Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2007-2014 Nicira, Inc.
4 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/init.h>
9#include <linux/module.h>
10#include <linux/if_arp.h>
11#include <linux/if_vlan.h>
12#include <linux/in.h>
13#include <linux/ip.h>
14#include <linux/jhash.h>
15#include <linux/delay.h>
16#include <linux/time.h>
17#include <linux/etherdevice.h>
18#include <linux/genetlink.h>
19#include <linux/kernel.h>
20#include <linux/kthread.h>
21#include <linux/mutex.h>
22#include <linux/percpu.h>
23#include <linux/rcupdate.h>
24#include <linux/tcp.h>
25#include <linux/udp.h>
26#include <linux/ethtool.h>
27#include <linux/wait.h>
28#include <asm/div64.h>
29#include <linux/highmem.h>
30#include <linux/netfilter_bridge.h>
31#include <linux/netfilter_ipv4.h>
32#include <linux/inetdevice.h>
33#include <linux/list.h>
34#include <linux/openvswitch.h>
35#include <linux/rculist.h>
36#include <linux/dmi.h>
37#include <net/genetlink.h>
38#include <net/net_namespace.h>
39#include <net/netns/generic.h>
40
41#include "datapath.h"
42#include "flow.h"
43#include "flow_table.h"
44#include "flow_netlink.h"
45#include "meter.h"
46#include "vport-internal_dev.h"
47#include "vport-netdev.h"
48
49unsigned int ovs_net_id __read_mostly;
50
51static struct genl_family dp_packet_genl_family;
52static struct genl_family dp_flow_genl_family;
53static struct genl_family dp_datapath_genl_family;
54
55static const struct nla_policy flow_policy[];
56
57static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
58 .name = OVS_FLOW_MCGROUP,
59};
60
61static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
62 .name = OVS_DATAPATH_MCGROUP,
63};
64
65static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
66 .name = OVS_VPORT_MCGROUP,
67};
68
69/* Check if need to build a reply message.
70 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
71static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
72 unsigned int group)
73{
74 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
75 genl_has_listeners(family, genl_info_net(info), group);
76}
77
78static void ovs_notify(struct genl_family *family,
79 struct sk_buff *skb, struct genl_info *info)
80{
81 genl_notify(family, skb, info, 0, GFP_KERNEL);
82}
83
84/**
85 * DOC: Locking:
86 *
87 * All writes e.g. Writes to device state (add/remove datapath, port, set
88 * operations on vports, etc.), Writes to other state (flow table
89 * modifications, set miscellaneous datapath parameters, etc.) are protected
90 * by ovs_lock.
91 *
92 * Reads are protected by RCU.
93 *
94 * There are a few special cases (mostly stats) that have their own
95 * synchronization but they nest under all of above and don't interact with
96 * each other.
97 *
98 * The RTNL lock nests inside ovs_mutex.
