]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame_incremental - net/openvswitch/datapath.c
projects / mirror_ubuntu-bionic-kernel.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
openvswitch: refactor do_output() to move NULL check out of fast path
[mirror_ubuntu-bionic-kernel.git] / net / openvswitch / datapath.c
... / ...
CommitLineData
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
61int ovs_net_id __read_mostly;
62EXPORT_SYMBOL(ovs_net_id);
63
64static struct genl_family dp_packet_genl_family;
65static struct genl_family dp_flow_genl_family;
66static struct genl_family dp_datapath_genl_family;
67
68static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
69 .name = OVS_FLOW_MCGROUP,
70};
71
72static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
73 .name = OVS_DATAPATH_MCGROUP,
74};
75
76static 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. */
82static 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
90static 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
114static DEFINE_MUTEX(ovs_mutex);
115
116void ovs_lock(void)
117{
118 mutex_lock(&ovs_mutex);
119}
120
121void ovs_unlock(void)
122{
123 mutex_unlock(&ovs_mutex);
124}
125
126#ifdef CONFIG_LOCKDEP
127int lockdep_ovsl_is_held(void)
128{
129 if (debug_locks)
130 return lockdep_is_held(&ovs_mutex);
131 else
132 return 1;
133}
134EXPORT_SYMBOL(lockdep_ovsl_is_held);
135#endif
136
137static struct vport *new_vport(const struct vport_parms *);
138static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 const struct dp_upcall_info *);
140static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 const struct dp_upcall_info *);
142
143/* Must be called with rcu_read_lock or ovs_mutex. */
144static struct datapath *get_dp(struct net *net, int dp_ifindex)
145{
146 struct datapath *dp = NULL;
147 struct net_device *dev;
148
149 rcu_read_lock();
150 dev = dev_get_by_index_rcu(net, dp_ifindex);
151 if (dev) {
152 struct vport *vport = ovs_internal_dev_get_vport(dev);
153 if (vport)
154 dp = vport->dp;
155 }
156 rcu_read_unlock();
157
158 return dp;
159}
160
161/* Must be called with rcu_read_lock or ovs_mutex. */
162const char *ovs_dp_name(const struct datapath *dp)
163{
164 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
165 return vport->ops->get_name(vport);
166}
167
168static int get_dpifindex(struct datapath *dp)
169{
170 struct vport *local;
171 int ifindex;
172
173 rcu_read_lock();
174
175 local = ovs_vport_rcu(dp, OVSP_LOCAL);
176 if (local)
177 ifindex = netdev_vport_priv(local)->dev->ifindex;
178 else
179 ifindex = 0;
180
181 rcu_read_unlock();
182
183 return ifindex;
184}
185
186static void destroy_dp_rcu(struct rcu_head *rcu)
187{
188 struct datapath *dp = container_of(rcu, struct datapath, rcu);
189
190 ovs_flow_tbl_destroy(&dp->table);
191 free_percpu(dp->stats_percpu);
192 release_net(ovs_dp_get_net(dp));
193 kfree(dp->ports);
194 kfree(dp);
195}
196
197static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
198 u16 port_no)
199{
200 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
201}
202
203/* Called with ovs_mutex or RCU read lock. */
204struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
205{
206 struct vport *vport;
207 struct hlist_head *head;
208
209 head = vport_hash_bucket(dp, port_no);
210 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
211 if (vport->port_no == port_no)
212 return vport;
213 }
214 return NULL;
215}
216
217/* Called with ovs_mutex. */
218static struct vport *new_vport(const struct vport_parms *parms)
219{
220 struct vport *vport;
221
222 vport = ovs_vport_add(parms);
223 if (!IS_ERR(vport)) {
224 struct datapath *dp = parms->dp;
225 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
226
227 hlist_add_head_rcu(&vport->dp_hash_node, head);
228 }
229 return vport;
230}
231
232void ovs_dp_detach_port(struct vport *p)
233{
234 ASSERT_OVSL();
235
236 /* First drop references to device. */
237 hlist_del_rcu(&p->dp_hash_node);
238
239 /* Then destroy it. */
240 ovs_vport_del(p);
241}
242
243/* Must be called with rcu_read_lock. */
244void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
245{
246 const struct vport *p = OVS_CB(skb)->input_vport;
247 struct datapath *dp = p->dp;
248 struct sw_flow *flow;
249 struct dp_stats_percpu *stats;
250 u64 *stats_counter;
251 u32 n_mask_hit;
252
253 stats = this_cpu_ptr(dp->stats_percpu);
254
255 /* Look up flow. */
256 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
257 if (unlikely(!flow)) {
258 struct dp_upcall_info upcall;
259 int error;
260
261 upcall.cmd = OVS_PACKET_CMD_MISS;
262 upcall.key = key;
263 upcall.userdata = NULL;
264 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
265 error = ovs_dp_upcall(dp, skb, &upcall);
266 if (unlikely(error))
267 kfree_skb(skb);
268 else
269 consume_skb(skb);
270 stats_counter = &stats->n_missed;
271 goto out;
272 }
273
274 OVS_CB(skb)->flow = flow;
275
276 ovs_flow_stats_update(OVS_CB(skb)->flow, key->tp.flags, skb);
277 ovs_execute_actions(dp, skb, key);
278 stats_counter = &stats->n_hit;
279
280out:
281 /* Update datapath statistics. */
282 u64_stats_update_begin(&stats->syncp);
283 (*stats_counter)++;
284 stats->n_mask_hit += n_mask_hit;
285 u64_stats_update_end(&stats->syncp);
286}
287
288int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
289 const struct dp_upcall_info *upcall_info)
290{
291 struct dp_stats_percpu *stats;
292 int err;
293
294 if (upcall_info->portid == 0) {
295 err = -ENOTCONN;
296 goto err;
297 }
298
299 if (!skb_is_gso(skb))
300 err = queue_userspace_packet(dp, skb, upcall_info);
301 else
302 err = queue_gso_packets(dp, skb, upcall_info);
303 if (err)
304 goto err;
305
306 return 0;
307
308err:
309 stats = this_cpu_ptr(dp->stats_percpu);
310
311 u64_stats_update_begin(&stats->syncp);
312 stats->n_lost++;
313 u64_stats_update_end(&stats->syncp);
314
315 return err;
316}
317
318static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
319 const struct dp_upcall_info *upcall_info)
320{
321 unsigned short gso_type = skb_shinfo(skb)->gso_type;
322 struct dp_upcall_info later_info;
323 struct sw_flow_key later_key;
324 struct sk_buff *segs, *nskb;
325 int err;
326
327 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
328 if (IS_ERR(segs))
329 return PTR_ERR(segs);
330 if (segs == NULL)
331 return -EINVAL;
332
333 /* Queue all of the segments. */
334 skb = segs;
335 do {
336 err = queue_userspace_packet(dp, skb, upcall_info);
337 if (err)
338 break;
339
340 if (skb == segs && gso_type & SKB_GSO_UDP) {
341 /* The initial flow key extracted by ovs_flow_extract()
342 * in this case is for a first fragment, so we need to
343 * properly mark later fragments.
