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