99 */
100
101static DEFINE_MUTEX(ovs_mutex);
102
103void ovs_lock(void)
104{
105 mutex_lock(&ovs_mutex);
106}
107
108void ovs_unlock(void)
109{
110 mutex_unlock(&ovs_mutex);
111}
112
113#ifdef CONFIG_LOCKDEP
114int lockdep_ovsl_is_held(void)
115{
116 if (debug_locks)
117 return lockdep_is_held(&ovs_mutex);
118 else
119 return 1;
120}
121#endif
122
123static struct vport *new_vport(const struct vport_parms *);
124static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
125 const struct sw_flow_key *,
126 const struct dp_upcall_info *,
127 uint32_t cutlen);
128static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
129 const struct sw_flow_key *,
130 const struct dp_upcall_info *,
131 uint32_t cutlen);
132
133/* Must be called with rcu_read_lock or ovs_mutex. */
134const char *ovs_dp_name(const struct datapath *dp)
135{
136 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
137 return ovs_vport_name(vport);
138}
139
140static int get_dpifindex(const struct datapath *dp)
141{
142 struct vport *local;
143 int ifindex;
144
145 rcu_read_lock();
146
147 local = ovs_vport_rcu(dp, OVSP_LOCAL);
148 if (local)
149 ifindex = local->dev->ifindex;
150 else
151 ifindex = 0;
152
153 rcu_read_unlock();
154
155 return ifindex;
156}
157
158static void destroy_dp_rcu(struct rcu_head *rcu)
159{
160 struct datapath *dp = container_of(rcu, struct datapath, rcu);
161
162 ovs_flow_tbl_destroy(&dp->table);
163 free_percpu(dp->stats_percpu);
164 kfree(dp->ports);
165 ovs_meters_exit(dp);
166 kfree(dp);
167}
168
169static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
170 u16 port_no)
171{
172 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
173}
174
175/* Called with ovs_mutex or RCU read lock. */
176struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
177{
178 struct vport *vport;
179 struct hlist_head *head;
180
181 head = vport_hash_bucket(dp, port_no);
182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
183 if (vport->port_no == port_no)
184 return vport;
185 }
186 return NULL;
187}
188
189/* Called with ovs_mutex. */
190static struct vport *new_vport(const struct vport_parms *parms)
191{
192 struct vport *vport;
193
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
198
199 hlist_add_head_rcu(&vport->dp_hash_node, head);
200 }
201 return vport;
202}
203
204void ovs_dp_detach_port(struct vport *p)
205{
206 ASSERT_OVSL();
207
208 /* First drop references to device. */
209 hlist_del_rcu(&p->dp_hash_node);
210
211 /* Then destroy it. */
212 ovs_vport_del(p);
213}
214
215/* Must be called with rcu_read_lock. */
216void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
217{
218 const struct vport *p = OVS_CB(skb)->input_vport;
219 struct datapath *dp = p->dp;
220 struct sw_flow *flow;
221 struct sw_flow_actions *sf_acts;
222 struct dp_stats_percpu *stats;
223 u64 *stats_counter;
224 u32 n_mask_hit;
225 int error;
226
227 stats = this_cpu_ptr(dp->stats_percpu);
228
229 /* Look up flow. */
230 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
231 if (unlikely(!flow)) {
232 struct dp_upcall_info upcall;
233
234 memset(&upcall, 0, sizeof(upcall));
235 upcall.cmd = OVS_PACKET_CMD_MISS;
236 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
237 upcall.mru = OVS_CB(skb)->mru;
238 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
239 if (unlikely(error))
240 kfree_skb(skb);
241 else
242 consume_skb(skb);
243 stats_counter = &stats->n_missed;
244 goto out;
245 }
246
247 ovs_flow_stats_update(flow, key->tp.flags, skb);
248 sf_acts = rcu_dereference(flow->sf_acts);
249 error = ovs_execute_actions(dp, skb, sf_acts, key);
250 if (unlikely(error))
251 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
252 ovs_dp_name(dp), error);
253
254 stats_counter = &stats->n_hit;
255
256out:
257 /* Update datapath statistics. */
258 u64_stats_update_begin(&stats->syncp);
259 (*stats_counter)++;
260 stats->n_mask_hit += n_mask_hit;
261 u64_stats_update_end(&stats->syncp);
262}
263
264int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
265 const struct sw_flow_key *key,
266 const struct dp_upcall_info *upcall_info,
267 uint32_t cutlen)
268{
269 struct dp_stats_percpu *stats;
270 int err;
271
272 if (upcall_info->portid == 0) {
273 err = -ENOTCONN;
274 goto err;
275 }
276
277 if (!skb_is_gso(skb))
278 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
279 else
280 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
281 if (err)
282 goto err;
283
284 return 0;
285
286err:
287 stats = this_cpu_ptr(dp->stats_percpu);
288
289 u64_stats_update_begin(&stats->syncp);
290 stats->n_lost++;
291 u64_stats_update_end(&stats->syncp);
292
293 return err;
294}
295
296static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
297 const struct sw_flow_key *key,
298 const struct dp_upcall_info *upcall_info,
299 uint32_t cutlen)
300{
301 unsigned int gso_type = skb_shinfo(skb)->gso_type;
302 struct sw_flow_key later_key;
303 struct sk_buff *segs, *nskb;
304 int err;
305
306 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
307 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
308 if (IS_ERR(segs))
309 return PTR_ERR(segs);
310 if (segs == NULL)
311 return -EINVAL;
312
313 if (gso_type & SKB_GSO_UDP) {
314 /* The initial flow key extracted by ovs_flow_key_extract()
315 * in this case is for a first fragment, so we need to
316 * properly mark later fragments.