344 */
345 later_key = *upcall_info->key;
346 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
347
348 later_info = *upcall_info;
349 later_info.key = &later_key;
350 upcall_info = &later_info;
351 }
352 } while ((skb = skb->next));
353
354 /* Free all of the segments. */
355 skb = segs;
356 do {
357 nskb = skb->next;
358 if (err)
359 kfree_skb(skb);
360 else
361 consume_skb(skb);
362 } while ((skb = nskb));
363 return err;
364}
365
366static size_t key_attr_size(void)
367{
368 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
369 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
370 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
371 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
372 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
373 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
374 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
375 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
376 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
377 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_OAM */
378 + nla_total_size(256) /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
379 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
380 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
381 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
382 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
383 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
384 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
385 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
386 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
387 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
388 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
389}
390
391static size_t upcall_msg_size(const struct nlattr *userdata,
392 unsigned int hdrlen)
393{
394 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
395 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
396 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
397
398 /* OVS_PACKET_ATTR_USERDATA */
399 if (userdata)
400 size += NLA_ALIGN(userdata->nla_len);
401
402 return size;
403}
404
405static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
406 const struct dp_upcall_info *upcall_info)
407{
408 struct ovs_header *upcall;
409 struct sk_buff *nskb = NULL;
410 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
411 struct nlattr *nla;
412 struct genl_info info = {
413 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
414 .snd_portid = upcall_info->portid,
415 };
416 size_t len;
417 unsigned int hlen;
418 int err, dp_ifindex;
419
420 dp_ifindex = get_dpifindex(dp);
421 if (!dp_ifindex)
422 return -ENODEV;
423
424 if (vlan_tx_tag_present(skb)) {
425 nskb = skb_clone(skb, GFP_ATOMIC);
426 if (!nskb)
427 return -ENOMEM;
428
429 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
430 if (!nskb)
431 return -ENOMEM;
432
433 nskb->vlan_tci = 0;
434 skb = nskb;
435 }
436
437 if (nla_attr_size(skb->len) > USHRT_MAX) {
438 err = -EFBIG;
439 goto out;
440 }
441
442 /* Complete checksum if needed */
443 if (skb->ip_summed == CHECKSUM_PARTIAL &&
444 (err = skb_checksum_help(skb)))
445 goto out;
446
447 /* Older versions of OVS user space enforce alignment of the last
448 * Netlink attribute to NLA_ALIGNTO which would require extensive
449 * padding logic. Only perform zerocopy if padding is not required.
450 */
451 if (dp->user_features & OVS_DP_F_UNALIGNED)
452 hlen = skb_zerocopy_headlen(skb);
453 else
454 hlen = skb->len;
455
456 len = upcall_msg_size(upcall_info->userdata, hlen);
457 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
458 if (!user_skb) {
459 err = -ENOMEM;
460 goto out;
461 }
462
463 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
464 0, upcall_info->cmd);
465 upcall->dp_ifindex = dp_ifindex;
466
467 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
468 err = ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
469 BUG_ON(err);
470 nla_nest_end(user_skb, nla);
471
472 if (upcall_info->userdata)
473 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
474 nla_len(upcall_info->userdata),
475 nla_data(upcall_info->userdata));
476
477 /* Only reserve room for attribute header, packet data is added
478 * in skb_zerocopy() */
479 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
480 err = -ENOBUFS;
481 goto out;
482 }
483 nla->nla_len = nla_attr_size(skb->len);
484
485 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
486 if (err)
487 goto out;
488
489 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
490 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
491 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
492
493 if (plen > 0)
494 memset(skb_put(user_skb, plen), 0, plen);
495 }
496
497 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
498
499 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
500 user_skb = NULL;
501out:
502 if (err)
503 skb_tx_error(skb);
504 kfree_skb(user_skb);
505 kfree_skb(nskb);
506 return err;
507}
508
509static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
510{
511 struct ovs_header *ovs_header = info->userhdr;
512 struct nlattr **a = info->attrs;
513 struct sw_flow_actions *acts;
514 struct sk_buff *packet;
515 struct sw_flow *flow;
516 struct datapath *dp;
517 struct ethhdr *eth;
518 struct vport *input_vport;
519 int len;
520 int err;
521
522 err = -EINVAL;
523 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
524 !a[OVS_PACKET_ATTR_ACTIONS])
525 goto err;
526
527 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
528 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
529 err = -ENOMEM;
530 if (!packet)
531 goto err;
532 skb_reserve(packet, NET_IP_ALIGN);
533
534 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
535
536 skb_reset_mac_header(packet);
537 eth = eth_hdr(packet);
538
539 /* Normally, setting the skb 'protocol' field would be handled by a
540 * call to eth_type_trans(), but it assumes there's a sending
541 * device, which we may not have. */
542 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
543 packet->protocol = eth->h_proto;
544 else
545 packet->protocol = htons(ETH_P_802_2);
546
547 /* Build an sw_flow for sending this packet. */
548 flow = ovs_flow_alloc();
549 err = PTR_ERR(flow);
550 if (IS_ERR(flow))
551 goto err_kfree_skb;
552
553 err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
554 &flow->key);
555 if (err)
556 goto err_flow_free;
557
558 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
559 err = PTR_ERR(acts);
560 if (IS_ERR(acts))
561 goto err_flow_free;
562
563 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
564 &flow->key, &acts);
565 if (err)
566 goto err_flow_free;
567
568 rcu_assign_pointer(flow->sf_acts, acts);
569
570 OVS_CB(packet)->egress_tun_info = NULL;
571 OVS_CB(packet)->flow = flow;
572 packet->priority = flow->key.phy.priority;
573 packet->mark = flow->key.phy.skb_mark;
574
575 rcu_read_lock();
576 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
577 err = -ENODEV;
578 if (!dp)
579 goto err_unlock;
580
581 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
582 if (!input_vport)
583 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
584
585 if (!input_vport)
586 goto err_unlock;
587
588 OVS_CB(packet)->input_vport = input_vport;
589
590 local_bh_disable();
591 err = ovs_execute_actions(dp, packet, &flow->key);
592 local_bh_enable();
593 rcu_read_unlock();
594
595 ovs_flow_free(flow, false);
596 return err;
597
598err_unlock:
599 rcu_read_unlock();
600err_flow_free:
601 ovs_flow_free(flow, false);
602err_kfree_skb:
603 kfree_skb(packet);
604err:
605 return err;
606}
607
608static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
609 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
610 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
611 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
612};
613
614static const struct genl_ops dp_packet_genl_ops[] = {
