announced user features\n",
1650 ovs_dp_name(dp));
1651 dp->user_features = 0;
1652}
1653
1654static int ovs_dp_set_upcall_portids(struct datapath *dp,
1655 const struct nlattr *ids)
1656{
1657 struct dp_nlsk_pids *old, *dp_nlsk_pids;
1658
1659 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1660 return -EINVAL;
1661
1662 old = ovsl_dereference(dp->upcall_portids);
1663
1664 dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1665 GFP_KERNEL);
1666 if (!dp_nlsk_pids)
1667 return -ENOMEM;
1668
1669 dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1670 nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1671
1672 rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1673
1674 kfree_rcu(old, rcu);
1675
1676 return 0;
1677}
1678
1679u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1680{
1681 struct dp_nlsk_pids *dp_nlsk_pids;
1682
1683 dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1684
1685 if (dp_nlsk_pids) {
1686 if (cpu_id < dp_nlsk_pids->n_pids) {
1687 return dp_nlsk_pids->pids[cpu_id];
1688 } else if (dp_nlsk_pids->n_pids > 0 &&
1689 cpu_id >= dp_nlsk_pids->n_pids) {
1690 /* If the number of netlink PIDs is mismatched with
1691 * the number of CPUs as seen by the kernel, log this
1692 * and send the upcall to an arbitrary socket (0) in
1693 * order to not drop packets
1694 */
1695 pr_info_ratelimited("cpu_id mismatch with handler threads");
1696 return dp_nlsk_pids->pids[cpu_id %
1697 dp_nlsk_pids->n_pids];
1698 } else {
1699 return 0;
1700 }
1701 } else {
1702 return 0;
1703 }
1704}
1705
1706static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1707{
1708 u32 user_features = 0, old_features = dp->user_features;
1709 int err;
1710
1711 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1712 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1713
1714 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1715 OVS_DP_F_UNALIGNED |
1716 OVS_DP_F_TC_RECIRC_SHARING |
1717 OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1718 return -EOPNOTSUPP;
1719
1720#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1721 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1722 return -EOPNOTSUPP;
1723#endif
1724 }
1725
1726 if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1727 int err;
1728 u32 cache_size;
1729
1730 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1731 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1732 if (err)
1733 return err;
1734 }
1735
1736 dp->user_features = user_features;
1737
1738 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1739 a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1740 /* Upcall Netlink Port IDs have been updated */
1741 err = ovs_dp_set_upcall_portids(dp,
1742 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1743 if (err)
1744 return err;
1745 }
1746
1747 if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1748 !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1749 tc_skb_ext_tc_enable();
1750 else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1751 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1752 tc_skb_ext_tc_disable();
1753
1754 return 0;
1755}
1756
1757static int ovs_dp_stats_init(struct datapath *dp)
1758{
1759 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1760 if (!dp->stats_percpu)
1761 return -ENOMEM;
1762
1763 return 0;
1764}
1765
1766static int ovs_dp_vport_init(struct datapath *dp)
1767{
1768 int i;
1769
1770 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1771 sizeof(struct hlist_head),
1772 GFP_KERNEL);
1773 if (!dp->ports)
1774 return -ENOMEM;
1775
1776 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1777 INIT_HLIST_HEAD(&dp->ports[i]);
1778
1779 return 0;
1780}
1781
1782static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1783{
