projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
openvswitch: Constify various function arguments
[deliverable/linux.git] / net / openvswitch / datapath.c
1 /*
2 * Copyright (c) 2007-2014 Nicira, Inc.
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 <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
53
54 #include "datapath.h"
55 #include "flow.h"
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
60
61 int ovs_net_id __read_mostly;
62 EXPORT_SYMBOL_GPL(ovs_net_id);
63
64 static struct genl_family dp_packet_genl_family;
65 static struct genl_family dp_flow_genl_family;
66 static struct genl_family dp_datapath_genl_family;
67
68 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
69 .name = OVS_FLOW_MCGROUP,
70 };
71
72 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
73 .name = OVS_DATAPATH_MCGROUP,
74 };
75
76 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
77 .name = OVS_VPORT_MCGROUP,
78 };
79
80 /* Check if need to build a reply message.
81 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
82 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
83 unsigned int group)
84 {
85 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
86 genl_has_listeners(family, genl_info_net(info)->genl_sock,
87 group);
88 }
89
90 static void ovs_notify(struct genl_family *family,
91 struct sk_buff *skb, struct genl_info *info)
92 {
93 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
94 0, info->nlhdr, GFP_KERNEL);
95 }
96
97 /**
98 * DOC: Locking:
99 *
100 * All writes e.g. Writes to device state (add/remove datapath, port, set
101 * operations on vports, etc.), Writes to other state (flow table
102 * modifications, set miscellaneous datapath parameters, etc.) are protected
103 * by ovs_lock.
104 *
105 * Reads are protected by RCU.
106 *
107 * There are a few special cases (mostly stats) that have their own
108 * synchronization but they nest under all of above and don't interact with
109 * each other.
110 *
111 * The RTNL lock nests inside ovs_mutex.
112 */
113
114 static DEFINE_MUTEX(ovs_mutex);
115
116 void ovs_lock(void)
117 {
118 mutex_lock(&ovs_mutex);
119 }
120
121 void ovs_unlock(void)
122 {
123 mutex_unlock(&ovs_mutex);
124 }
125
126 #ifdef CONFIG_LOCKDEP
127 int lockdep_ovsl_is_held(void)
128 {
129 if (debug_locks)
130 return lockdep_is_held(&ovs_mutex);
131 else
132 return 1;
133 }
134 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
135 #endif
136
137 static struct vport *new_vport(const struct vport_parms *);
138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 const struct sw_flow_key *,
140 const struct dp_upcall_info *);
141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
142 const struct sw_flow_key *,
143 const struct dp_upcall_info *);
144
145 /* Must be called with rcu_read_lock. */
146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
147 {
148 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
149
150 if (dev) {
151 struct vport *vport = ovs_internal_dev_get_vport(dev);
152 if (vport)
153 return vport->dp;
154 }
155
156 return NULL;
157 }
158
159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160 * returned dp pointer valid.
161 */
162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
163 {
164 struct datapath *dp;
165
166 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
167 rcu_read_lock();
168 dp = get_dp_rcu(net, dp_ifindex);
169 rcu_read_unlock();
170
171 return dp;
172 }
173
174 /* Must be called with rcu_read_lock or ovs_mutex. */
175 const char *ovs_dp_name(const struct datapath *dp)
176 {
177 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
178 return vport->ops->get_name(vport);
179 }
180
181 static int get_dpifindex(const struct datapath *dp)
182 {
183 struct vport *local;
184 int ifindex;
185
186 rcu_read_lock();
187
188 local = ovs_vport_rcu(dp, OVSP_LOCAL);
189 if (local)
190 ifindex = netdev_vport_priv(local)->dev->ifindex;
191 else
192 ifindex = 0;
193
194 rcu_read_unlock();
195
196 return ifindex;
197 }
198
199 static void destroy_dp_rcu(struct rcu_head *rcu)
200 {
201 struct datapath *dp = container_of(rcu, struct datapath, rcu);
202
203 ovs_flow_tbl_destroy(&dp->table);
204 free_percpu(dp->stats_percpu);
205 release_net(ovs_dp_get_net(dp));
206 kfree(dp->ports);
207 kfree(dp);
208 }
209
210 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
211 u16 port_no)
212 {
213 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
214 }
215
216 /* Called with ovs_mutex or RCU read lock. */
217 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
218 {
219 struct vport *vport;
220 struct hlist_head *head;
221
222 head = vport_hash_bucket(dp, port_no);
223 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
224 if (vport->port_no == port_no)
225 return vport;
226 }
227 return NULL;
228 }
229
230 /* Called with ovs_mutex. */
231 static struct vport *new_vport(const struct vport_parms *parms)
232 {
233 struct vport *vport;
234
235 vport = ovs_vport_add(parms);
236 if (!IS_ERR(vport)) {
237 struct datapath *dp = parms->dp;
238 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
239
240 hlist_add_head_rcu(&vport->dp_hash_node, head);
241 }
242 return vport;
243 }
244
245 void ovs_dp_detach_port(struct vport *p)
246 {
247 ASSERT_OVSL();
248
249 /* First drop references to device. */
250 hlist_del_rcu(&p->dp_hash_node);
251
252 /* Then destroy it. */
253 ovs_vport_del(p);
254 }
255
256 /* Must be called with rcu_read_lock. */
257 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
258 {
259 const struct vport *p = OVS_CB(skb)->input_vport;
260 struct datapath *dp = p->dp;
261 struct sw_flow *flow;
262 struct sw_flow_actions *sf_acts;
263 struct dp_stats_percpu *stats;
264 u64 *stats_counter;
265 u32 n_mask_hit;
266
267 stats = this_cpu_ptr(dp->stats_percpu);
268
269 /* Look up flow. */
270 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
271 if (unlikely(!flow)) {
272 struct dp_upcall_info upcall;
273 int error;
274
275 upcall.cmd = OVS_PACKET_CMD_MISS;
276 upcall.userdata = NULL;
277 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
278 upcall.egress_tun_info = NULL;
279 error = ovs_dp_upcall(dp, skb, key, &upcall);
280 if (unlikely(error))
281 kfree_skb(skb);
282 else
283 consume_skb(skb);
284 stats_counter = &stats->n_missed;
285 goto out;
286 }
287
288 ovs_flow_stats_update(flow, key->tp.flags, skb);
289 sf_acts = rcu_dereference(flow->sf_acts);
290 ovs_execute_actions(dp, skb, sf_acts, key);
291
292 stats_counter = &stats->n_hit;
293
294 out:
295 /* Update datapath statistics. */
296 u64_stats_update_begin(&stats->syncp);
297 (*stats_counter)++;
298 stats->n_mask_hit += n_mask_hit;
299 u64_stats_update_end(&stats->syncp);
300 }
301
302 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
303 const struct sw_flow_key *key,
304 const struct dp_upcall_info *upcall_info)
305 {
306 struct dp_stats_percpu *stats;
307 int err;
308
309 if (upcall_info->portid == 0) {
310 err = -ENOTCONN;
311 goto err;
312 }
313
314 if (!skb_is_gso(skb))
315 err = queue_userspace_packet(dp, skb, key, upcall_info);
316 else
317 err = queue_gso_packets(dp, skb, key, upcall_info);
318 if (err)
319 goto err;
320
321 return 0;
322
323 err:
324 stats = this_cpu_ptr(dp->stats_percpu);
325
326 u64_stats_update_begin(&stats->syncp);
327 stats->n_lost++;
328 u64_stats_update_end(&stats->syncp);
329
330 return err;
331 }
332
333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
334 const struct sw_flow_key *key,
335 const struct dp_upcall_info *upcall_info)
336 {
337 unsigned short gso_type = skb_shinfo(skb)->gso_type;
338 struct sw_flow_key later_key;
339 struct sk_buff *segs, *nskb;
340 struct ovs_skb_cb ovs_cb;
341 int err;
342
343 ovs_cb = *OVS_CB(skb);
344 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
345 *OVS_CB(skb) = ovs_cb;
346 if (IS_ERR(segs))
347 return PTR_ERR(segs);
348 if (segs == NULL)
349 return -EINVAL;
350
351 if (gso_type & SKB_GSO_UDP) {
352 /* The initial flow key extracted by ovs_flow_key_extract()
353 * in this case is for a first fragment, so we need to
354 * properly mark later fragments.