615 { .cmd = OVS_PACKET_CMD_EXECUTE,
616 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
617 .policy = packet_policy,
618 .doit = ovs_packet_cmd_execute
619 }
620};
621
622static struct genl_family dp_packet_genl_family = {
623 .id = GENL_ID_GENERATE,
624 .hdrsize = sizeof(struct ovs_header),
625 .name = OVS_PACKET_FAMILY,
626 .version = OVS_PACKET_VERSION,
627 .maxattr = OVS_PACKET_ATTR_MAX,
628 .netnsok = true,
629 .parallel_ops = true,
630 .ops = dp_packet_genl_ops,
631 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
632};
633
634static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
635 struct ovs_dp_megaflow_stats *mega_stats)
636{
637 int i;
638
639 memset(mega_stats, 0, sizeof(*mega_stats));
640
641 stats->n_flows = ovs_flow_tbl_count(&dp->table);
642 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
643
644 stats->n_hit = stats->n_missed = stats->n_lost = 0;
645
646 for_each_possible_cpu(i) {
647 const struct dp_stats_percpu *percpu_stats;
648 struct dp_stats_percpu local_stats;
649 unsigned int start;
650
651 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
652
653 do {
654 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
655 local_stats = *percpu_stats;
656 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
657
658 stats->n_hit += local_stats.n_hit;
659 stats->n_missed += local_stats.n_missed;
660 stats->n_lost += local_stats.n_lost;
661 mega_stats->n_mask_hit += local_stats.n_mask_hit;
662 }
663}
664
665static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
666{
667 return NLMSG_ALIGN(sizeof(struct ovs_header))
668 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
669 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
670 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
671 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
672 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
673 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
674}
675
676/* Called with ovs_mutex or RCU read lock. */
677static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
678 struct sk_buff *skb, u32 portid,
679 u32 seq, u32 flags, u8 cmd)
680{
681 const int skb_orig_len = skb->len;
682 struct nlattr *start;
683 struct ovs_flow_stats stats;
684 __be16 tcp_flags;
685 unsigned long used;
686 struct ovs_header *ovs_header;
687 struct nlattr *nla;
688 int err;
689
690 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
691 if (!ovs_header)
692 return -EMSGSIZE;
693
694 ovs_header->dp_ifindex = dp_ifindex;
695
696 /* Fill flow key. */
697 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
698 if (!nla)
699 goto nla_put_failure;
700
701 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
702 if (err)
703 goto error;
704 nla_nest_end(skb, nla);
705
706 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
707 if (!nla)
708 goto nla_put_failure;
709
710 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
711 if (err)
712 goto error;
713
714 nla_nest_end(skb, nla);
715
716 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
717
718 if (used &&
719 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
720 goto nla_put_failure;
721
722 if (stats.n_packets &&
723 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
724 goto nla_put_failure;
725
726 if ((u8)ntohs(tcp_flags) &&
727 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
728 goto nla_put_failure;
729
730 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
731 * this is the first flow to be dumped into 'skb'. This is unusual for
732 * Netlink but individual action lists can be longer than
733 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
734 * The userspace caller can always fetch the actions separately if it
735 * really wants them. (Most userspace callers in fact don't care.)
736 *
737 * This can only fail for dump operations because the skb is always
738 * properly sized for single flows.
739 */
740 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
741 if (start) {
742 const struct sw_flow_actions *sf_acts;
743
744 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
745 err = ovs_nla_put_actions(sf_acts->actions,
746 sf_acts->actions_len, skb);
747
748 if (!err)
749 nla_nest_end(skb, start);
750 else {
751 if (skb_orig_len)
752 goto error;
753
754 nla_nest_cancel(skb, start);
755 }
756 } else if (skb_orig_len)
757 goto nla_put_failure;
758
759 return genlmsg_end(skb, ovs_header);
760
761nla_put_failure:
762 err = -EMSGSIZE;
763error:
764 genlmsg_cancel(skb, ovs_header);
765 return err;
766}
767
768/* May not be called with RCU read lock. */
769static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
770 struct genl_info *info,
771 bool always)
772{
773 struct sk_buff *skb;
774
775 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
776 return NULL;
777
778 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
779 if (!skb)
780 return ERR_PTR(-ENOMEM);
781
782 return skb;
783}
784
785/* Called with ovs_mutex. */
786static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
787 int dp_ifindex,
788 struct genl_info *info, u8 cmd,
789 bool always)
790{
791 struct sk_buff *skb;
792 int retval;
793
794 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
795 always);
796 if (IS_ERR_OR_NULL(skb))
797 return skb;
798
799 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
800 info->snd_portid, info->snd_seq, 0,
801 cmd);
802 BUG_ON(retval < 0);
803 return skb;
804}
805
806static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
807{
808 struct nlattr **a = info->attrs;
809 struct ovs_header *ovs_header = info->userhdr;
810 struct sw_flow *flow, *new_flow;
811 struct sw_flow_mask mask;
812 struct sk_buff *reply;
813 struct datapath *dp;
814 struct sw_flow_actions *acts;
815 struct sw_flow_match match;
816 int error;
817
818 /* Must have key and actions. */
819 error = -EINVAL;
820 if (!a[OVS_FLOW_ATTR_KEY]) {
821 OVS_NLERR("Flow key attribute not present in new flow.\n");
822 goto error;
823 }
824 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
825 OVS_NLERR("Flow actions attribute not present in new flow.\n");
826 goto error;
827 }
828
829 /* Most of the time we need to allocate a new flow, do it before
830 * locking.
831 */
832 new_flow = ovs_flow_alloc();
833 if (IS_ERR(new_flow)) {
834 error = PTR_ERR(new_flow);
835 goto error;
836 }
837
838 /* Extract key. */
839 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
840 error = ovs_nla_get_match(&match,
841 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
842 if (error)
843 goto err_kfree_flow;
844
845 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
846
847 /* Validate actions. */
848 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
849 error = PTR_ERR(acts);
850 if (IS_ERR(acts))
851 goto err_kfree_flow;
852
853 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
854 &acts);
855 if (error) {
856 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
857 goto err_kfree_acts;
858 }
859
860 reply = ovs_flow_cmd_alloc_info(acts, info, false);
861 if (IS_ERR(reply)) {
862 error = PTR_ERR(reply);
863 goto err_kfree_acts;
864 }
865
866 ovs_lock();
867 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
868 if (unlikely(!dp)) {
869 error = -ENODEV;
870 goto err_unlock_ovs;
871 }
872 /* Check if this is a duplicate flow */
873 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
874 if (likely(!flow)) {
875 rcu_assign_pointer(new_flow->sf_acts, acts);
876
877 /* Put flow in bucket. */
878 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
879 if (unlikely(error)) {
880 acts = NULL;
881 goto err_unlock_ovs;
882 }
883
884 if (unlikely(reply)) {
885 error = ovs_flow_cmd_fill_info(new_flow,
886 ovs_header->dp_ifindex,
887 reply, info->snd_portid,
888 info->snd_seq, 0,
889 OVS_FLOW_CMD_NEW);
890 BUG_ON(error < 0);
891 }
892 ovs_unlock();
893 } else {
894 struct sw_flow_actions *old_acts;
895
896 /* Bail out if we're not allowed to modify an existing flow.