1784 struct nlattr **a = info->attrs;
1785 struct vport_parms parms;
1786 struct sk_buff *reply;
1787 struct datapath *dp;
1788 struct vport *vport;
1789 struct ovs_net *ovs_net;
1790 int err;
1791
1792 err = -EINVAL;
1793 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1794 goto err;
1795
1796 reply = ovs_dp_cmd_alloc_info();
1797 if (!reply)
1798 return -ENOMEM;
1799
1800 err = -ENOMEM;
1801 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1802 if (dp == NULL)
1803 goto err_destroy_reply;
1804
1805 ovs_dp_set_net(dp, sock_net(skb->sk));
1806
1807 /* Allocate table. */
1808 err = ovs_flow_tbl_init(&dp->table);
1809 if (err)
1810 goto err_destroy_dp;
1811
1812 err = ovs_dp_stats_init(dp);
1813 if (err)
1814 goto err_destroy_table;
1815
1816 err = ovs_dp_vport_init(dp);
1817 if (err)
1818 goto err_destroy_stats;
1819
1820 err = ovs_meters_init(dp);
1821 if (err)
1822 goto err_destroy_ports;
1823
1824 /* Set up our datapath device. */
1825 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1826 parms.type = OVS_VPORT_TYPE_INTERNAL;
1827 parms.options = NULL;
1828 parms.dp = dp;
1829 parms.port_no = OVSP_LOCAL;
1830 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1831 parms.desired_ifindex = nla_get_s32_default(a[OVS_DP_ATTR_IFINDEX], 0);
1832
1833 /* So far only local changes have been made, now need the lock. */
1834 ovs_lock();
1835
1836 err = ovs_dp_change(dp, a);
1837 if (err)
1838 goto err_unlock_and_destroy_meters;
1839
1840 vport = new_vport(&parms);
1841 if (IS_ERR(vport)) {
1842 err = PTR_ERR(vport);
1843 if (err == -EBUSY)
1844 err = -EEXIST;
1845
1846 if (err == -EEXIST) {
1847 /* An outdated user space instance that does not understand
1848 * the concept of user_features has attempted to create a new
1849 * datapath and is likely to reuse it. Drop all user features.
1850 */
1851 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1852 ovs_dp_reset_user_features(skb, info);
1853 }
1854
1855 goto err_destroy_portids;
1856 }
1857
1858 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1859 info->snd_seq, 0, OVS_DP_CMD_NEW);
1860 BUG_ON(err < 0);
1861
1862 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1863 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1864
1865 ovs_unlock();
1866
1867 ovs_notify(&dp_datapath_genl_family, reply, info);
1868 return 0;
1869
1870err_destroy_portids:
1871 kfree(rcu_dereference_raw(dp->upcall_portids));
1872err_unlock_and_destroy_meters:
1873 ovs_unlock();
1874 ovs_meters_exit(dp);
1875err_destroy_ports:
1876 kfree(dp->ports);
1877err_destroy_stats:
1878 free_percpu(dp->stats_percpu);
1879err_destroy_table:
1880 ovs_flow_tbl_destroy(&dp->table);
1881err_destroy_dp:
1882 kfree(dp);
1883err_destroy_reply:
1884 kfree_skb(reply);
1885err:
1886 return err;
1887}
1888
1889/* Called with ovs_mutex. */
1890static void __dp_destroy(struct datapath *dp)
1891{
1892 struct flow_table *table = &dp->table;
1893 int i;
1894
1895 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1896 tc_skb_ext_tc_disable();
1897
1898 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1899 struct vport *vport;
1900 struct hlist_node *n;
1901
1902 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1903 if (vport->port_no != OVSP_LOCAL)
1904 ovs_dp_detach_port(vport);
1905 }
1906
1907 list_del_rcu(&dp->list_node);
1908
1909 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1910 * all ports in datapath are destroyed first before freeing datapath.
1911 */
1912 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1913
1914 /* Flush sw_flow in the tables. RCU cb only releases resource
1915 * such as dp, ports and tables. That may avoid some issues
1916 * such as RCU usage warning.
