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