355 */
356 later_key = *key;
357 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
358 }
359
360 /* Queue all of the segments. */
361 skb = segs;
362 do {
363 *OVS_CB(skb) = ovs_cb;
364 if (gso_type & SKB_GSO_UDP && skb != segs)
365 key = &later_key;
366
367 err = queue_userspace_packet(dp, skb, key, upcall_info);
368 if (err)
369 break;
370
371 } while ((skb = skb->next));
372
373 /* Free all of the segments. */
374 skb = segs;
375 do {
376 nskb = skb->next;
377 if (err)
378 kfree_skb(skb);
379 else
380 consume_skb(skb);
381 } while ((skb = nskb));
382 return err;
383 }
384
385 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
386 unsigned int hdrlen)
387 {
388 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
389 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
390 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391
392 /* OVS_PACKET_ATTR_USERDATA */
393 if (upcall_info->userdata)
394 size += NLA_ALIGN(upcall_info->userdata->nla_len);
395
396 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
397 if (upcall_info->egress_tun_info)
398 size += nla_total_size(ovs_tun_key_attr_size());
399
400 return size;
401 }
402
403 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
404 const struct sw_flow_key *key,
405 const struct dp_upcall_info *upcall_info)
406 {
407 struct ovs_header *upcall;
408 struct sk_buff *nskb = NULL;
409 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
410 struct nlattr *nla;
411 struct genl_info info = {
412 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
413 .snd_portid = upcall_info->portid,
414 };
415 size_t len;
416 unsigned int hlen;
417 int err, dp_ifindex;
418
419 dp_ifindex = get_dpifindex(dp);
420 if (!dp_ifindex)
421 return -ENODEV;
422
423 if (vlan_tx_tag_present(skb)) {
424 nskb = skb_clone(skb, GFP_ATOMIC);
425 if (!nskb)
426 return -ENOMEM;
427
428 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
429 if (!nskb)
430 return -ENOMEM;
431
432 nskb->vlan_tci = 0;
433 skb = nskb;
434 }
435
436 if (nla_attr_size(skb->len) > USHRT_MAX) {
437 err = -EFBIG;
438 goto out;
439 }
440
441 /* Complete checksum if needed */
442 if (skb->ip_summed == CHECKSUM_PARTIAL &&
443 (err = skb_checksum_help(skb)))
444 goto out;
445
446 /* Older versions of OVS user space enforce alignment of the last
447 * Netlink attribute to NLA_ALIGNTO which would require extensive
448 * padding logic. Only perform zerocopy if padding is not required.
449 */
450 if (dp->user_features & OVS_DP_F_UNALIGNED)
451 hlen = skb_zerocopy_headlen(skb);
452 else
453 hlen = skb->len;
454
455 len = upcall_msg_size(upcall_info, hlen);
456 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
457 if (!user_skb) {
458 err = -ENOMEM;
459 goto out;
460 }
461
462 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
463 0, upcall_info->cmd);
464 upcall->dp_ifindex = dp_ifindex;
465
466 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
467 err = ovs_nla_put_flow(key, key, user_skb);
468 BUG_ON(err);
469 nla_nest_end(user_skb, nla);
470
471 if (upcall_info->userdata)
472 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
473 nla_len(upcall_info->userdata),
474 nla_data(upcall_info->userdata));
475
476 if (upcall_info->egress_tun_info) {
477 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
478 err = ovs_nla_put_egress_tunnel_key(user_skb,
479 upcall_info->egress_tun_info);
480 BUG_ON(err);
481 nla_nest_end(user_skb, nla);
482 }
483
484 /* Only reserve room for attribute header, packet data is added
485 * in skb_zerocopy() */
486 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
487 err = -ENOBUFS;
488 goto out;
489 }
490 nla->nla_len = nla_attr_size(skb->len);
491
492 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
493 if (err)
494 goto out;
495
496 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
497 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
498 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
499
500 if (plen > 0)
501 memset(skb_put(user_skb, plen), 0, plen);
502 }
503
504 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
505
506 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
507 user_skb = NULL;
508 out:
509 if (err)
510 skb_tx_error(skb);
511 kfree_skb(user_skb);
512 kfree_skb(nskb);
513 return err;
514 }
515
516 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
517 {
518 struct ovs_header *ovs_header = info->userhdr;
519 struct nlattr **a = info->attrs;
520 struct sw_flow_actions *acts;
521 struct sk_buff *packet;
522 struct sw_flow *flow;
523 struct sw_flow_actions *sf_acts;
524 struct datapath *dp;
525 struct ethhdr *eth;
526 struct vport *input_vport;
527 int len;
528 int err;
529
530 err = -EINVAL;
531 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
532 !a[OVS_PACKET_ATTR_ACTIONS])
533 goto err;
534
535 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
536 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
537 err = -ENOMEM;
538 if (!packet)
539 goto err;
540 skb_reserve(packet, NET_IP_ALIGN);
541
542 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
543
544 skb_reset_mac_header(packet);
545 eth = eth_hdr(packet);
546
547 /* Normally, setting the skb 'protocol' field would be handled by a
548 * call to eth_type_trans(), but it assumes there's a sending
549 * device, which we may not have. */
550 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
551 packet->protocol = eth->h_proto;
552 else
553 packet->protocol = htons(ETH_P_802_2);
554
555 /* Build an sw_flow for sending this packet. */
556 flow = ovs_flow_alloc();
557 err = PTR_ERR(flow);
558 if (IS_ERR(flow))
559 goto err_kfree_skb;
560
561 err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
562 &flow->key);
563 if (err)
564 goto err_flow_free;
565
566 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
567 &flow->key, &acts);
568 if (err)
569 goto err_flow_free;
570
571 rcu_assign_pointer(flow->sf_acts, acts);
572 OVS_CB(packet)->egress_tun_info = NULL;
573 packet->priority = flow->key.phy.priority;
574 packet->mark = flow->key.phy.skb_mark;
575
576 rcu_read_lock();
577 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
578 err = -ENODEV;
579 if (!dp)
580 goto err_unlock;
581
582 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
583 if (!input_vport)
584 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
585
586 if (!input_vport)
587 goto err_unlock;
588
589 OVS_CB(packet)->input_vport = input_vport;
590 sf_acts = rcu_dereference(flow->sf_acts);
591
592 local_bh_disable();
593 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
594 local_bh_enable();
595 rcu_read_unlock();
596
597 ovs_flow_free(flow, false);
598 return err;
599
600 err_unlock:
601 rcu_read_unlock();
602 err_flow_free:
603 ovs_flow_free(flow, false);
604 err_kfree_skb:
605 kfree_skb(packet);
606 err:
607 return err;
608 }
609
610 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
611 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
612 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
613 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