897 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
898 * because Generic Netlink treats the latter as a dump
899 * request. We also accept NLM_F_EXCL in case that bug ever
900 * gets fixed.
901 */
902 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
903 | NLM_F_EXCL))) {
904 error = -EEXIST;
905 goto err_unlock_ovs;
906 }
907 /* The unmasked key has to be the same for flow updates. */
908 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
909 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
910 if (!flow) {
911 error = -ENOENT;
912 goto err_unlock_ovs;
913 }
914 }
915 /* Update actions. */
916 old_acts = ovsl_dereference(flow->sf_acts);
917 rcu_assign_pointer(flow->sf_acts, acts);
918
919 if (unlikely(reply)) {
920 error = ovs_flow_cmd_fill_info(flow,
921 ovs_header->dp_ifindex,
922 reply, info->snd_portid,
923 info->snd_seq, 0,
924 OVS_FLOW_CMD_NEW);
925 BUG_ON(error < 0);
926 }
927 ovs_unlock();
928
929 ovs_nla_free_flow_actions(old_acts);
930 ovs_flow_free(new_flow, false);
931 }
932
933 if (reply)
934 ovs_notify(&dp_flow_genl_family, reply, info);
935 return 0;
936
937err_unlock_ovs:
938 ovs_unlock();
939 kfree_skb(reply);
940err_kfree_acts:
941 kfree(acts);
942err_kfree_flow:
943 ovs_flow_free(new_flow, false);
944error:
945 return error;
946}
947
948static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
949 const struct sw_flow_key *key,
950 const struct sw_flow_mask *mask)
951{
952 struct sw_flow_actions *acts;
953 struct sw_flow_key masked_key;
954 int error;
955
956 acts = ovs_nla_alloc_flow_actions(nla_len(a));
957 if (IS_ERR(acts))
958 return acts;
959
960 ovs_flow_mask_key(&masked_key, key, mask);
961 error = ovs_nla_copy_actions(a, &masked_key, &acts);
962 if (error) {
963 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
964 kfree(acts);
965 return ERR_PTR(error);
966 }
967
968 return acts;
969}
970
971static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
972{
973 struct nlattr **a = info->attrs;
974 struct ovs_header *ovs_header = info->userhdr;
975 struct sw_flow_key key;
976 struct sw_flow *flow;
977 struct sw_flow_mask mask;
978 struct sk_buff *reply = NULL;
979 struct datapath *dp;
980 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
981 struct sw_flow_match match;
982 int error;
983
984 /* Extract key. */
985 error = -EINVAL;
986 if (!a[OVS_FLOW_ATTR_KEY]) {
987 OVS_NLERR("Flow key attribute not present in set flow.\n");
988 goto error;
989 }
990
991 ovs_match_init(&match, &key, &mask);
992 error = ovs_nla_get_match(&match,
993 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
994 if (error)
995 goto error;
996
997 /* Validate actions. */
998 if (a[OVS_FLOW_ATTR_ACTIONS]) {
999 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask);
1000 if (IS_ERR(acts)) {
1001 error = PTR_ERR(acts);
1002 goto error;
1003 }
1004 }
1005
1006 /* Can allocate before locking if have acts. */
1007 if (acts) {
1008 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1009 if (IS_ERR(reply)) {
1010 error = PTR_ERR(reply);
1011 goto err_kfree_acts;
1012 }
1013 }
1014
1015 ovs_lock();
1016 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1017 if (unlikely(!dp)) {
1018 error = -ENODEV;
1019 goto err_unlock_ovs;
1020 }
1021 /* Check that the flow exists. */
1022 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1023 if (unlikely(!flow)) {
1024 error = -ENOENT;
1025 goto err_unlock_ovs;
1026 }
1027
1028 /* Update actions, if present. */
1029 if (likely(acts)) {
1030 old_acts = ovsl_dereference(flow->sf_acts);
1031 rcu_assign_pointer(flow->sf_acts, acts);
1032
1033 if (unlikely(reply)) {
1034 error = ovs_flow_cmd_fill_info(flow,
1035 ovs_header->dp_ifindex,
1036 reply, info->snd_portid,
1037 info->snd_seq, 0,
1038 OVS_FLOW_CMD_NEW);
1039 BUG_ON(error < 0);
1040 }
1041 } else {
1042 /* Could not alloc without acts before locking. */
1043 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1044 info, OVS_FLOW_CMD_NEW, false);
1045 if (unlikely(IS_ERR(reply))) {
1046 error = PTR_ERR(reply);
1047 goto err_unlock_ovs;
1048 }
1049 }
1050
1051 /* Clear stats. */
1052 if (a[OVS_FLOW_ATTR_CLEAR])
1053 ovs_flow_stats_clear(flow);
1054 ovs_unlock();
1055
1056 if (reply)
1057 ovs_notify(&dp_flow_genl_family, reply, info);
1058 if (old_acts)
1059 ovs_nla_free_flow_actions(old_acts);
1060
1061 return 0;
1062
1063err_unlock_ovs:
1064 ovs_unlock();
1065 kfree_skb(reply);
1066err_kfree_acts:
1067 kfree(acts);
1068error:
1069 return error;
1070}
1071
1072static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1073{
1074 struct nlattr **a = info->attrs;
1075 struct ovs_header *ovs_header = info->userhdr;
1076 struct sw_flow_key key;
1077 struct sk_buff *reply;
1078 struct sw_flow *flow;
1079 struct datapath *dp;
1080 struct sw_flow_match match;
1081 int err;
1082
1083 if (!a[OVS_FLOW_ATTR_KEY]) {
1084 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1085 return -EINVAL;
1086 }
1087
1088 ovs_match_init(&match, &key, NULL);
1089 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1090 if (err)
1091 return err;
1092
1093 ovs_lock();
1094 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1095 if (!dp) {
1096 err = -ENODEV;
1097 goto unlock;
1098 }
1099
1100 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1101 if (!flow) {
1102 err = -ENOENT;
1103 goto unlock;
1104 }
1105
1106 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1107 OVS_FLOW_CMD_NEW, true);
1108 if (IS_ERR(reply)) {
1109 err = PTR_ERR(reply);
1110 goto unlock;
1111 }
1112
1113 ovs_unlock();
1114 return genlmsg_reply(reply, info);
1115unlock:
1116 ovs_unlock();
1117 return err;
1118}
1119
1120static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1121{
1122 struct nlattr **a = info->attrs;
1123 struct ovs_header *ovs_header = info->userhdr;
1124 struct sw_flow_key key;
1125 struct sk_buff *reply;
1126 struct sw_flow *flow;
1127 struct datapath *dp;
1128 struct sw_flow_match match;
1129 int err;
1130
1131 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1132 ovs_match_init(&match, &key, NULL);
1133 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1134 if (unlikely(err))
1135 return err;
1136 }
1137
1138 ovs_lock();
1139 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1140 if (unlikely(!dp)) {
1141 err = -ENODEV;
1142 goto unlock;
1143 }
1144
1145 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1146 err = ovs_flow_tbl_flush(&dp->table);
1147 goto unlock;
1148 }
1149
1150 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1151 if (unlikely(!flow)) {
1152 err = -ENOENT;
1153 goto unlock;
1154 }
1155
1156 ovs_flow_tbl_remove(&dp->table, flow);
1157 ovs_unlock();
1158
1159 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1160 info, false);
1161 if (likely(reply)) {
1162 if (likely(!IS_ERR(reply))) {
1163 rcu_read_lock(); /*To keep RCU checker happy. */
1164 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1165 reply, info->snd_portid,
1166 info->snd_seq, 0,
1167 OVS_FLOW_CMD_DEL);
1168 rcu_read_unlock();
1169 BUG_ON(err < 0);
1170
1171 ovs_notify(&dp_flow_genl_family, reply, info);
1172 } else {