1917 */
1918 table_instance_flow_flush(table, ovsl_dereference(table->ti),
1919 ovsl_dereference(table->ufid_ti));
1920
1921 /* RCU destroy the ports, meters and flow tables. */
1922 call_rcu(&dp->rcu, destroy_dp_rcu);
1923}
1924
1925static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1926{
1927 struct sk_buff *reply;
1928 struct datapath *dp;
1929 int err;
1930
1931 reply = ovs_dp_cmd_alloc_info();
1932 if (!reply)
1933 return -ENOMEM;
1934
1935 ovs_lock();
1936 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1937 info->attrs);
1938 err = PTR_ERR(dp);
1939 if (IS_ERR(dp))
1940 goto err_unlock_free;
1941
1942 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1943 info->snd_seq, 0, OVS_DP_CMD_DEL);
1944 BUG_ON(err < 0);
1945
1946 __dp_destroy(dp);
1947 ovs_unlock();
1948
1949 ovs_notify(&dp_datapath_genl_family, reply, info);
1950
1951 return 0;
1952
1953err_unlock_free:
1954 ovs_unlock();
1955 kfree_skb(reply);
1956 return err;
1957}
1958
1959static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1960{
1961 struct sk_buff *reply;
1962 struct datapath *dp;
1963 int err;
1964
1965 reply = ovs_dp_cmd_alloc_info();
1966 if (!reply)
1967 return -ENOMEM;
1968
1969 ovs_lock();
1970 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1971 info->attrs);
1972 err = PTR_ERR(dp);
1973 if (IS_ERR(dp))
1974 goto err_unlock_free;
1975
1976 err = ovs_dp_change(dp, info->attrs);
1977 if (err)
1978 goto err_unlock_free;
1979
1980 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1981 info->snd_seq, 0, OVS_DP_CMD_SET);
1982 BUG_ON(err < 0);
1983
1984 ovs_unlock();
1985 ovs_notify(&dp_datapath_genl_family, reply, info);
1986
1987 return 0;
1988
1989err_unlock_free:
1990 ovs_unlock();
1991 kfree_skb(reply);
1992 return err;
1993}
1994
1995static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1996{
1997 struct sk_buff *reply;
1998 struct datapath *dp;
1999 int err;
2000
2001 reply = ovs_dp_cmd_alloc_info();
2002 if (!reply)
2003 return -ENOMEM;
2004
2005 ovs_lock();
2006 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2007 info->attrs);
2008 if (IS_ERR(dp)) {
2009 err = PTR_ERR(dp);
2010 goto err_unlock_free;
2011 }
2012 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2013 info->snd_seq, 0, OVS_DP_CMD_GET);
2014 BUG_ON(err < 0);
2015 ovs_unlock();
2016
2017 return genlmsg_reply(reply, info);
2018
2019err_unlock_free:
2020 ovs_unlock();
2021 kfree_skb(reply);
2022 return err;
2023}
2024
2025static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2026{
2027 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2028 struct datapath *dp;
2029 int skip = cb->args[0];
2030 int i = 0;
2031
2032 ovs_lock();
2033 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2034 if (i >= skip &&
2035 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2036 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2037 OVS_DP_CMD_GET) < 0)
2038 break;
2039 i++;
2040 }
2041 ovs_unlock();
2042
2043 cb->args[0] = i;
2044
2045 return skb->len;
2046}
2047
2048static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2049 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2050 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2051 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2052 [OVS_DP_ATTR_MASKS_CACHE_SIZE] = NLA_POLICY_RANGE(NLA_U32, 0,
2053 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2054 [OVS_DP_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2055};
2056
2057static const struct genl_small_ops dp_datapath_genl_ops[] = {
2058 { .cmd = OVS_DP_CMD_NEW,
2059 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2060 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2061 .doit = ovs_dp_cmd_new
2062 },
2063 { .cmd = OVS_DP_CMD_DEL,
2064 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2065 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2066 .doit = ovs_dp_cmd_del
2067 },
2068 { .cmd = OVS_DP_CMD_GET,
2069 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2070 .flags = 0, /* OK for unprivileged users. */
2071 .doit = ovs_dp_cmd_get,
2072 .dumpit = ovs_dp_cmd_dump