614 };
615
616 static const struct genl_ops dp_packet_genl_ops[] = {
617 { .cmd = OVS_PACKET_CMD_EXECUTE,
618 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
619 .policy = packet_policy,
620 .doit = ovs_packet_cmd_execute
621 }
622 };
623
624 static struct genl_family dp_packet_genl_family = {
625 .id = GENL_ID_GENERATE,
626 .hdrsize = sizeof(struct ovs_header),
627 .name = OVS_PACKET_FAMILY,
628 .version = OVS_PACKET_VERSION,
629 .maxattr = OVS_PACKET_ATTR_MAX,
630 .netnsok = true,
631 .parallel_ops = true,
632 .ops = dp_packet_genl_ops,
633 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
634 };
635
636 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
637 struct ovs_dp_megaflow_stats *mega_stats)
638 {
639 int i;
640
641 memset(mega_stats, 0, sizeof(*mega_stats));
642
643 stats->n_flows = ovs_flow_tbl_count(&dp->table);
644 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
645
646 stats->n_hit = stats->n_missed = stats->n_lost = 0;
647
648 for_each_possible_cpu(i) {
649 const struct dp_stats_percpu *percpu_stats;
650 struct dp_stats_percpu local_stats;
651 unsigned int start;
652
653 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
654
655 do {
656 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
657 local_stats = *percpu_stats;
658 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
659
660 stats->n_hit += local_stats.n_hit;
661 stats->n_missed += local_stats.n_missed;
662 stats->n_lost += local_stats.n_lost;
663 mega_stats->n_mask_hit += local_stats.n_mask_hit;
664 }
665 }
666
667 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
668 {
669 return NLMSG_ALIGN(sizeof(struct ovs_header))
670 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
671 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
672 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
673 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
674 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
675 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
676 }
677
678 /* Called with ovs_mutex or RCU read lock. */
679 static int ovs_flow_cmd_fill_match(const struct sw_flow *flow,
680 struct sk_buff *skb)
681 {
682 struct nlattr *nla;
683 int err;
684
685 /* Fill flow key. */
686 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
687 if (!nla)
688 return -EMSGSIZE;
689
690 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
691 if (err)
692 return err;
693
694 nla_nest_end(skb, nla);
695
696 /* Fill flow mask. */
697 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
698 if (!nla)
699 return -EMSGSIZE;
700
701 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
702 if (err)
703 return err;
704
705 nla_nest_end(skb, nla);
706 return 0;
707 }
708
709 /* Called with ovs_mutex or RCU read lock. */
710 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
711 struct sk_buff *skb)
712 {
713 struct ovs_flow_stats stats;
714 __be16 tcp_flags;
715 unsigned long used;
716
717 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
718
719 if (used &&
720 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
721 return -EMSGSIZE;
722
723 if (stats.n_packets &&
724 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
725 return -EMSGSIZE;
726
727 if ((u8)ntohs(tcp_flags) &&
728 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
729 return -EMSGSIZE;
730
731 return 0;
732 }
733
734 /* Called with ovs_mutex or RCU read lock. */
735 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
736 struct sk_buff *skb, int skb_orig_len)
737 {
738 struct nlattr *start;
739 int err;
740
741 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
742 * this is the first flow to be dumped into 'skb'. This is unusual for
743 * Netlink but individual action lists can be longer than
744 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
745 * The userspace caller can always fetch the actions separately if it
746 * really wants them. (Most userspace callers in fact don't care.)
747 *
748 * This can only fail for dump operations because the skb is always
749 * properly sized for single flows.
750 */
751 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
752 if (start) {
753 const struct sw_flow_actions *sf_acts;
754
755 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
756 err = ovs_nla_put_actions(sf_acts->actions,
757 sf_acts->actions_len, skb);
758
759 if (!err)
760 nla_nest_end(skb, start);
761 else {
762 if (skb_orig_len)
763 return err;
764
765 nla_nest_cancel(skb, start);
766 }
767 } else if (skb_orig_len) {
768 return -EMSGSIZE;
769 }
770
771 return 0;
772 }
773
774 /* Called with ovs_mutex or RCU read lock. */
775 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
776 struct sk_buff *skb, u32 portid,
777 u32 seq, u32 flags, u8 cmd)
778 {
779 const int skb_orig_len = skb->len;
780 struct ovs_header *ovs_header;
781 int err;
782
783 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
784 flags, cmd);
785 if (!ovs_header)
786 return -EMSGSIZE;
787
788 ovs_header->dp_ifindex = dp_ifindex;
789
790 err = ovs_flow_cmd_fill_match(flow, skb);
791 if (err)
792 goto error;
793
794 err = ovs_flow_cmd_fill_stats(flow, skb);
795 if (err)
796 goto error;
797
798 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
799 if (err)
800 goto error;
801
802 return genlmsg_end(skb, ovs_header);
803
804 error:
805 genlmsg_cancel(skb, ovs_header);
806 return err;
807 }
808
809 /* May not be called with RCU read lock. */
810 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
811 struct genl_info *info,
812 bool always)
813 {
814 struct sk_buff *skb;
815
816 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
817 return NULL;
818
819 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
820 if (!skb)
821 return ERR_PTR(-ENOMEM);
822
823 return skb;
824 }
825
826 /* Called with ovs_mutex. */
827 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
828 int dp_ifindex,
829 struct genl_info *info, u8 cmd,
830 bool always)
831 {
832 struct sk_buff *skb;
833 int retval;
834
835 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
836 always);
837 if (IS_ERR_OR_NULL(skb))
838 return skb;
839
840 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
841 info->snd_portid, info->snd_seq, 0,
842 cmd);
843 BUG_ON(retval < 0);
844 return skb;
845 }
846
847 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
848 {
849 struct nlattr **a = info->attrs;
850 struct ovs_header *ovs_header = info->userhdr;
851 struct sw_flow *flow, *new_flow;
852 struct sw_flow_mask mask;
853 struct sk_buff *reply;
854 struct datapath *dp;
855 struct sw_flow_actions *acts;
856 struct sw_flow_match match;
857 int error;
858
859 /* Must have key and actions. */
860 error = -EINVAL;
861 if (!a[OVS_FLOW_ATTR_KEY]) {
862 OVS_NLERR("Flow key attribute not present in new flow.\n");
863 goto error;
864 }
865 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
866 OVS_NLERR("Flow actions attribute not present in new flow.\n");
867 goto error;
868 }
869
870 /* Most of the time we need to allocate a new flow, do it before
871 * locking.