1173 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1174 }
1175 }
1176
1177 ovs_flow_free(flow, true);
1178 return 0;
1179unlock:
1180 ovs_unlock();
1181 return err;
1182}
1183
1184static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1185{
1186 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1187 struct table_instance *ti;
1188 struct datapath *dp;
1189
1190 rcu_read_lock();
1191 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1192 if (!dp) {
1193 rcu_read_unlock();
1194 return -ENODEV;
1195 }
1196
1197 ti = rcu_dereference(dp->table.ti);
1198 for (;;) {
1199 struct sw_flow *flow;
1200 u32 bucket, obj;
1201
1202 bucket = cb->args[0];
1203 obj = cb->args[1];
1204 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1205 if (!flow)
1206 break;
1207
1208 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1209 NETLINK_CB(cb->skb).portid,
1210 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1211 OVS_FLOW_CMD_NEW) < 0)
1212 break;
1213
1214 cb->args[0] = bucket;
1215 cb->args[1] = obj;
1216 }
1217 rcu_read_unlock();
1218 return skb->len;
1219}
1220
1221static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1222 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1223 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1224 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1225};
1226
1227static const struct genl_ops dp_flow_genl_ops[] = {
1228 { .cmd = OVS_FLOW_CMD_NEW,
1229 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1230 .policy = flow_policy,
1231 .doit = ovs_flow_cmd_new
1232 },
1233 { .cmd = OVS_FLOW_CMD_DEL,
1234 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1235 .policy = flow_policy,
1236 .doit = ovs_flow_cmd_del
1237 },
1238 { .cmd = OVS_FLOW_CMD_GET,
1239 .flags = 0, /* OK for unprivileged users. */
1240 .policy = flow_policy,
1241 .doit = ovs_flow_cmd_get,
1242 .dumpit = ovs_flow_cmd_dump
1243 },
1244 { .cmd = OVS_FLOW_CMD_SET,
1245 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1246 .policy = flow_policy,
1247 .doit = ovs_flow_cmd_set,
1248 },
1249};
1250
1251static struct genl_family dp_flow_genl_family = {
1252 .id = GENL_ID_GENERATE,
1253 .hdrsize = sizeof(struct ovs_header),
1254 .name = OVS_FLOW_FAMILY,
1255 .version = OVS_FLOW_VERSION,
1256 .maxattr = OVS_FLOW_ATTR_MAX,
1257 .netnsok = true,
1258 .parallel_ops = true,
1259 .ops = dp_flow_genl_ops,
1260 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1261 .mcgrps = &ovs_dp_flow_multicast_group,
1262 .n_mcgrps = 1,
1263};
1264
1265static size_t ovs_dp_cmd_msg_size(void)
1266{
1267 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1268
1269 msgsize += nla_total_size(IFNAMSIZ);
1270 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1271 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1272 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1273
1274 return msgsize;
1275}
1276
1277/* Called with ovs_mutex or RCU read lock. */
1278static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1279 u32 portid, u32 seq, u32 flags, u8 cmd)
1280{
1281 struct ovs_header *ovs_header;
1282 struct ovs_dp_stats dp_stats;
1283 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1284 int err;
1285
1286 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1287 flags, cmd);
1288 if (!ovs_header)
1289 goto error;
1290
1291 ovs_header->dp_ifindex = get_dpifindex(dp);
1292
1293 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1294 if (err)
1295 goto nla_put_failure;
1296
1297 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1298 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1299 &dp_stats))
1300 goto nla_put_failure;
1301
1302 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1303 sizeof(struct ovs_dp_megaflow_stats),
1304 &dp_megaflow_stats))
1305 goto nla_put_failure;
1306
1307 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1308 goto nla_put_failure;
1309
1310 return genlmsg_end(skb, ovs_header);
1311
1312nla_put_failure:
1313 genlmsg_cancel(skb, ovs_header);
1314error:
1315 return -EMSGSIZE;
1316}
1317
1318static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1319{
1320 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1321}
1322
1323/* Called with rcu_read_lock or ovs_mutex. */
1324static struct datapath *lookup_datapath(struct net *net,
1325 struct ovs_header *ovs_header,
1326 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1327{
1328 struct datapath *dp;
1329
1330 if (!a[OVS_DP_ATTR_NAME])
1331 dp = get_dp(net, ovs_header->dp_ifindex);
1332 else {
1333 struct vport *vport;
1334
1335 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1336 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1337 }
1338 return dp ? dp : ERR_PTR(-ENODEV);
1339}
1340
1341static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1342{
1343 struct datapath *dp;
1344
1345 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1346 if (IS_ERR(dp))
1347 return;
1348
1349 WARN(dp->user_features, "Dropping previously announced user features\n");
1350 dp->user_features = 0;
1351}
1352
1353static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1354{
1355 if (a[OVS_DP_ATTR_USER_FEATURES])
1356 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1357}
1358
1359static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1360{
1361 struct nlattr **a = info->attrs;
1362 struct vport_parms parms;
1363 struct sk_buff *reply;
1364 struct datapath *dp;
1365 struct vport *vport;
1366 struct ovs_net *ovs_net;
1367 int err, i;
1368
1369 err = -EINVAL;
1370 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1371 goto err;
1372
1373 reply = ovs_dp_cmd_alloc_info(info);
1374 if (!reply)
1375 return -ENOMEM;
1376
1377 err = -ENOMEM;
1378 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1379 if (dp == NULL)
1380 goto err_free_reply;
1381
1382 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1383
1384 /* Allocate table. */
1385 err = ovs_flow_tbl_init(&dp->table);
1386 if (err)
1387 goto err_free_dp;
1388
1389 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1390 if (!dp->stats_percpu) {
1391 err = -ENOMEM;
1392 goto err_destroy_table;
1393 }
1394
1395 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1396 GFP_KERNEL);
1397 if (!dp->ports) {
1398 err = -ENOMEM;
1399 goto err_destroy_percpu;
1400 }
1401
1402 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1403 INIT_HLIST_HEAD(&dp->ports[i]);
1404
1405 /* Set up our datapath device. */
1406 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1407 parms.type = OVS_VPORT_TYPE_INTERNAL;
1408 parms.options = NULL;
1409 parms.dp = dp;
1410 parms.port_no = OVSP_LOCAL;
1411 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1412
1413 ovs_dp_change(dp, a);
1414
1415 /* So far only local changes have been made, now need the lock. */
1416 ovs_lock();
1417
1418 vport = new_vport(&parms);
1419 if (IS_ERR(vport)) {
1420 err = PTR_ERR(vport);
1421 if (err == -EBUSY)
1422 err = -EEXIST;
1423
1424 if (err == -EEXIST) {
1425 /* An outdated user space instance that does not understand
1426 * the concept of user_features has attempted to create a new
1427 * datapath and is likely to reuse it. Drop all user features.