317 */
318 later_key = *key;
319 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
320 }
321
322 /* Queue all of the segments. */
323 skb = segs;
324 do {
325 if (gso_type & SKB_GSO_UDP && skb != segs)
326 key = &later_key;
327
328 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
329 if (err)
330 break;
331
332 } while ((skb = skb->next));
333
334 /* Free all of the segments. */
335 skb = segs;
336 do {
337 nskb = skb->next;
338 if (err)
339 kfree_skb(skb);
340 else
341 consume_skb(skb);
342 } while ((skb = nskb));
343 return err;
344}
345
346static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
347 unsigned int hdrlen, int actions_attrlen)
348{
349 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
350 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
351 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
352 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
353
354 /* OVS_PACKET_ATTR_USERDATA */
355 if (upcall_info->userdata)
356 size += NLA_ALIGN(upcall_info->userdata->nla_len);
357
358 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
359 if (upcall_info->egress_tun_info)
360 size += nla_total_size(ovs_tun_key_attr_size());
361
362 /* OVS_PACKET_ATTR_ACTIONS */
363 if (upcall_info->actions_len)
364 size += nla_total_size(actions_attrlen);
365
366 /* OVS_PACKET_ATTR_MRU */
367 if (upcall_info->mru)
368 size += nla_total_size(sizeof(upcall_info->mru));
369
370 return size;
371}
372
373static void pad_packet(struct datapath *dp, struct sk_buff *skb)
374{
375 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
376 size_t plen = NLA_ALIGN(skb->len) - skb->len;
377
378 if (plen > 0)
379 skb_put_zero(skb, plen);
380 }
381}
382
383static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
384 const struct sw_flow_key *key,
385 const struct dp_upcall_info *upcall_info,
386 uint32_t cutlen)
387{
388 struct ovs_header *upcall;
389 struct sk_buff *nskb = NULL;
390 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
391 struct nlattr *nla;
392 size_t len;
393 unsigned int hlen;
394 int err, dp_ifindex;
395
396 dp_ifindex = get_dpifindex(dp);
397 if (!dp_ifindex)
398 return -ENODEV;
399
400 if (skb_vlan_tag_present(skb)) {
401 nskb = skb_clone(skb, GFP_ATOMIC);
402 if (!nskb)
403 return -ENOMEM;
404
405 nskb = __vlan_hwaccel_push_inside(nskb);
406 if (!nskb)
407 return -ENOMEM;
408
409 skb = nskb;
410 }
411
412 if (nla_attr_size(skb->len) > USHRT_MAX) {
413 err = -EFBIG;
414 goto out;
415 }
416
417 /* Complete checksum if needed */
418 if (skb->ip_summed == CHECKSUM_PARTIAL &&
419 (err = skb_csum_hwoffload_help(skb, 0)))
420 goto out;
421
422 /* Older versions of OVS user space enforce alignment of the last
423 * Netlink attribute to NLA_ALIGNTO which would require extensive
424 * padding logic. Only perform zerocopy if padding is not required.