872 */
873 new_flow = ovs_flow_alloc();
874 if (IS_ERR(new_flow)) {
875 error = PTR_ERR(new_flow);
876 goto error;
877 }
878
879 /* Extract key. */
880 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
881 error = ovs_nla_get_match(&match,
882 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
883 if (error)
884 goto err_kfree_flow;
885
886 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
887
888 /* Validate actions. */
889 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
890 &acts);
891 if (error) {
892 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
893 goto err_kfree_flow;
894 }
895
896 reply = ovs_flow_cmd_alloc_info(acts, info, false);
897 if (IS_ERR(reply)) {
898 error = PTR_ERR(reply);
899 goto err_kfree_acts;
900 }
901
902 ovs_lock();
903 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
904 if (unlikely(!dp)) {
905 error = -ENODEV;
906 goto err_unlock_ovs;
907 }
908 /* Check if this is a duplicate flow */
909 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
910 if (likely(!flow)) {
911 rcu_assign_pointer(new_flow->sf_acts, acts);
912
913 /* Put flow in bucket. */
914 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
915 if (unlikely(error)) {
916 acts = NULL;
917 goto err_unlock_ovs;
918 }
919
920 if (unlikely(reply)) {
921 error = ovs_flow_cmd_fill_info(new_flow,
922 ovs_header->dp_ifindex,
923 reply, info->snd_portid,
924 info->snd_seq, 0,
925 OVS_FLOW_CMD_NEW);
926 BUG_ON(error < 0);
927 }
928 ovs_unlock();
929 } else {
930 struct sw_flow_actions *old_acts;
931
932 /* Bail out if we're not allowed to modify an existing flow.
933 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
934 * because Generic Netlink treats the latter as a dump
935 * request. We also accept NLM_F_EXCL in case that bug ever
936 * gets fixed.
937 */
938 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
939 | NLM_F_EXCL))) {
940 error = -EEXIST;
941 goto err_unlock_ovs;
942 }
943 /* The unmasked key has to be the same for flow updates. */
944 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
945 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
946 if (!flow) {
947 error = -ENOENT;
948 goto err_unlock_ovs;
949 }
950 }
951 /* Update actions. */
952 old_acts = ovsl_dereference(flow->sf_acts);
953 rcu_assign_pointer(flow->sf_acts, acts);
954
955 if (unlikely(reply)) {
956 error = ovs_flow_cmd_fill_info(flow,
957 ovs_header->dp_ifindex,
958 reply, info->snd_portid,
959 info->snd_seq, 0,
960 OVS_FLOW_CMD_NEW);
961 BUG_ON(error < 0);
962 }
963 ovs_unlock();
964
965 ovs_nla_free_flow_actions(old_acts);
966 ovs_flow_free(new_flow, false);
967 }
968
969 if (reply)
970 ovs_notify(&dp_flow_genl_family, reply, info);
971 return 0;
972
973 err_unlock_ovs:
974 ovs_unlock();
975 kfree_skb(reply);
976 err_kfree_acts:
977 kfree(acts);
978 err_kfree_flow:
979 ovs_flow_free(new_flow, false);
980 error:
981 return error;
982 }
983
984 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
985 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
986 const struct sw_flow_key *key,
987 const struct sw_flow_mask *mask)
988 {
989 struct sw_flow_actions *acts;
990 struct sw_flow_key masked_key;
991 int error;
992
993 ovs_flow_mask_key(&masked_key, key, mask);
994 error = ovs_nla_copy_actions(a, &masked_key, &acts);
995 if (error) {
996 OVS_NLERR("Actions may not be safe on all matching packets.\n");
997 return ERR_PTR(error);
998 }
999
1000 return acts;
1001 }
1002
1003 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1004 {
1005 struct nlattr **a = info->attrs;
1006 struct ovs_header *ovs_header = info->userhdr;
1007 struct sw_flow_key key;
1008 struct sw_flow *flow;
1009 struct sw_flow_mask mask;
1010 struct sk_buff *reply = NULL;
1011 struct datapath *dp;
1012 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1013 struct sw_flow_match match;
1014 int error;
1015
1016 /* Extract key. */
1017 error = -EINVAL;
1018 if (!a[OVS_FLOW_ATTR_KEY]) {
1019 OVS_NLERR("Flow key attribute not present in set flow.\n");
1020 goto error;
1021 }
1022
1023 ovs_match_init(&match, &key, &mask);
1024 error = ovs_nla_get_match(&match,
1025 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1026 if (error)
1027 goto error;
1028
1029 /* Validate actions. */
1030 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1031 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask);
1032 if (IS_ERR(acts)) {
1033 error = PTR_ERR(acts);
1034 goto error;
1035 }
1036
1037 /* Can allocate before locking if have acts. */
1038 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1039 if (IS_ERR(reply)) {
1040 error = PTR_ERR(reply);
1041 goto err_kfree_acts;
1042 }
1043 }
1044
1045 ovs_lock();
1046 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1047 if (unlikely(!dp)) {
1048 error = -ENODEV;
1049 goto err_unlock_ovs;
1050 }
1051 /* Check that the flow exists. */
1052 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1053 if (unlikely(!flow)) {
1054 error = -ENOENT;
1055 goto err_unlock_ovs;
1056 }
1057
1058 /* Update actions, if present. */
1059 if (likely(acts)) {
1060 old_acts = ovsl_dereference(flow->sf_acts);
1061 rcu_assign_pointer(flow->sf_acts, acts);
1062
1063 if (unlikely(reply)) {
1064 error = ovs_flow_cmd_fill_info(flow,
1065 ovs_header->dp_ifindex,
1066 reply, info->snd_portid,
1067 info->snd_seq, 0,
1068 OVS_FLOW_CMD_NEW);
1069 BUG_ON(error < 0);
1070 }
1071 } else {
1072 /* Could not alloc without acts before locking. */
1073 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1074 info, OVS_FLOW_CMD_NEW, false);
1075 if (unlikely(IS_ERR(reply))) {
1076 error = PTR_ERR(reply);
1077 goto err_unlock_ovs;
1078 }
1079 }
1080
1081 /* Clear stats. */
1082 if (a[OVS_FLOW_ATTR_CLEAR])
1083 ovs_flow_stats_clear(flow);
1084 ovs_unlock();
1085
1086 if (reply)
1087 ovs_notify(&dp_flow_genl_family, reply, info);
1088 if (old_acts)
1089 ovs_nla_free_flow_actions(old_acts);
1090
1091 return 0;
1092
1093 err_unlock_ovs:
1094 ovs_unlock();
1095 kfree_skb(reply);
1096 err_kfree_acts:
1097 kfree(acts);
1098 error:
1099 return error;
1100 }
1101
1102 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1103 {
1104 struct nlattr **a = info->attrs;
1105 struct ovs_header *ovs_header = info->userhdr;
1106 struct sw_flow_key key;
1107 struct sk_buff *reply;
1108 struct sw_flow *flow;
1109 struct datapath *dp;
1110 struct sw_flow_match match;
1111 int err;
1112
1113 if (!a[OVS_FLOW_ATTR_KEY]) {
1114 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1115 return -EINVAL;
1116 }
1117
1118 ovs_match_init(&match, &key, NULL);
1119 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1120 if (err)
1121 return err;
1122
1123 ovs_lock();
1124 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1125 if (!dp) {
1126 err = -ENODEV;
1127 goto unlock;
1128 }
1129
1130 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1131 if (!flow) {
1132 err = -ENOENT;
1133 goto unlock;
1134 }
1135
1136 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1137 OVS_FLOW_CMD_NEW, true);
1138 if (IS_ERR(reply)) {
1139 err = PTR_ERR(reply);
1140 goto unlock;
1141 }
1142
1143 ovs_unlock();
1144 return genlmsg_reply(reply, info);
1145 unlock:
1146 ovs_unlock();
1147 return err;
1148 }
1149
1150 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1151 {
1152 struct nlattr **a = info->attrs;
1153 struct ovs_header *ovs_header = info->userhdr;
1154 struct sw_flow_key key;
1155 struct sk_buff *reply;
1156 struct sw_flow *flow;
1157 struct datapath *dp;
1158 struct sw_flow_match match;
1159 int err;
1160
1161 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1162 ovs_match_init(&match, &key, NULL);
1163 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1164 if (unlikely(err))
1165 return err;
1166 }
1167
1168 ovs_lock();
1169 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1170 if (unlikely(!dp)) {
1171 err = -ENODEV;
1172 goto unlock;
1173 }
1174