2073 },
2074 { .cmd = OVS_DP_CMD_SET,
2075 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2076 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2077 .doit = ovs_dp_cmd_set,
2078 },
2079};
2080
2081static struct genl_family dp_datapath_genl_family __ro_after_init = {
2082 .hdrsize = sizeof(struct ovs_header),
2083 .name = OVS_DATAPATH_FAMILY,
2084 .version = OVS_DATAPATH_VERSION,
2085 .maxattr = OVS_DP_ATTR_MAX,
2086 .policy = datapath_policy,
2087 .netnsok = true,
2088 .parallel_ops = true,
2089 .small_ops = dp_datapath_genl_ops,
2090 .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2091 .resv_start_op = OVS_DP_CMD_SET + 1,
2092 .mcgrps = &ovs_dp_datapath_multicast_group,
2093 .n_mcgrps = 1,
2094 .module = THIS_MODULE,
2095};
2096
2097/* Called with ovs_mutex or RCU read lock. */
2098static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2099 struct net *net, u32 portid, u32 seq,
2100 u32 flags, u8 cmd, gfp_t gfp)
2101{
2102 struct ovs_header *ovs_header;
2103 struct ovs_vport_stats vport_stats;
2104 struct net *net_vport;
2105 int err;
2106
2107 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2108 flags, cmd);
2109 if (!ovs_header)
2110 return -EMSGSIZE;
2111
2112 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2113
2114 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2115 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2116 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2117 ovs_vport_name(vport)) ||
2118 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2119 goto nla_put_failure;
2120
2121 rcu_read_lock();
2122 net_vport = dev_net_rcu(vport->dev);
2123 if (!net_eq(net, net_vport)) {
2124 int id = peernet2id_alloc(net, net_vport, GFP_ATOMIC);
2125
2126 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2127 goto nla_put_failure_unlock;
2128 }
2129 rcu_read_unlock();
2130
2131 ovs_vport_get_stats(vport, &vport_stats);
2132 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2133 sizeof(struct ovs_vport_stats), &vport_stats,
2134 OVS_VPORT_ATTR_PAD))
2135 goto nla_put_failure;
2136
2137 if (ovs_vport_get_upcall_stats(vport, skb))
2138 goto nla_put_failure;
2139
2140 if (ovs_vport_get_upcall_portids(vport, skb))
2141 goto nla_put_failure;
2142
2143 err = ovs_vport_get_options(vport, skb);
2144 if (err == -EMSGSIZE)
2145 goto error;
2146
2147 genlmsg_end(skb, ovs_header);
2148 return 0;
2149
2150nla_put_failure_unlock:
2151 rcu_read_unlock();
2152nla_put_failure:
2153 err = -EMSGSIZE;
2154error:
2155 genlmsg_cancel(skb, ovs_header);
2156 return err;
2157}
2158
2159static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2160{
2161 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2162}
2163
2164/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2165struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2166 u32 portid, u32 seq, u8 cmd)
2167{
2168 struct sk_buff *skb;
2169 int retval;
2170
2171 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2172 if (!skb)
2173 return ERR_PTR(-ENOMEM);
2174
2175 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2176 GFP_KERNEL);
2177 BUG_ON(retval < 0);
2178
2179 return skb;
2180}
2181
2182/* Called with ovs_mutex or RCU read lock. */
2183static struct vport *lookup_vport(struct net *net,
2184 const struct ovs_header *ovs_header,
2185 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2186{
2187 struct datapath *dp;
2188 struct vport *vport;
2189
2190 if (a[OVS_VPORT_ATTR_IFINDEX])
2191 return ERR_PTR(-EOPNOTSUPP);
2192 if (a[OVS_VPORT_ATTR_NAME]) {
2193 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2194 if (!vport)
2195 return ERR_PTR(-ENODEV);
2196 if (ovs_header->dp_ifindex &&
2197 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2198 return ERR_PTR(-ENODEV);
2199 return vport;
2200 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2201 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2202
2203 if (port_no >= DP_MAX_PORTS)
2204 return ERR_PTR(-EFBIG);
2205
2206 dp = get_dp(net, ovs_header->dp_ifindex);
2207 if (!dp)
2208 return ERR_PTR(-ENODEV);