425 */
426 if (dp->user_features & OVS_DP_F_UNALIGNED)
427 hlen = skb_zerocopy_headlen(skb);
428 else
429 hlen = skb->len;
430
431 len = upcall_msg_size(upcall_info, hlen - cutlen,
432 OVS_CB(skb)->acts_origlen);
433 user_skb = genlmsg_new(len, GFP_ATOMIC);
434 if (!user_skb) {
435 err = -ENOMEM;
436 goto out;
437 }
438
439 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
440 0, upcall_info->cmd);
441 if (!upcall) {
442 err = -EINVAL;
443 goto out;
444 }
445 upcall->dp_ifindex = dp_ifindex;
446
447 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
448 if (err)
449 goto out;
450
451 if (upcall_info->userdata)
452 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
453 nla_len(upcall_info->userdata),
454 nla_data(upcall_info->userdata));
455
456 if (upcall_info->egress_tun_info) {
457 nla = nla_nest_start_noflag(user_skb,
458 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
459 if (!nla) {
460 err = -EMSGSIZE;
461 goto out;
462 }
463 err = ovs_nla_put_tunnel_info(user_skb,
464 upcall_info->egress_tun_info);
465 if (err)
466 goto out;
467
468 nla_nest_end(user_skb, nla);
469 }
470
471 if (upcall_info->actions_len) {
472 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
473 if (!nla) {
474 err = -EMSGSIZE;
475 goto out;
476 }
477 err = ovs_nla_put_actions(upcall_info->actions,
478 upcall_info->actions_len,
479 user_skb);
480 if (!err)
481 nla_nest_end(user_skb, nla);
482 else
483 nla_nest_cancel(user_skb, nla);
484 }
485
486 /* Add OVS_PACKET_ATTR_MRU */
487 if (upcall_info->mru) {
488 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
489 upcall_info->mru)) {
490 err = -ENOBUFS;
491 goto out;
492 }
493 pad_packet(dp, user_skb);
494 }
495
496 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
497 if (cutlen > 0) {
498 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
499 skb->len)) {
500 err = -ENOBUFS;
501 goto out;
502 }
503 pad_packet(dp, user_skb);
504 }
505
506 /* Only reserve room for attribute header, packet data is added
507 * in skb_zerocopy() */
508 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
509 err = -ENOBUFS;
510 goto out;
511 }
512 nla->nla_len = nla_attr_size(skb->len - cutlen);
513
514 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
515 if (err)
516 goto out;
517
518 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
519 pad_packet(dp, user_skb);
520
521 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
522
523 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
524 user_skb = NULL;
525out:
526 if (err)
527 skb_tx_error(skb);
528 kfree_skb(user_skb);
529 kfree_skb(nskb);
530 return err;
531}
532
533static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
534{
535 struct ovs_header *ovs_header = info->userhdr;
536 struct net *net = sock_net(skb->sk);
537 struct nlattr **a = info->attrs;
538 struct sw_flow_actions *acts;
539 struct sk_buff *packet;
540 struct sw_flow *flow;
541 struct sw_flow_actions *sf_acts;
542 struct datapath *dp;
543 struct vport *input_vport;
544 u16 mru = 0;
545 int len;
546 int err;
547 bool log = !a[OVS_PACKET_ATTR_PROBE];
548
549 err = -EINVAL;
550 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
551 !a[OVS_PACKET_ATTR_ACTIONS])
552 goto err;
553
554 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
555 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
556 err = -ENOMEM;
557 if (!packet)
558 goto err;
559 skb_reserve(packet, NET_IP_ALIGN);
560
561 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
562
563 /* Set packet's mru */
564 if (a[OVS_PACKET_ATTR_MRU]) {
565 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
566 packet->ignore_df = 1;
567 }
568 OVS_CB(packet)->mru = mru;
569
570 /* Build an sw_flow for sending this packet. */
571 flow = ovs_flow_alloc();
572 err = PTR_ERR(flow);
573 if (IS_ERR(flow))
574 goto err_kfree_skb;
575
576 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
577 packet, &flow->key, log);
578 if (err)
579 goto err_flow_free;
580
581 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
582 &flow->key, &acts, log);
583 if (err)
584 goto err_flow_free;
585
586 rcu_assign_pointer(flow->sf_acts, acts);
587 packet->priority = flow->key.phy.priority;
588 packet->mark = flow->key.phy.skb_mark;
589
590 rcu_read_lock();
591 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
592 err = -ENODEV;
593 if (!dp)
594 goto err_unlock;
595
596 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
597 if (!input_vport)
598 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
599
600 if (!input_vport)
601 goto err_unlock;
602
603 packet->dev = input_vport->dev;
604 OVS_CB(packet)->input_vport = input_vport;
605 sf_acts = rcu_dereference(flow->sf_acts);
606
607 local_bh_disable();
608 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
609 local_bh_enable();
610 rcu_read_unlock();
611