1428 */
1429 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1430 ovs_dp_reset_user_features(skb, info);
1431 }
1432
1433 goto err_destroy_ports_array;
1434 }
1435
1436 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1437 info->snd_seq, 0, OVS_DP_CMD_NEW);
1438 BUG_ON(err < 0);
1439
1440 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1441 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1442
1443 ovs_unlock();
1444
1445 ovs_notify(&dp_datapath_genl_family, reply, info);
1446 return 0;
1447
1448err_destroy_ports_array:
1449 ovs_unlock();
1450 kfree(dp->ports);
1451err_destroy_percpu:
1452 free_percpu(dp->stats_percpu);
1453err_destroy_table:
1454 ovs_flow_tbl_destroy(&dp->table);
1455err_free_dp:
1456 release_net(ovs_dp_get_net(dp));
1457 kfree(dp);
1458err_free_reply:
1459 kfree_skb(reply);
1460err:
1461 return err;
1462}
1463
1464/* Called with ovs_mutex. */
1465static void __dp_destroy(struct datapath *dp)
1466{
1467 int i;
1468
1469 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1470 struct vport *vport;
1471 struct hlist_node *n;
1472
1473 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1474 if (vport->port_no != OVSP_LOCAL)
1475 ovs_dp_detach_port(vport);
1476 }
1477
1478 list_del_rcu(&dp->list_node);
1479
1480 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1481 * all ports in datapath are destroyed first before freeing datapath.
1482 */
1483 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1484
1485 /* RCU destroy the flow table */
1486 call_rcu(&dp->rcu, destroy_dp_rcu);
1487}
1488
1489static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1490{
1491 struct sk_buff *reply;
1492 struct datapath *dp;
1493 int err;
1494
1495 reply = ovs_dp_cmd_alloc_info(info);
1496 if (!reply)
1497 return -ENOMEM;
1498
1499 ovs_lock();
1500 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1501 err = PTR_ERR(dp);
1502 if (IS_ERR(dp))
1503 goto err_unlock_free;
1504
1505 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1506 info->snd_seq, 0, OVS_DP_CMD_DEL);
1507 BUG_ON(err < 0);
1508
1509 __dp_destroy(dp);
1510 ovs_unlock();
1511
1512 ovs_notify(&dp_datapath_genl_family, reply, info);
1513
1514 return 0;
1515
1516err_unlock_free:
1517 ovs_unlock();
1518 kfree_skb(reply);
1519 return err;
1520}
1521
1522static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1523{
1524 struct sk_buff *reply;
1525 struct datapath *dp;
1526 int err;
1527
1528 reply = ovs_dp_cmd_alloc_info(info);
1529 if (!reply)
1530 return -ENOMEM;
1531
1532 ovs_lock();
1533 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1534 err = PTR_ERR(dp);
1535 if (IS_ERR(dp))
1536 goto err_unlock_free;
1537
1538 ovs_dp_change(dp, info->attrs);
1539
1540 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1541 info->snd_seq, 0, OVS_DP_CMD_NEW);
1542 BUG_ON(err < 0);
1543
1544 ovs_unlock();
1545 ovs_notify(&dp_datapath_genl_family, reply, info);
1546
1547 return 0;
1548
1549err_unlock_free:
1550 ovs_unlock();
1551 kfree_skb(reply);
1552 return err;
1553}
1554
1555static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1556{
1557 struct sk_buff *reply;
1558 struct datapath *dp;
1559 int err;
1560
1561 reply = ovs_dp_cmd_alloc_info(info);
1562 if (!reply)
1563 return -ENOMEM;
1564
1565 rcu_read_lock();
1566 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1567 if (IS_ERR(dp)) {
1568 err = PTR_ERR(dp);
1569 goto err_unlock_free;
1570 }
1571 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1572 info->snd_seq, 0, OVS_DP_CMD_NEW);
1573 BUG_ON(err < 0);
1574 rcu_read_unlock();
1575
1576 return genlmsg_reply(reply, info);
1577
1578err_unlock_free:
1579 rcu_read_unlock();
1580 kfree_skb(reply);
1581 return err;
1582}
1583
1584static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1585{
1586 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1587 struct datapath *dp;
1588 int skip = cb->args[0];
1589 int i = 0;
1590
1591 rcu_read_lock();
1592 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1593 if (i >= skip &&
1594 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1595 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1596 OVS_DP_CMD_NEW) < 0)
1597 break;
1598 i++;
1599 }
1600 rcu_read_unlock();
1601
1602 cb->args[0] = i;
1603
1604 return skb->len;
1605}
1606
1607static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1608 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1609 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1610 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1611};
1612
1613static const struct genl_ops dp_datapath_genl_ops[] = {
1614 { .cmd = OVS_DP_CMD_NEW,
1615 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1616 .policy = datapath_policy,
1617 .doit = ovs_dp_cmd_new
1618 },
1619 { .cmd = OVS_DP_CMD_DEL,
1620 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1621 .policy = datapath_policy,
1622 .doit = ovs_dp_cmd_del
1623 },
1624 { .cmd = OVS_DP_CMD_GET,
1625 .flags = 0, /* OK for unprivileged users. */
1626 .policy = datapath_policy,
1627 .doit = ovs_dp_cmd_get,
1628 .dumpit = ovs_dp_cmd_dump
1629 },
1630 { .cmd = OVS_DP_CMD_SET,
1631 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1632 .policy = datapath_policy,
1633 .doit = ovs_dp_cmd_set,
1634 },
1635};
1636
1637static struct genl_family dp_datapath_genl_family = {
1638 .id = GENL_ID_GENERATE,
1639 .hdrsize = sizeof(struct ovs_header),
1640 .name = OVS_DATAPATH_FAMILY,
1641 .version = OVS_DATAPATH_VERSION,
1642 .maxattr = OVS_DP_ATTR_MAX,
1643 .netnsok = true,
1644 .parallel_ops = true,
1645 .ops = dp_datapath_genl_ops,
1646 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1647 .mcgrps = &ovs_dp_datapath_multicast_group,
1648 .n_mcgrps = 1,
1649};
1650
1651/* Called with ovs_mutex or RCU read lock. */
1652static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1653 u32 portid, u32 seq, u32 flags, u8 cmd)
1654{
1655 struct ovs_header *ovs_header;
1656 struct ovs_vport_stats vport_stats;
1657 int err;
1658
1659 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1660 flags, cmd);
1661 if (!ovs_header)
1662 return -EMSGSIZE;
1663
1664 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1665
1666 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1667 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1668 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1669 vport->ops->get_name(vport)))
1670 goto nla_put_failure;
1671
1672 ovs_vport_get_stats(vport, &vport_stats);