2209
2210 vport = ovs_vport_ovsl_rcu(dp, port_no);
2211 if (!vport)
2212 return ERR_PTR(-ENODEV);
2213 return vport;
2214 } else
2215 return ERR_PTR(-EINVAL);
2216
2217}
2218
2219static unsigned int ovs_get_max_headroom(struct datapath *dp)
2220{
2221 unsigned int dev_headroom, max_headroom = 0;
2222 struct net_device *dev;
2223 struct vport *vport;
2224 int i;
2225
2226 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2227 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2228 lockdep_ovsl_is_held()) {
2229 dev = vport->dev;
2230 dev_headroom = netdev_get_fwd_headroom(dev);
2231 if (dev_headroom > max_headroom)
2232 max_headroom = dev_headroom;
2233 }
2234 }
2235
2236 return max_headroom;
2237}
2238
2239/* Called with ovs_mutex */
2240static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2241{
2242 struct vport *vport;
2243 int i;
2244
2245 dp->max_headroom = new_headroom;
2246 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2247 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2248 lockdep_ovsl_is_held())
2249 netdev_set_rx_headroom(vport->dev, new_headroom);
2250 }
2251}
2252
2253static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2254{
2255 struct nlattr **a = info->attrs;
2256 struct ovs_header *ovs_header = genl_info_userhdr(info);
2257 struct vport_parms parms;
2258 struct sk_buff *reply;
2259 struct vport *vport;
2260 struct datapath *dp;
2261 unsigned int new_headroom;
2262 u32 port_no;
2263 int err;
2264
2265 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2266 !a[OVS_VPORT_ATTR_UPCALL_PID])
2267 return -EINVAL;
2268
2269 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2270
2271 if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2272 return -EOPNOTSUPP;
2273
2274 port_no = nla_get_u32_default(a[OVS_VPORT_ATTR_PORT_NO], 0);
2275 if (port_no >= DP_MAX_PORTS)
2276 return -EFBIG;
2277
2278 reply = ovs_vport_cmd_alloc_info();
2279 if (!reply)
2280 return -ENOMEM;
2281
2282 ovs_lock();
2283restart:
2284 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2285 err = -ENODEV;
2286 if (!dp)
2287 goto exit_unlock_free;
2288
2289 if (port_no) {
2290 vport = ovs_vport_ovsl(dp, port_no);
2291 err = -EBUSY;
2292 if (vport)
2293 goto exit_unlock_free;
2294 } else {
2295 for (port_no = 1; ; port_no++) {
2296 if (port_no >= DP_MAX_PORTS) {
2297 err = -EFBIG;
2298 goto exit_unlock_free;
2299 }
2300 vport = ovs_vport_ovsl(dp, port_no);
2301 if (!vport)
2302 break;
2303 }
2304 }
2305
2306 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2307 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2308 parms.dp = dp;
2309 parms.port_no = port_no;
2310 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2311 parms.desired_ifindex = nla_get_s32_default(a[OVS_VPORT_ATTR_IFINDEX],
2312 0);
2313
2314 vport = new_vport(&parms);
2315 err = PTR_ERR(vport);
2316 if (IS_ERR(vport)) {
2317 if (err == -EAGAIN)
2318 goto restart;
2319 goto exit_unlock_free;
2320 }
2321
2322 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2323 info->snd_portid, info->snd_seq, 0,
2324 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2325
2326 new_headroom = netdev_get_fwd_headroom(vport->dev);
2327
2328 if (new_headroom > dp->max_headroom)
2329 ovs_update_headroom(dp, new_headroom);
2330 else
2331 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2332
2333 BUG_ON(err < 0);
2334 ovs_unlock();
2335
2336 ovs_notify(&dp_vport_genl_family, reply, info);
2337 return 0;
2338
2339exit_unlock_free:
2340 ovs_unlock();
2341 kfree_skb(reply);
2342 return err;
2343}
2344
2345static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2346{
2347 struct nlattr **a = info->attrs;
2348 struct sk_buff *reply;
2349 struct vport *vport;
2350 int err;
2351
2352 reply = ovs_vport_cmd_alloc_info();
2353 if (!reply)
2354 return -ENOMEM;
2355
2356 ovs_lock();
2357 vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2358 err = PTR_ERR(vport);
2359 if (IS_ERR(vport))
2360 goto exit_unlock_free;
2361
2362 if (a[OVS_VPORT_ATTR_TYPE] &&
2363 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2364 err = -EINVAL;