1175 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1176 err = ovs_flow_tbl_flush(&dp->table);
1177 goto unlock;
1178 }
1179
1180 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1181 if (unlikely(!flow)) {
1182 err = -ENOENT;
1183 goto unlock;
1184 }
1185
1186 ovs_flow_tbl_remove(&dp->table, flow);
1187 ovs_unlock();
1188
1189 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1190 info, false);
1191 if (likely(reply)) {
1192 if (likely(!IS_ERR(reply))) {
1193 rcu_read_lock(); /*To keep RCU checker happy. */
1194 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1195 reply, info->snd_portid,
1196 info->snd_seq, 0,
1197 OVS_FLOW_CMD_DEL);
1198 rcu_read_unlock();
1199 BUG_ON(err < 0);
1200
1201 ovs_notify(&dp_flow_genl_family, reply, info);
1202 } else {
1203 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1204 }
1205 }
1206
1207 ovs_flow_free(flow, true);
1208 return 0;
1209 unlock:
1210 ovs_unlock();
1211 return err;
1212 }
1213
1214 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1215 {
1216 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1217 struct table_instance *ti;
1218 struct datapath *dp;
1219
1220 rcu_read_lock();
1221 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1222 if (!dp) {
1223 rcu_read_unlock();
1224 return -ENODEV;
1225 }
1226
1227 ti = rcu_dereference(dp->table.ti);
1228 for (;;) {
1229 struct sw_flow *flow;
1230 u32 bucket, obj;
1231
1232 bucket = cb->args[0];
1233 obj = cb->args[1];
1234 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1235 if (!flow)
1236 break;
1237
1238 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1239 NETLINK_CB(cb->skb).portid,
1240 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1241 OVS_FLOW_CMD_NEW) < 0)
1242 break;
1243
1244 cb->args[0] = bucket;
1245 cb->args[1] = obj;
1246 }
1247 rcu_read_unlock();
1248 return skb->len;
1249 }
1250
1251 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1252 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1253 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1254 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1255 };
1256
1257 static const struct genl_ops dp_flow_genl_ops[] = {
1258 { .cmd = OVS_FLOW_CMD_NEW,
1259 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1260 .policy = flow_policy,
1261 .doit = ovs_flow_cmd_new
1262 },
1263 { .cmd = OVS_FLOW_CMD_DEL,
1264 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1265 .policy = flow_policy,
1266 .doit = ovs_flow_cmd_del
1267 },
1268 { .cmd = OVS_FLOW_CMD_GET,
1269 .flags = 0, /* OK for unprivileged users. */
1270 .policy = flow_policy,
1271 .doit = ovs_flow_cmd_get,
1272 .dumpit = ovs_flow_cmd_dump
1273 },
1274 { .cmd = OVS_FLOW_CMD_SET,
1275 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1276 .policy = flow_policy,
1277 .doit = ovs_flow_cmd_set,
1278 },
1279 };
1280
1281 static struct genl_family dp_flow_genl_family = {
1282 .id = GENL_ID_GENERATE,
1283 .hdrsize = sizeof(struct ovs_header),
1284 .name = OVS_FLOW_FAMILY,
1285 .version = OVS_FLOW_VERSION,
1286 .maxattr = OVS_FLOW_ATTR_MAX,
1287 .netnsok = true,
1288 .parallel_ops = true,
1289 .ops = dp_flow_genl_ops,
1290 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1291 .mcgrps = &ovs_dp_flow_multicast_group,
1292 .n_mcgrps = 1,
1293 };
1294
1295 static size_t ovs_dp_cmd_msg_size(void)
1296 {
1297 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1298
1299 msgsize += nla_total_size(IFNAMSIZ);
1300 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1301 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1302 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1303
1304 return msgsize;
1305 }
1306
1307 /* Called with ovs_mutex or RCU read lock. */
1308 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1309 u32 portid, u32 seq, u32 flags, u8 cmd)
1310 {
1311 struct ovs_header *ovs_header;
1312 struct ovs_dp_stats dp_stats;
1313 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1314 int err;
1315
1316 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1317 flags, cmd);
1318 if (!ovs_header)
1319 goto error;
1320
1321 ovs_header->dp_ifindex = get_dpifindex(dp);
1322
1323 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1324 if (err)
1325 goto nla_put_failure;
1326
1327 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1328 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1329 &dp_stats))
1330 goto nla_put_failure;
1331
1332 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1333 sizeof(struct ovs_dp_megaflow_stats),
1334 &dp_megaflow_stats))
1335 goto nla_put_failure;
1336
1337 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1338 goto nla_put_failure;
1339
1340 return genlmsg_end(skb, ovs_header);
1341
1342 nla_put_failure:
1343 genlmsg_cancel(skb, ovs_header);
1344 error:
1345 return -EMSGSIZE;
1346 }
1347
1348 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1349 {
1350 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1351 }
1352
1353 /* Called with rcu_read_lock or ovs_mutex. */
1354 static struct datapath *lookup_datapath(struct net *net,
1355 const struct ovs_header *ovs_header,
1356 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1357 {
1358 struct datapath *dp;
1359
1360 if (!a[OVS_DP_ATTR_NAME])
1361 dp = get_dp(net, ovs_header->dp_ifindex);
1362 else {
1363 struct vport *vport;
1364
1365 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1366 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1367 }
1368 return dp ? dp : ERR_PTR(-ENODEV);
1369 }
1370
1371 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1372 {
1373 struct datapath *dp;
1374
1375 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1376 if (IS_ERR(dp))
1377 return;
1378
1379 WARN(dp->user_features, "Dropping previously announced user features\n");
1380 dp->user_features = 0;
1381 }
1382
1383 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1384 {
1385 if (a[OVS_DP_ATTR_USER_FEATURES])
1386 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1387 }
1388
1389 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1390 {
1391 struct nlattr **a = info->attrs;
1392 struct vport_parms parms;
1393 struct sk_buff *reply;
1394 struct datapath *dp;
1395 struct vport *vport;
1396 struct ovs_net *ovs_net;
1397 int err, i;
1398
1399 err = -EINVAL;
1400 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1401 goto err;
1402
1403 reply = ovs_dp_cmd_alloc_info(info);
1404 if (!reply)
1405 return -ENOMEM;
1406
1407 err = -ENOMEM;
1408 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1409 if (dp == NULL)
1410 goto err_free_reply;
1411
1412 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1413
1414 /* Allocate table. */
1415 err = ovs_flow_tbl_init(&dp->table);
1416 if (err)
1417 goto err_free_dp;
1418
1419 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1420 if (!dp->stats_percpu) {
1421 err = -ENOMEM;
1422 goto err_destroy_table;
1423 }
1424
1425 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1426 GFP_KERNEL);
1427 if (!dp->ports) {
1428 err = -ENOMEM;
1429 goto err_destroy_percpu;
1430 }
1431
1432 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1433 INIT_HLIST_HEAD(&dp->ports[i]);
1434
1435 /* Set up our datapath device. */
1436 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1437 parms.type = OVS_VPORT_TYPE_INTERNAL;
1438 parms.options = NULL;
1439 parms.dp = dp;
1440 parms.port_no = OVSP_LOCAL;
1441 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1442
1443 ovs_dp_change(dp, a);
1444
1445 /* So far only local changes have been made, now need the lock. */
1446 ovs_lock();
1447
1448 vport = new_vport(&parms);
1449 if (IS_ERR(vport)) {
1450 err = PTR_ERR(vport);
1451 if (err == -EBUSY)
1452 err = -EEXIST;
1453
1454 if (err == -EEXIST) {
1455 /* An outdated user space instance that does not understand
1456 * the concept of user_features has attempted to create a new
1457 * datapath and is likely to reuse it. Drop all user features.