612 ovs_flow_free(flow, false);
613 return err;
614
615err_unlock:
616 rcu_read_unlock();
617err_flow_free:
618 ovs_flow_free(flow, false);
619err_kfree_skb:
620 kfree_skb(packet);
621err:
622 return err;
623}
624
625static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
626 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
627 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
628 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
629 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
630 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
631};
632
633static const struct genl_ops dp_packet_genl_ops[] = {
634 { .cmd = OVS_PACKET_CMD_EXECUTE,
635 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
636 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
637 .doit = ovs_packet_cmd_execute
638 }
639};
640
641static struct genl_family dp_packet_genl_family __ro_after_init = {
642 .hdrsize = sizeof(struct ovs_header),
643 .name = OVS_PACKET_FAMILY,
644 .version = OVS_PACKET_VERSION,
645 .maxattr = OVS_PACKET_ATTR_MAX,
646 .policy = packet_policy,
647 .netnsok = true,
648 .parallel_ops = true,
649 .ops = dp_packet_genl_ops,
650 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
651 .module = THIS_MODULE,
652};
653
654static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
655 struct ovs_dp_megaflow_stats *mega_stats)
656{
657 int i;
658
659 memset(mega_stats, 0, sizeof(*mega_stats));
660
661 stats->n_flows = ovs_flow_tbl_count(&dp->table);
662 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
663
664 stats->n_hit = stats->n_missed = stats->n_lost = 0;
665
666 for_each_possible_cpu(i) {
667 const struct dp_stats_percpu *percpu_stats;
668 struct dp_stats_percpu local_stats;
669 unsigned int start;
670
671 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
672
673 do {
674 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
675 local_stats = *percpu_stats;
676 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
677
678 stats->n_hit += local_stats.n_hit;
679 stats->n_missed += local_stats.n_missed;
680 stats->n_lost += local_stats.n_lost;
681 mega_stats->n_mask_hit += local_stats.n_mask_hit;
682 }
683}
684
685static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
686{
687 return ovs_identifier_is_ufid(sfid) &&
688 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
689}
690
691static bool should_fill_mask(uint32_t ufid_flags)
692{
693 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
694}
695
696static bool should_fill_actions(uint32_t ufid_flags)
697{
698 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
699}
700
701static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
702 const struct sw_flow_id *sfid,
703 uint32_t ufid_flags)
704{
705 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
706
707 /* OVS_FLOW_ATTR_UFID */
708 if (sfid && ovs_identifier_is_ufid(sfid))
709 len += nla_total_size(sfid->ufid_len);
710
711 /* OVS_FLOW_ATTR_KEY */
712 if (!sfid || should_fill_key(sfid, ufid_flags))
713 len += nla_total_size(ovs_key_attr_size());
714
715 /* OVS_FLOW_ATTR_MASK */
716 if (should_fill_mask(ufid_flags))
717 len += nla_total_size(ovs_key_attr_size());
718
719 /* OVS_FLOW_ATTR_ACTIONS */
720 if (should_fill_actions(ufid_flags))
721 len += nla_total_size(acts->orig_len);
722
723 return len
724 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
725 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
726 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
727}
728
729/* Called with ovs_mutex or RCU read lock. */
730static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
731 struct sk_buff *skb)
732{
733 struct ovs_flow_stats stats;
734 __be16 tcp_flags;
735 unsigned long used;
736
737 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
738
739 if (used &&
740 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
741 OVS_FLOW_ATTR_PAD))
742 return -EMSGSIZE;
743
744 if (stats.n_packets &&
745 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
746 sizeof(struct ovs_flow_stats), &stats,
747 OVS_FLOW_ATTR_PAD))
748 return -EMSGSIZE;
749
750 if ((u8)ntohs(tcp_flags) &&
751 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
752 return -EMSGSIZE;
753
754 return 0;
755}
756
757/* Called with ovs_mutex or RCU read lock. */
758static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
759 struct sk_buff *skb, int skb_orig_len)
760{
761 struct nlattr *start;
762 int err;
763
764 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
765 * this is the first flow to be dumped into 'skb'. This is unusual for
766 * Netlink but individual action lists can be longer than
767 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
768 * The userspace caller can always fetch the actions separately if it
769 * really wants them. (Most userspace callers in fact don't care.)