2365 goto exit_unlock_free;
2366 }
2367
2368 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2369 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2370 if (err)
2371 goto exit_unlock_free;
2372 }
2373
2374
2375 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2376 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2377
2378 err = ovs_vport_set_upcall_portids(vport, ids);
2379 if (err)
2380 goto exit_unlock_free;
2381 }
2382
2383 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2384 info->snd_portid, info->snd_seq, 0,
2385 OVS_VPORT_CMD_SET, GFP_KERNEL);
2386 BUG_ON(err < 0);
2387
2388 ovs_unlock();
2389 ovs_notify(&dp_vport_genl_family, reply, info);
2390 return 0;
2391
2392exit_unlock_free:
2393 ovs_unlock();
2394 kfree_skb(reply);
2395 return err;
2396}
2397
2398static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2399{
2400 bool update_headroom = false;
2401 struct nlattr **a = info->attrs;
2402 struct sk_buff *reply;
2403 struct datapath *dp;
2404 struct vport *vport;
2405 unsigned int new_headroom;
2406 int err;
2407
2408 reply = ovs_vport_cmd_alloc_info();
2409 if (!reply)
2410 return -ENOMEM;
2411
2412 ovs_lock();
2413 vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2414 err = PTR_ERR(vport);
2415 if (IS_ERR(vport))
2416 goto exit_unlock_free;
2417
2418 if (vport->port_no == OVSP_LOCAL) {
2419 err = -EINVAL;
2420 goto exit_unlock_free;
2421 }
2422
2423 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2424 info->snd_portid, info->snd_seq, 0,
2425 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2426 BUG_ON(err < 0);
2427
2428 /* the vport deletion may trigger dp headroom update */
2429 dp = vport->dp;
2430 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2431 update_headroom = true;
2432
2433 netdev_reset_rx_headroom(vport->dev);
2434 ovs_dp_detach_port(vport);
2435
2436 if (update_headroom) {
2437 new_headroom = ovs_get_max_headroom(dp);
2438
2439 if (new_headroom < dp->max_headroom)
2440 ovs_update_headroom(dp, new_headroom);
2441 }
2442 ovs_unlock();
2443
2444 ovs_notify(&dp_vport_genl_family, reply, info);
2445 return 0;
2446
2447exit_unlock_free:
2448 ovs_unlock();
2449 kfree_skb(reply);
2450 return err;
2451}
2452
2453static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2454{
2455 struct nlattr **a = info->attrs;
2456 struct ovs_header *ovs_header = genl_info_userhdr(info);
2457 struct sk_buff *reply;
2458 struct vport *vport;
2459 int err;
2460
2461 reply = ovs_vport_cmd_alloc_info();
2462 if (!reply)
2463 return -ENOMEM;
2464
2465 rcu_read_lock();
2466 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2467 err = PTR_ERR(vport);
2468 if (IS_ERR(vport))
2469 goto exit_unlock_free;
2470 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2471 info->snd_portid, info->snd_seq, 0,
2472 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2473 BUG_ON(err < 0);
2474 rcu_read_unlock();
2475
2476 return genlmsg_reply(reply, info);
2477
2478exit_unlock_free:
2479 rcu_read_unlock();
2480 kfree_skb(reply);
2481 return err;
2482}
2483
2484static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2485{
2486 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2487 struct datapath *dp;
2488 int bucket = cb->args[0], skip = cb->args[1];
2489 int i, j = 0;
2490
2491 rcu_read_lock();
2492 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2493 if (!dp) {
2494 rcu_read_unlock();
2495 return -ENODEV;
2496 }
2497 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2498 struct vport *vport;
2499
2500 j = 0;
2501 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2502 if (j >= skip &&
2503 ovs_vport_cmd_fill_info(vport, skb,
2504 sock_net(skb->sk),
2505 NETLINK_CB(cb->skb).portid,
2506 cb->nlh->nlmsg_seq,
2507 NLM_F_MULTI,
2508 OVS_VPORT_CMD_GET,
2509 GFP_ATOMIC) < 0)
2510 goto out;
2511
2512 j++;
2513 }
2514 skip = 0;
2515 }
2516out:
2517 rcu_read_unlock();
2518
2519 cb->args[0] = i;
2520 cb->args[1] = j;
2521
2522 return skb->len;
2523}
2524
2525static void ovs_dp_masks_rebalance(struct work_struct *work)
2526{