1458 */
1459 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1460 ovs_dp_reset_user_features(skb, info);
1461 }
1462
1463 goto err_destroy_ports_array;
1464 }
1465
1466 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1467 info->snd_seq, 0, OVS_DP_CMD_NEW);
1468 BUG_ON(err < 0);
1469
1470 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1471 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1472
1473 ovs_unlock();
1474
1475 ovs_notify(&dp_datapath_genl_family, reply, info);
1476 return 0;
1477
1478 err_destroy_ports_array:
1479 ovs_unlock();
1480 kfree(dp->ports);
1481 err_destroy_percpu:
1482 free_percpu(dp->stats_percpu);
1483 err_destroy_table:
1484 ovs_flow_tbl_destroy(&dp->table);
1485 err_free_dp:
1486 release_net(ovs_dp_get_net(dp));
1487 kfree(dp);
1488 err_free_reply:
1489 kfree_skb(reply);
1490 err:
1491 return err;
1492 }
1493
1494 /* Called with ovs_mutex. */
1495 static void __dp_destroy(struct datapath *dp)
1496 {
1497 int i;
1498
1499 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1500 struct vport *vport;
1501 struct hlist_node *n;
1502
1503 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1504 if (vport->port_no != OVSP_LOCAL)
1505 ovs_dp_detach_port(vport);
1506 }
1507
1508 list_del_rcu(&dp->list_node);
1509
1510 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1511 * all ports in datapath are destroyed first before freeing datapath.
1512 */
1513 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1514
1515 /* RCU destroy the flow table */
1516 call_rcu(&dp->rcu, destroy_dp_rcu);
1517 }
1518
1519 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1520 {
1521 struct sk_buff *reply;
1522 struct datapath *dp;
1523 int err;
1524
1525 reply = ovs_dp_cmd_alloc_info(info);
1526 if (!reply)
1527 return -ENOMEM;
1528
1529 ovs_lock();
1530 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1531 err = PTR_ERR(dp);
1532 if (IS_ERR(dp))
1533 goto err_unlock_free;
1534
1535 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1536 info->snd_seq, 0, OVS_DP_CMD_DEL);
1537 BUG_ON(err < 0);
1538
1539 __dp_destroy(dp);
1540 ovs_unlock();
1541
1542 ovs_notify(&dp_datapath_genl_family, reply, info);
1543
1544 return 0;
1545
1546 err_unlock_free:
1547 ovs_unlock();
1548 kfree_skb(reply);
1549 return err;
1550 }
1551
1552 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1553 {
1554 struct sk_buff *reply;
1555 struct datapath *dp;
1556 int err;
1557
1558 reply = ovs_dp_cmd_alloc_info(info);
1559 if (!reply)
1560 return -ENOMEM;
1561
1562 ovs_lock();
1563 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1564 err = PTR_ERR(dp);
1565 if (IS_ERR(dp))
1566 goto err_unlock_free;
1567
1568 ovs_dp_change(dp, info->attrs);
1569
1570 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1571 info->snd_seq, 0, OVS_DP_CMD_NEW);
1572 BUG_ON(err < 0);
1573
1574 ovs_unlock();
1575 ovs_notify(&dp_datapath_genl_family, reply, info);
1576
1577 return 0;
1578
1579 err_unlock_free:
1580 ovs_unlock();
1581 kfree_skb(reply);
1582 return err;
1583 }
1584
1585 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1586 {
1587 struct sk_buff *reply;
1588 struct datapath *dp;
1589 int err;
1590
1591 reply = ovs_dp_cmd_alloc_info(info);
1592 if (!reply)
1593 return -ENOMEM;
1594
1595 rcu_read_lock();
1596 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1597 if (IS_ERR(dp)) {
1598 err = PTR_ERR(dp);
1599 goto err_unlock_free;
1600 }
1601 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1602 info->snd_seq, 0, OVS_DP_CMD_NEW);
1603 BUG_ON(err < 0);
1604 rcu_read_unlock();
1605
1606 return genlmsg_reply(reply, info);
1607
1608 err_unlock_free:
1609 rcu_read_unlock();
1610 kfree_skb(reply);
1611 return err;
1612 }
1613
1614 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1615 {
1616 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1617 struct datapath *dp;
1618 int skip = cb->args[0];
1619 int i = 0;
1620
1621 rcu_read_lock();
1622 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1623 if (i >= skip &&
1624 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1625 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1626 OVS_DP_CMD_NEW) < 0)
1627 break;
1628 i++;
1629 }
1630 rcu_read_unlock();
1631
1632 cb->args[0] = i;
1633
1634 return skb->len;
1635 }
1636
1637 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1638 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1639 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1640 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1641 };
1642
1643 static const struct genl_ops dp_datapath_genl_ops[] = {
1644 { .cmd = OVS_DP_CMD_NEW,
1645 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1646 .policy = datapath_policy,
1647 .doit = ovs_dp_cmd_new
1648 },
1649 { .cmd = OVS_DP_CMD_DEL,
1650 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1651 .policy = datapath_policy,
1652 .doit = ovs_dp_cmd_del
1653 },
1654 { .cmd = OVS_DP_CMD_GET,
1655 .flags = 0, /* OK for unprivileged users. */
1656 .policy = datapath_policy,
1657 .doit = ovs_dp_cmd_get,
1658 .dumpit = ovs_dp_cmd_dump
1659 },
1660 { .cmd = OVS_DP_CMD_SET,
1661 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1662 .policy = datapath_policy,
1663 .doit = ovs_dp_cmd_set,
1664 },
1665 };
1666
1667 static struct genl_family dp_datapath_genl_family = {
1668 .id = GENL_ID_GENERATE,
1669 .hdrsize = sizeof(struct ovs_header),
1670 .name = OVS_DATAPATH_FAMILY,
1671 .version = OVS_DATAPATH_VERSION,
1672 .maxattr = OVS_DP_ATTR_MAX,
1673 .netnsok = true,
1674 .parallel_ops = true,
1675 .ops = dp_datapath_genl_ops,
1676 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1677 .mcgrps = &ovs_dp_datapath_multicast_group,
1678 .n_mcgrps = 1,
1679 };
1680
1681 /* Called with ovs_mutex or RCU read lock. */
1682 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1683 u32 portid, u32 seq, u32 flags, u8 cmd)
1684 {
1685 struct ovs_header *ovs_header;
1686 struct ovs_vport_stats vport_stats;
1687 int err;
1688
1689 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1690 flags, cmd);
1691 if (!ovs_header)
1692 return -EMSGSIZE;
1693
1694 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1695
1696 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1697 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1698 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1699 vport->ops->get_name(vport)))
1700 goto nla_put_failure;
1701
1702 ovs_vport_get_stats(vport, &vport_stats);