1673 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1674 &vport_stats))
1675 goto nla_put_failure;
1676
1677 if (ovs_vport_get_upcall_portids(vport, skb))
1678 goto nla_put_failure;
1679
1680 err = ovs_vport_get_options(vport, skb);
1681 if (err == -EMSGSIZE)
1682 goto error;
1683
1684 return genlmsg_end(skb, ovs_header);
1685
1686nla_put_failure:
1687 err = -EMSGSIZE;
1688error:
1689 genlmsg_cancel(skb, ovs_header);
1690 return err;
1691}
1692
1693static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1694{
1695 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1696}
1697
1698/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1699struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1700 u32 seq, u8 cmd)
1701{
1702 struct sk_buff *skb;
1703 int retval;
1704
1705 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1706 if (!skb)
1707 return ERR_PTR(-ENOMEM);
1708
1709 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1710 BUG_ON(retval < 0);
1711
1712 return skb;
1713}
1714
1715/* Called with ovs_mutex or RCU read lock. */
1716static struct vport *lookup_vport(struct net *net,
1717 struct ovs_header *ovs_header,
1718 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1719{
1720 struct datapath *dp;
1721 struct vport *vport;
1722
1723 if (a[OVS_VPORT_ATTR_NAME]) {
1724 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1725 if (!vport)
1726 return ERR_PTR(-ENODEV);
1727 if (ovs_header->dp_ifindex &&
1728 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1729 return ERR_PTR(-ENODEV);
1730 return vport;
1731 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1732 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1733
1734 if (port_no >= DP_MAX_PORTS)
1735 return ERR_PTR(-EFBIG);
1736
1737 dp = get_dp(net, ovs_header->dp_ifindex);
1738 if (!dp)
1739 return ERR_PTR(-ENODEV);
1740
1741 vport = ovs_vport_ovsl_rcu(dp, port_no);
1742 if (!vport)
1743 return ERR_PTR(-ENODEV);
1744 return vport;
1745 } else
1746 return ERR_PTR(-EINVAL);
1747}
1748
1749static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1750{
1751 struct nlattr **a = info->attrs;
1752 struct ovs_header *ovs_header = info->userhdr;
1753 struct vport_parms parms;
1754 struct sk_buff *reply;
1755 struct vport *vport;
1756 struct datapath *dp;
1757 u32 port_no;
1758 int err;
1759
1760 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1761 !a[OVS_VPORT_ATTR_UPCALL_PID])
1762 return -EINVAL;
1763
1764 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1765 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1766 if (port_no >= DP_MAX_PORTS)
1767 return -EFBIG;
1768
1769 reply = ovs_vport_cmd_alloc_info();
1770 if (!reply)
1771 return -ENOMEM;
1772
1773 ovs_lock();
1774restart:
1775 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1776 err = -ENODEV;
1777 if (!dp)
1778 goto exit_unlock_free;
1779
1780 if (port_no) {
1781 vport = ovs_vport_ovsl(dp, port_no);
1782 err = -EBUSY;
1783 if (vport)
1784 goto exit_unlock_free;
1785 } else {
1786 for (port_no = 1; ; port_no++) {
1787 if (port_no >= DP_MAX_PORTS) {
1788 err = -EFBIG;
1789 goto exit_unlock_free;
1790 }
1791 vport = ovs_vport_ovsl(dp, port_no);
1792 if (!vport)
1793 break;
1794 }
1795 }
1796
1797 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1798 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1799 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1800 parms.dp = dp;
1801 parms.port_no = port_no;
1802 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1803
1804 vport = new_vport(&parms);
1805 err = PTR_ERR(vport);
1806 if (IS_ERR(vport)) {
1807 if (err == -EAGAIN)
1808 goto restart;
1809 goto exit_unlock_free;
1810 }
1811
1812 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1813 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1814 BUG_ON(err < 0);
1815 ovs_unlock();
1816
1817 ovs_notify(&dp_vport_genl_family, reply, info);
1818 return 0;
1819
1820exit_unlock_free:
1821 ovs_unlock();
1822 kfree_skb(reply);
1823 return err;
1824}
1825
1826static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1827{
1828 struct nlattr **a = info->attrs;
1829 struct sk_buff *reply;
1830 struct vport *vport;
1831 int err;
1832
1833 reply = ovs_vport_cmd_alloc_info();
1834 if (!reply)
1835 return -ENOMEM;
1836
1837 ovs_lock();
1838 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1839 err = PTR_ERR(vport);
1840 if (IS_ERR(vport))
1841 goto exit_unlock_free;
1842
1843 if (a[OVS_VPORT_ATTR_TYPE] &&
1844 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1845 err = -EINVAL;
1846 goto exit_unlock_free;
1847 }
1848
1849 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1850 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1851 if (err)
1852 goto exit_unlock_free;
1853 }
1854
1855
1856 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1857 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1858
1859 err = ovs_vport_set_upcall_portids(vport, ids);
1860 if (err)
1861 goto exit_unlock_free;
1862 }
1863
1864 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1865 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1866 BUG_ON(err < 0);
1867
1868 ovs_unlock();
1869 ovs_notify(&dp_vport_genl_family, reply, info);
1870 return 0;
1871
1872exit_unlock_free:
1873 ovs_unlock();
1874 kfree_skb(reply);
1875 return err;
1876}
1877
1878static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1879{
1880 struct nlattr **a = info->attrs;
1881 struct sk_buff *reply;
1882 struct vport *vport;
1883 int err;
1884
1885 reply = ovs_vport_cmd_alloc_info();
1886 if (!reply)
1887 return -ENOMEM;
1888
1889 ovs_lock();
1890 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1891 err = PTR_ERR(vport);
1892 if (IS_ERR(vport))
1893 goto exit_unlock_free;
1894
1895 if (vport->port_no == OVSP_LOCAL) {
1896 err = -EINVAL;
1897 goto exit_unlock_free;
1898 }
1899
1900 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1901 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1902 BUG_ON(err < 0);
1903 ovs_dp_detach_port(vport);
1904 ovs_unlock();
1905
1906 ovs_notify(&dp_vport_genl_family, reply, info);
1907 return 0;
1908
1909exit_unlock_free:
1910 ovs_unlock();
1911 kfree_skb(reply);
1912 return err;
1913}
1914
1915static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1916{
1917 struct nlattr **a = info->attrs;
1918 struct ovs_header *ovs_header = info->userhdr;
1919 struct sk_buff *reply;
1920 struct vport *vport;