2527 struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2528 masks_rebalance.work);
2529 struct datapath *dp;
2530
2531 ovs_lock();
2532
2533 list_for_each_entry(dp, &ovs_net->dps, list_node)
2534 ovs_flow_masks_rebalance(&dp->table);
2535
2536 ovs_unlock();
2537
2538 schedule_delayed_work(&ovs_net->masks_rebalance,
2539 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2540}
2541
2542static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2543 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2544 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2545 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2546 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2547 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2548 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2549 [OVS_VPORT_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2550 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2551 [OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
2552};
2553
2554static const struct genl_small_ops dp_vport_genl_ops[] = {
2555 { .cmd = OVS_VPORT_CMD_NEW,
2556 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2557 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2558 .doit = ovs_vport_cmd_new
2559 },
2560 { .cmd = OVS_VPORT_CMD_DEL,
2561 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2562 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2563 .doit = ovs_vport_cmd_del
2564 },
2565 { .cmd = OVS_VPORT_CMD_GET,
2566 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2567 .flags = 0, /* OK for unprivileged users. */
2568 .doit = ovs_vport_cmd_get,
2569 .dumpit = ovs_vport_cmd_dump
2570 },
2571 { .cmd = OVS_VPORT_CMD_SET,
2572 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2573 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2574 .doit = ovs_vport_cmd_set,
2575 },
2576};
2577
2578struct genl_family dp_vport_genl_family __ro_after_init = {
2579 .hdrsize = sizeof(struct ovs_header),
2580 .name = OVS_VPORT_FAMILY,
2581 .version = OVS_VPORT_VERSION,
2582 .maxattr = OVS_VPORT_ATTR_MAX,
2583 .policy = vport_policy,
2584 .netnsok = true,
2585 .parallel_ops = true,
2586 .small_ops = dp_vport_genl_ops,
2587 .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2588 .resv_start_op = OVS_VPORT_CMD_SET + 1,
2589 .mcgrps = &ovs_dp_vport_multicast_group,
2590 .n_mcgrps = 1,
2591 .module = THIS_MODULE,
2592};
2593
2594static struct genl_family * const dp_genl_families[] = {
2595 &dp_datapath_genl_family,
2596 &dp_vport_genl_family,
2597 &dp_flow_genl_family,
2598 &dp_packet_genl_family,
2599 &dp_meter_genl_family,
2600#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2601 &dp_ct_limit_genl_family,
2602#endif
2603};
2604
2605static void dp_unregister_genl(int n_families)
2606{
2607 int i;
2608
2609 for (i = 0; i < n_families; i++)
2610 genl_unregister_family(dp_genl_families[i]);
2611}
2612
2613static int __init dp_register_genl(void)
2614{
2615 int err;
2616 int i;
2617
2618 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2619
2620 err = genl_register_family(dp_genl_families[i]);
2621 if (err)
2622 goto error;
2623 }
2624
2625 return 0;
2626
2627error:
2628 dp_unregister_genl(i);
2629 return err;
2630}
2631
2632static int __net_init ovs_init_net(struct net *net)
2633{
2634 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2635 int err;
2636
2637 INIT_LIST_HEAD(&ovs_net->dps);
2638 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2639 INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2640
2641 err = ovs_ct_init(net);
2642 if (err)
2643 return err;
2644
2645 schedule_delayed_work(&ovs_net->masks_rebalance,
2646 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2647 return 0;
2648}
2649
2650static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2651 struct list_head *head)
2652{
2653 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2654 struct datapath *dp;
2655
2656 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2657 int i;
2658
2659 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2660 struct vport *vport;