1703 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1704 &vport_stats))
1705 goto nla_put_failure;
1706
1707 if (ovs_vport_get_upcall_portids(vport, skb))
1708 goto nla_put_failure;
1709
1710 err = ovs_vport_get_options(vport, skb);
1711 if (err == -EMSGSIZE)
1712 goto error;
1713
1714 return genlmsg_end(skb, ovs_header);
1715
1716 nla_put_failure:
1717 err = -EMSGSIZE;
1718 error:
1719 genlmsg_cancel(skb, ovs_header);
1720 return err;
1721 }
1722
1723 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1724 {
1725 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1726 }
1727
1728 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1729 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1730 u32 seq, u8 cmd)
1731 {
1732 struct sk_buff *skb;
1733 int retval;
1734
1735 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1736 if (!skb)
1737 return ERR_PTR(-ENOMEM);
1738
1739 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1740 BUG_ON(retval < 0);
1741
1742 return skb;
1743 }
1744
1745 /* Called with ovs_mutex or RCU read lock. */
1746 static struct vport *lookup_vport(struct net *net,
1747 const struct ovs_header *ovs_header,
1748 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1749 {
1750 struct datapath *dp;
1751 struct vport *vport;
1752
1753 if (a[OVS_VPORT_ATTR_NAME]) {
1754 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1755 if (!vport)
1756 return ERR_PTR(-ENODEV);
1757 if (ovs_header->dp_ifindex &&
1758 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1759 return ERR_PTR(-ENODEV);
1760 return vport;
1761 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1762 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1763
1764 if (port_no >= DP_MAX_PORTS)
1765 return ERR_PTR(-EFBIG);
1766
1767 dp = get_dp(net, ovs_header->dp_ifindex);
1768 if (!dp)
1769 return ERR_PTR(-ENODEV);
1770
1771 vport = ovs_vport_ovsl_rcu(dp, port_no);
1772 if (!vport)
1773 return ERR_PTR(-ENODEV);
1774 return vport;
1775 } else
1776 return ERR_PTR(-EINVAL);
1777 }
1778
1779 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1780 {
1781 struct nlattr **a = info->attrs;
1782 struct ovs_header *ovs_header = info->userhdr;
1783 struct vport_parms parms;
1784 struct sk_buff *reply;
1785 struct vport *vport;
1786 struct datapath *dp;
1787 u32 port_no;
1788 int err;
1789
1790 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1791 !a[OVS_VPORT_ATTR_UPCALL_PID])
1792 return -EINVAL;
1793
1794 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1795 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1796 if (port_no >= DP_MAX_PORTS)
1797 return -EFBIG;
1798
1799 reply = ovs_vport_cmd_alloc_info();
1800 if (!reply)
1801 return -ENOMEM;
1802
1803 ovs_lock();
1804 restart:
1805 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1806 err = -ENODEV;
1807 if (!dp)
1808 goto exit_unlock_free;
1809
1810 if (port_no) {
1811 vport = ovs_vport_ovsl(dp, port_no);
1812 err = -EBUSY;
1813 if (vport)
1814 goto exit_unlock_free;
1815 } else {
1816 for (port_no = 1; ; port_no++) {
1817 if (port_no >= DP_MAX_PORTS) {
1818 err = -EFBIG;
1819 goto exit_unlock_free;
1820 }
1821 vport = ovs_vport_ovsl(dp, port_no);
1822 if (!vport)
1823 break;
1824 }
1825 }
1826
1827 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1828 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1829 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1830 parms.dp = dp;
1831 parms.port_no = port_no;
1832 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1833
1834 vport = new_vport(&parms);
1835 err = PTR_ERR(vport);
1836 if (IS_ERR(vport)) {
1837 if (err == -EAGAIN)
1838 goto restart;
1839 goto exit_unlock_free;
1840 }
1841
1842 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1843 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1844 BUG_ON(err < 0);
1845 ovs_unlock();
1846
1847 ovs_notify(&dp_vport_genl_family, reply, info);
1848 return 0;
1849
1850 exit_unlock_free:
1851 ovs_unlock();
1852 kfree_skb(reply);
1853 return err;
1854 }
1855
1856 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1857 {
1858 struct nlattr **a = info->attrs;
1859 struct sk_buff *reply;
1860 struct vport *vport;
1861 int err;
1862
1863 reply = ovs_vport_cmd_alloc_info();
1864 if (!reply)
1865 return -ENOMEM;
1866
1867 ovs_lock();
1868 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1869 err = PTR_ERR(vport);
1870 if (IS_ERR(vport))
1871 goto exit_unlock_free;
1872
1873 if (a[OVS_VPORT_ATTR_TYPE] &&
1874 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1875 err = -EINVAL;
1876 goto exit_unlock_free;
1877 }
1878
1879 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1880 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1881 if (err)
1882 goto exit_unlock_free;
1883 }
1884
1885
1886 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1887 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1888
1889 err = ovs_vport_set_upcall_portids(vport, ids);
1890 if (err)
1891 goto exit_unlock_free;
1892 }
1893
1894 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1895 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1896 BUG_ON(err < 0);
1897
1898 ovs_unlock();
1899 ovs_notify(&dp_vport_genl_family, reply, info);
1900 return 0;
1901
1902 exit_unlock_free:
1903 ovs_unlock();
1904 kfree_skb(reply);
1905 return err;
1906 }
1907
1908 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1909 {
1910 struct nlattr **a = info->attrs;
1911 struct sk_buff *reply;
1912 struct vport *vport;
1913 int err;
1914
1915 reply = ovs_vport_cmd_alloc_info();
1916 if (!reply)
1917 return -ENOMEM;
1918
1919 ovs_lock();
1920 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1921 err = PTR_ERR(vport);
1922 if (IS_ERR(vport))
1923 goto exit_unlock_free;
1924
1925 if (vport->port_no == OVSP_LOCAL) {
1926 err = -EINVAL;
1927 goto exit_unlock_free;
1928 }
1929
1930 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1931 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1932 BUG_ON(err < 0);
1933 ovs_dp_detach_port(vport);
1934 ovs_unlock();
1935
1936 ovs_notify(&dp_vport_genl_family, reply, info);
1937 return 0;
1938
1939 exit_unlock_free:
1940 ovs_unlock();
1941 kfree_skb(reply);
1942 return err;
1943 }
1944
1945 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1946 {
1947 struct nlattr **a = info->attrs;
1948 struct ovs_header *ovs_header = info->userhdr;
1949 struct sk_buff *reply;
1950 struct vport *vport;
1951 int err;
1952
1953 reply = ovs_vport_cmd_alloc_info();