770 *
771 * This can only fail for dump operations because the skb is always
772 * properly sized for single flows.
773 */
774 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
775 if (start) {
776 const struct sw_flow_actions *sf_acts;
777
778 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
779 err = ovs_nla_put_actions(sf_acts->actions,
780 sf_acts->actions_len, skb);
781
782 if (!err)
783 nla_nest_end(skb, start);
784 else {
785 if (skb_orig_len)
786 return err;
787
788 nla_nest_cancel(skb, start);
789 }
790 } else if (skb_orig_len) {
791 return -EMSGSIZE;
792 }
793
794 return 0;
795}
796
797/* Called with ovs_mutex or RCU read lock. */
798static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
799 struct sk_buff *skb, u32 portid,
800 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
801{
802 const int skb_orig_len = skb->len;
803 struct ovs_header *ovs_header;
804 int err;
805
806 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
807 flags, cmd);
808 if (!ovs_header)
809 return -EMSGSIZE;
810
811 ovs_header->dp_ifindex = dp_ifindex;
812
813 err = ovs_nla_put_identifier(flow, skb);
814 if (err)
815 goto error;
816
817 if (should_fill_key(&flow->id, ufid_flags)) {
818 err = ovs_nla_put_masked_key(flow, skb);
819 if (err)
820 goto error;
821 }
822
823 if (should_fill_mask(ufid_flags)) {
824 err = ovs_nla_put_mask(flow, skb);
825 if (err)
826 goto error;
827 }
828
829 err = ovs_flow_cmd_fill_stats(flow, skb);
830 if (err)
831 goto error;
832
833 if (should_fill_actions(ufid_flags)) {
834 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
835 if (err)
836 goto error;
837 }
838
839 genlmsg_end(skb, ovs_header);
840 return 0;
841
842error:
843 genlmsg_cancel(skb, ovs_header);
844 return err;
845}
846
847/* May not be called with RCU read lock. */
848static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
849 const struct sw_flow_id *sfid,
850 struct genl_info *info,
851 bool always,
852 uint32_t ufid_flags)
853{
854 struct sk_buff *skb;
855 size_t len;
856
857 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
858 return NULL;
859
860 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
861 skb = genlmsg_new(len, GFP_KERNEL);
862 if (!skb)
863 return ERR_PTR(-ENOMEM);
864
865 return skb;
866}
867
868/* Called with ovs_mutex. */
869static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
870 int dp_ifindex,
871 struct genl_info *info, u8 cmd,
872 bool always, u32 ufid_flags)
873{
874 struct sk_buff *skb;
875 int retval;
876
877 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
878 &flow->id, info, always, ufid_flags);
879 if (IS_ERR_OR_NULL(skb))
880 return skb;
881
882 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
883 info->snd_portid, info->snd_seq, 0,
884 cmd, ufid_flags);
885 BUG_ON(retval < 0);
886 return skb;
887}
888
889static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
890{
891 struct net *net = sock_net(skb->sk);
892 struct nlattr **a = info->attrs;
893 struct ovs_header *ovs_header = info->userhdr;
894 struct sw_flow *flow = NULL, *new_flow;
895 struct sw_flow_mask mask;
896 struct sk_buff *reply;
897 struct datapath *dp;
898 struct sw_flow_actions *acts;
899 struct sw_flow_match match;
900 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
901 int error;
902 bool log = !a[OVS_FLOW_ATTR_PROBE];
903
904 /* Must have key and actions. */
905 error = -EINVAL;
906 if (!a[OVS_FLOW_ATTR_KEY]) {
907 OVS_NLERR(log, "Flow key attr not present in new flow.");
908 goto error;
909 }
910 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
911 OVS_NLERR(log, "Flow actions attr not present in new flow.");
912 goto error;
913 }
914
915 /* Most of the time we need to allocate a new flow, do it before
916 * locking.