1921 int err;
1922
1923 reply = ovs_vport_cmd_alloc_info();
1924 if (!reply)
1925 return -ENOMEM;
1926
1927 rcu_read_lock();
1928 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1929 err = PTR_ERR(vport);
1930 if (IS_ERR(vport))
1931 goto exit_unlock_free;
1932 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1933 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1934 BUG_ON(err < 0);
1935 rcu_read_unlock();
1936
1937 return genlmsg_reply(reply, info);
1938
1939exit_unlock_free:
1940 rcu_read_unlock();
1941 kfree_skb(reply);
1942 return err;
1943}
1944
1945static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1946{
1947 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1948 struct datapath *dp;
1949 int bucket = cb->args[0], skip = cb->args[1];
1950 int i, j = 0;
1951
1952 rcu_read_lock();
1953 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1954 if (!dp) {
1955 rcu_read_unlock();
1956 return -ENODEV;
1957 }
1958 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1959 struct vport *vport;
1960
1961 j = 0;
1962 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1963 if (j >= skip &&
1964 ovs_vport_cmd_fill_info(vport, skb,
1965 NETLINK_CB(cb->skb).portid,
1966 cb->nlh->nlmsg_seq,
1967 NLM_F_MULTI,
1968 OVS_VPORT_CMD_NEW) < 0)
1969 goto out;
1970
1971 j++;
1972 }
1973 skip = 0;
1974 }
1975out:
1976 rcu_read_unlock();
1977
1978 cb->args[0] = i;
1979 cb->args[1] = j;
1980
1981 return skb->len;
1982}
1983
1984static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1985 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1986 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1987 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1988 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1989 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1990 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1991};
1992
1993static const struct genl_ops dp_vport_genl_ops[] = {
1994 { .cmd = OVS_VPORT_CMD_NEW,
1995 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1996 .policy = vport_policy,
1997 .doit = ovs_vport_cmd_new
1998 },
1999 { .cmd = OVS_VPORT_CMD_DEL,
2000 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2001 .policy = vport_policy,
2002 .doit = ovs_vport_cmd_del
2003 },
2004 { .cmd = OVS_VPORT_CMD_GET,
2005 .flags = 0, /* OK for unprivileged users. */
2006 .policy = vport_policy,
2007 .doit = ovs_vport_cmd_get,
2008 .dumpit = ovs_vport_cmd_dump
2009 },
2010 { .cmd = OVS_VPORT_CMD_SET,
2011 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2012 .policy = vport_policy,
2013 .doit = ovs_vport_cmd_set,
2014 },
2015};
2016
2017struct genl_family dp_vport_genl_family = {
2018 .id = GENL_ID_GENERATE,
2019 .hdrsize = sizeof(struct ovs_header),
2020 .name = OVS_VPORT_FAMILY,
2021 .version = OVS_VPORT_VERSION,
2022 .maxattr = OVS_VPORT_ATTR_MAX,
2023 .netnsok = true,
2024 .parallel_ops = true,
2025 .ops = dp_vport_genl_ops,
2026 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2027 .mcgrps = &ovs_dp_vport_multicast_group,
2028 .n_mcgrps = 1,
2029};
2030
2031static struct genl_family * const dp_genl_families[] = {
2032 &dp_datapath_genl_family,
2033 &dp_vport_genl_family,
2034 &dp_flow_genl_family,
2035 &dp_packet_genl_family,
2036};
2037
2038static void dp_unregister_genl(int n_families)
2039{
2040 int i;
2041
2042 for (i = 0; i < n_families; i++)
2043 genl_unregister_family(dp_genl_families[i]);
2044}
2045
2046static int dp_register_genl(void)
2047{
2048 int err;
2049 int i;
2050
2051 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2052
2053 err = genl_register_family(dp_genl_families[i]);
2054 if (err)
2055 goto error;
2056 }
2057
2058 return 0;
2059
2060error:
2061 dp_unregister_genl(i);
2062 return err;
2063}
2064
2065static int __net_init ovs_init_net(struct net *net)
2066{
2067 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2068
2069 INIT_LIST_HEAD(&ovs_net->dps);
2070 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2071 return 0;
2072}
2073
2074static void __net_exit ovs_exit_net(struct net *net)
2075{
2076 struct datapath *dp, *dp_next;
2077 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2078
2079 ovs_lock();
2080 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2081 __dp_destroy(dp);
2082 ovs_unlock();
2083
2084 cancel_work_sync(&ovs_net->dp_notify_work);
2085}
2086
2087static struct pernet_operations ovs_net_ops = {
2088 .init = ovs_init_net,
2089 .exit = ovs_exit_net,
2090 .id = &ovs_net_id,
2091 .size = sizeof(struct ovs_net),
2092};
2093
2094static int __init dp_init(void)
2095{
2096 int err;
2097
2098 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2099
2100 pr_info("Open vSwitch switching datapath\n");
2101
2102 err = action_fifos_init();
2103 if (err)
2104 goto error;
2105
2106 err = ovs_internal_dev_rtnl_link_register();
2107 if (err)
2108 goto error_action_fifos_exit;
2109
2110 err = ovs_flow_init();
2111 if (err)
2112 goto error_unreg_rtnl_link;
2113
2114 err = ovs_vport_init();
2115 if (err)
2116 goto error_flow_exit;
2117
2118 err = register_pernet_device(&ovs_net_ops);
2119 if (err)
2120 goto error_vport_exit;
2121
2122 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2123 if (err)
2124 goto error_netns_exit;
2125
2126 err = ovs_netdev_init();
2127 if (err)
2128 goto error_unreg_notifier;
2129
2130 err = dp_register_genl();
2131 if (err < 0)
2132 goto error_unreg_netdev;
2133
2134 return 0;
2135
2136error_unreg_netdev:
2137 ovs_netdev_exit();
2138error_unreg_notifier:
2139 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2140error_netns_exit:
2141 unregister_pernet_device(&ovs_net_ops);
2142error_vport_exit:
2143 ovs_vport_exit();
2144error_flow_exit:
2145 ovs_flow_exit();
2146error_unreg_rtnl_link:
2147 ovs_internal_dev_rtnl_link_unregister();
2148error_action_fifos_exit:
2149 action_fifos_exit();
2150error:
2151 return err;
2152}
2153
2154static void dp_cleanup(void)
2155{
2156 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2157 ovs_netdev_exit();
2158 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2159 unregister_pernet_device(&ovs_net_ops);
2160 rcu_barrier();
2161 ovs_vport_exit();
2162 ovs_flow_exit();
2163 ovs_internal_dev_rtnl_link_unregister();
2164 action_fifos_exit();
2165}
2166
2167module_init(dp_init);
2168module_exit(dp_cleanup);
2169
2170MODULE_DESCRIPTION("Open vSwitch switching datapath");
2171MODULE_LICENSE("GPL");
ubuntu-bionic-kernel mirror