1954 if (!reply)
1955 return -ENOMEM;
1956
1957 rcu_read_lock();
1958 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1959 err = PTR_ERR(vport);
1960 if (IS_ERR(vport))
1961 goto exit_unlock_free;
1962 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1963 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1964 BUG_ON(err < 0);
1965 rcu_read_unlock();
1966
1967 return genlmsg_reply(reply, info);
1968
1969 exit_unlock_free:
1970 rcu_read_unlock();
1971 kfree_skb(reply);
1972 return err;
1973 }
1974
1975 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1976 {
1977 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1978 struct datapath *dp;
1979 int bucket = cb->args[0], skip = cb->args[1];
1980 int i, j = 0;
1981
1982 rcu_read_lock();
1983 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1984 if (!dp) {
1985 rcu_read_unlock();
1986 return -ENODEV;
1987 }
1988 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1989 struct vport *vport;
1990
1991 j = 0;
1992 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1993 if (j >= skip &&
1994 ovs_vport_cmd_fill_info(vport, skb,
1995 NETLINK_CB(cb->skb).portid,
1996 cb->nlh->nlmsg_seq,
1997 NLM_F_MULTI,
1998 OVS_VPORT_CMD_NEW) < 0)
1999 goto out;
2000
2001 j++;
2002 }
2003 skip = 0;
2004 }
2005 out:
2006 rcu_read_unlock();
2007
2008 cb->args[0] = i;
2009 cb->args[1] = j;
2010
2011 return skb->len;
2012 }
2013
2014 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2015 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2016 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2017 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2018 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2019 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2020 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2021 };
2022
2023 static const struct genl_ops dp_vport_genl_ops[] = {
2024 { .cmd = OVS_VPORT_CMD_NEW,
2025 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2026 .policy = vport_policy,
2027 .doit = ovs_vport_cmd_new
2028 },
2029 { .cmd = OVS_VPORT_CMD_DEL,
2030 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2031 .policy = vport_policy,
2032 .doit = ovs_vport_cmd_del
2033 },
2034 { .cmd = OVS_VPORT_CMD_GET,
2035 .flags = 0, /* OK for unprivileged users. */
2036 .policy = vport_policy,
2037 .doit = ovs_vport_cmd_get,
2038 .dumpit = ovs_vport_cmd_dump
2039 },
2040 { .cmd = OVS_VPORT_CMD_SET,
2041 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2042 .policy = vport_policy,
2043 .doit = ovs_vport_cmd_set,
2044 },
2045 };
2046
2047 struct genl_family dp_vport_genl_family = {
2048 .id = GENL_ID_GENERATE,
2049 .hdrsize = sizeof(struct ovs_header),
2050 .name = OVS_VPORT_FAMILY,
2051 .version = OVS_VPORT_VERSION,
2052 .maxattr = OVS_VPORT_ATTR_MAX,
2053 .netnsok = true,
2054 .parallel_ops = true,
2055 .ops = dp_vport_genl_ops,
2056 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2057 .mcgrps = &ovs_dp_vport_multicast_group,
2058 .n_mcgrps = 1,
2059 };
2060
2061 static struct genl_family * const dp_genl_families[] = {
2062 &dp_datapath_genl_family,
2063 &dp_vport_genl_family,
2064 &dp_flow_genl_family,
2065 &dp_packet_genl_family,
2066 };
2067
2068 static void dp_unregister_genl(int n_families)
2069 {
2070 int i;
2071
2072 for (i = 0; i < n_families; i++)
2073 genl_unregister_family(dp_genl_families[i]);
2074 }
2075
2076 static int dp_register_genl(void)
2077 {
2078 int err;
2079 int i;
2080
2081 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2082
2083 err = genl_register_family(dp_genl_families[i]);
2084 if (err)
2085 goto error;
2086 }
2087
2088 return 0;
2089
2090 error:
2091 dp_unregister_genl(i);
2092 return err;
2093 }
2094
2095 static int __net_init ovs_init_net(struct net *net)
2096 {
2097 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2098
2099 INIT_LIST_HEAD(&ovs_net->dps);
2100 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2101 return 0;
2102 }
2103
2104 static void __net_exit ovs_exit_net(struct net *net)
2105 {
2106 struct datapath *dp, *dp_next;
2107 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2108
2109 ovs_lock();
2110 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2111 __dp_destroy(dp);
2112 ovs_unlock();
2113
2114 cancel_work_sync(&ovs_net->dp_notify_work);
2115 }
2116
2117 static struct pernet_operations ovs_net_ops = {
2118 .init = ovs_init_net,
2119 .exit = ovs_exit_net,
2120 .id = &ovs_net_id,
2121 .size = sizeof(struct ovs_net),
2122 };
2123
2124 static int __init dp_init(void)
2125 {
2126 int err;
2127
2128 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2129
2130 pr_info("Open vSwitch switching datapath\n");
2131
2132 err = action_fifos_init();
2133 if (err)
2134 goto error;
2135
2136 err = ovs_internal_dev_rtnl_link_register();
2137 if (err)
2138 goto error_action_fifos_exit;
2139
2140 err = ovs_flow_init();
2141 if (err)
2142 goto error_unreg_rtnl_link;
2143
2144 err = ovs_vport_init();
2145 if (err)
2146 goto error_flow_exit;
2147
2148 err = register_pernet_device(&ovs_net_ops);
2149 if (err)
2150 goto error_vport_exit;
2151
2152 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2153 if (err)
2154 goto error_netns_exit;
2155
2156 err = ovs_netdev_init();
2157 if (err)
2158 goto error_unreg_notifier;
2159
2160 err = dp_register_genl();
2161 if (err < 0)
2162 goto error_unreg_netdev;
2163
2164 return 0;
2165
2166 error_unreg_netdev:
2167 ovs_netdev_exit();
2168 error_unreg_notifier:
2169 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2170 error_netns_exit:
2171 unregister_pernet_device(&ovs_net_ops);
2172 error_vport_exit:
2173 ovs_vport_exit();
2174 error_flow_exit:
2175 ovs_flow_exit();
2176 error_unreg_rtnl_link:
2177 ovs_internal_dev_rtnl_link_unregister();
2178 error_action_fifos_exit:
2179 action_fifos_exit();
2180 error:
2181 return err;
2182 }
2183
2184 static void dp_cleanup(void)
2185 {
2186 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2187 ovs_netdev_exit();
2188 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2189 unregister_pernet_device(&ovs_net_ops);
2190 rcu_barrier();
2191 ovs_vport_exit();
2192 ovs_flow_exit();
2193 ovs_internal_dev_rtnl_link_unregister();
2194 action_fifos_exit();
2195 }
2196
2197 module_init(dp_init);
2198 module_exit(dp_cleanup);
2199
2200 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2201 MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.102986 seconds and 5 git commands to generate.