917 */
918 new_flow = ovs_flow_alloc();
919 if (IS_ERR(new_flow)) {
920 error = PTR_ERR(new_flow);
921 goto error;
922 }
923
924 /* Extract key. */
925 ovs_match_init(&match, &new_flow->key, false, &mask);
926 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
927 a[OVS_FLOW_ATTR_MASK], log);
928 if (error)
929 goto err_kfree_flow;
930
931 /* Extract flow identifier. */
932 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
933 &new_flow->key, log);
934 if (error)
935 goto err_kfree_flow;
936
937 /* unmasked key is needed to match when ufid is not used. */
938 if (ovs_identifier_is_key(&new_flow->id))
939 match.key = new_flow->id.unmasked_key;
940
941 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
942
943 /* Validate actions. */
944 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
945 &new_flow->key, &acts, log);
946 if (error) {
947 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
948 goto err_kfree_flow;
949 }
950
951 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
952 ufid_flags);
953 if (IS_ERR(reply)) {
954 error = PTR_ERR(reply);
955 goto err_kfree_acts;
956 }
957
958 ovs_lock();
959 dp = get_dp(net, ovs_header->dp_ifindex);
960 if (unlikely(!dp)) {
961 error = -ENODEV;
962 goto err_unlock_ovs;
963 }
964
965 /* Check if this is a duplicate flow */
966 if (ovs_identifier_is_ufid(&new_flow->id))
967 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
968 if (!flow)
969 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
970 if (likely(!flow)) {
971 rcu_assign_pointer(new_flow->sf_acts, acts);
972
973 /* Put flow in bucket. */
974 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
975 if (unlikely(error)) {
976 acts = NULL;
977 goto err_unlock_ovs;
978 }
979
980 if (unlikely(reply)) {
981 error = ovs_flow_cmd_fill_info(new_flow,
982 ovs_header->dp_ifindex,
983 reply, info->snd_portid,
984 info->snd_seq, 0,
985 OVS_FLOW_CMD_NEW,
986 ufid_flags);
987 BUG_ON(error < 0);
988 }
989 ovs_unlock();
990 } else {
991 struct sw_flow_actions *old_acts;
992
993 /* Bail out if we're not allowed to modify an existing flow.
994 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
995 * because Generic Netlink treats the latter as a dump
996 * request. We also accept NLM_F_EXCL in case that bug ever
997 * gets fixed.
998 */
999 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1000 | NLM_F_EXCL))) {
1001 error = -EEXIST;
1002 goto err_unlock_ovs;
1003 }
1004 /* The flow identifier has to be the same for flow updates.
1005 * Look for any overlapping flow.