2661
2662 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2663 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2664 continue;
2665
2666 if (dev_net(vport->dev) == dnet)
2667 list_add(&vport->detach_list, head);
2668 }
2669 }
2670 }
2671}
2672
2673static void __net_exit ovs_exit_net(struct net *dnet)
2674{
2675 struct datapath *dp, *dp_next;
2676 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2677 struct vport *vport, *vport_next;
2678 struct net *net;
2679 LIST_HEAD(head);
2680
2681 ovs_lock();
2682
2683 ovs_ct_exit(dnet);
2684
2685 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2686 __dp_destroy(dp);
2687
2688 down_read(&net_rwsem);
2689 for_each_net(net)
2690 list_vports_from_net(net, dnet, &head);
2691 up_read(&net_rwsem);
2692
2693 /* Detach all vports from given namespace. */
2694 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2695 list_del(&vport->detach_list);
2696 ovs_dp_detach_port(vport);
2697 }
2698
2699 ovs_unlock();
2700
2701 cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2702 cancel_work_sync(&ovs_net->dp_notify_work);
2703}
2704
2705static struct pernet_operations ovs_net_ops = {
2706 .init = ovs_init_net,
2707 .exit = ovs_exit_net,
2708 .id = &ovs_net_id,
2709 .size = sizeof(struct ovs_net),
2710};
2711
2712static const char * const ovs_drop_reasons[] = {
2713#define S(x) [(x) & ~SKB_DROP_REASON_SUBSYS_MASK] = (#x),
2714 OVS_DROP_REASONS(S)
2715#undef S
2716};
2717
2718static struct drop_reason_list drop_reason_list_ovs = {
2719 .reasons = ovs_drop_reasons,
2720 .n_reasons = ARRAY_SIZE(ovs_drop_reasons),
2721};
2722
2723static int __init dp_init(void)
2724{
2725 int err;
2726
2727 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2728 sizeof_field(struct sk_buff, cb));
2729
2730 pr_info("Open vSwitch switching datapath\n");
2731
2732 err = action_fifos_init();
2733 if (err)
2734 goto error;
2735
2736 err = ovs_internal_dev_rtnl_link_register();
2737 if (err)
2738 goto error_action_fifos_exit;
2739
2740 err = ovs_flow_init();
2741 if (err)
2742 goto error_unreg_rtnl_link;
2743
2744 err = ovs_vport_init();
2745 if (err)
2746 goto error_flow_exit;
2747
2748 err = register_pernet_device(&ovs_net_ops);
2749 if (err)
2750 goto error_vport_exit;
2751
2752 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2753 if (err)
2754 goto error_netns_exit;
2755
2756 err = ovs_netdev_init();
2757 if (err)
2758 goto error_unreg_notifier;
2759
2760 err = dp_register_genl();
2761 if (err < 0)
2762 goto error_unreg_netdev;
2763
2764 drop_reasons_register_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH,
2765 &drop_reason_list_ovs);
2766
2767 return 0;
2768
2769error_unreg_netdev:
2770 ovs_netdev_exit();
2771error_unreg_notifier:
2772 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2773error_netns_exit:
2774 unregister_pernet_device(&ovs_net_ops);
2775error_vport_exit:
2776 ovs_vport_exit();
2777error_flow_exit:
2778 ovs_flow_exit();
2779error_unreg_rtnl_link:
2780 ovs_internal_dev_rtnl_link_unregister();
2781error_action_fifos_exit:
2782 action_fifos_exit();
2783error:
2784 return err;
2785}
2786
2787static void dp_cleanup(void)
2788{
2789 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2790 ovs_netdev_exit();
2791 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2792 unregister_pernet_device(&ovs_net_ops);
2793 drop_reasons_unregister_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH);
2794 rcu_barrier();
2795 ovs_vport_exit();
2796 ovs_flow_exit();
2797 ovs_internal_dev_rtnl_link_unregister();
2798 action_fifos_exit();
2799}
2800
2801module_init(dp_init);
2802module_exit(dp_cleanup);
2803
2804MODULE_DESCRIPTION("Open vSwitch switching datapath");
2805MODULE_LICENSE("GPL");
2806MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2807MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2808MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2809MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2810MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2811MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);