1006 */
1007 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1008 if (ovs_identifier_is_key(&flow->id))
1009 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1010 &match);
1011 else /* UFID matches but key is different */
1012 flow = NULL;
1013 if (!flow) {
1014 error = -ENOENT;
1015 goto err_unlock_ovs;
1016 }
1017 }
1018 /* Update actions. */
1019 old_acts = ovsl_dereference(flow->sf_acts);
1020 rcu_assign_pointer(flow->sf_acts, acts);
1021
1022 if (unlikely(reply)) {
1023 error = ovs_flow_cmd_fill_info(flow,
1024 ovs_header->dp_ifindex,
1025 reply, info->snd_portid,
1026 info->snd_seq, 0,
1027 OVS_FLOW_CMD_NEW,
1028 ufid_flags);
1029 BUG_ON(error < 0);
1030 }
1031 ovs_unlock();
1032
1033 ovs_nla_free_flow_actions_rcu(old_acts);
1034 ovs_flow_free(new_flow, false);
1035 }
1036
1037 if (reply)
1038 ovs_notify(&dp_flow_genl_family, reply, info);
1039 return 0;
1040
1041err_unlock_ovs:
1042 ovs_unlock();
1043 kfree_skb(reply);
1044err_kfree_acts:
1045 ovs_nla_free_flow_actions(acts);
1046err_kfree_flow:
1047 ovs_flow_free(new_flow, false);
1048error:
1049 return error;
1050}
1051
1052/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1053static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1054 const struct nlattr *a,
1055 const struct sw_flow_key *key,
1056 const struct sw_flow_mask *mask,
1057 bool log)
1058{
1059 struct sw_flow_actions *acts;
1060 struct sw_flow_key masked_key;
1061 int error;
1062
1063 ovs_flow_mask_key(&masked_key, key, true, mask);
1064 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1065 if (error) {
1066 OVS_NLERR(log,
1067 "Actions may not be safe on all matching packets");
1068 return ERR_PTR(error);
1069 }
1070
1071 return acts;
1072}
1073
1074/* Factor out match-init and action-copy to avoid
1075 * "Wframe-larger-than=1024" warning. Because mask is only
1076 * used to get actions, we new a function to save some
1077 * stack space.
1078 *
1079 * If there are not key and action attrs, we return 0
1080 * directly. In the case, the caller will also not use the
1081 * match as before. If there is action attr, we try to get
1082 * actions and save them to *acts. Before returning from
1083 * the function, we reset the match->mask pointer. Because
1084 * we should not to return match object with dangling reference
1085 * to mask.
1086 * */
1087static noinline_for_stack int
1088ovs_nla_init_match_and_action(struct net *net,
1089 struct sw_flow_match *match,
1090 struct sw_flow_key *key,
1091 struct nlattr **a,
1092 struct sw_flow_actions **acts,
1093 bool log)
1094{
1095 struct sw_flow_mask mask;
1096 int error = 0;
1097
1098 if (a[OVS_FLOW_ATTR_KEY]) {
1099 ovs_match_init(match, key, true, &mask);
1100 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1101 a[OVS_FLOW_ATTR_MASK], log);
1102 if (error)
1103 goto error;
1104 }
1105
1106 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1107 if (!a[OVS_FLOW_ATTR_KEY]) {
1108 OVS_NLERR(log,
1109 "Flow key attribute not present in set flow.");
1110 error = -EINVAL;
1111 goto error;
1112 }
1113
1114 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1115 &mask, log);
1116 if (IS_ERR(*acts)) {
1117 error = PTR_ERR(*acts);
1118 goto error;
1119 }
1120 }
1121
1122 /* On success, error is 0. */
1123error:
1124 match->mask = NULL;
1125 return error;
1126}
1127
1128static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1129{
1130 struct net *net = sock_net(skb->sk);
1131 struct nlattr **a = info->attrs;
1132 struct ovs_header *ovs_header = info->userhdr;
1133 struct sw_flow_key key;
1134 struct sw_flow *flow;
1135 struct sk_buff *reply = NULL;
1136 struct datapath *dp;
1137 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1138 struct sw_flow_match match;
1139 struct sw_flow_id sfid;
1140 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1141 int error = 0;
1142 bool log = !a[OVS_FLOW_ATTR_PROBE];
1143 bool ufid_present;
1144
1145 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1146 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1147 OVS_NLERR(log,
1148 "Flow set message rejected, Key attribute missing.");
1149 return -EINVAL;
1150 }
1151
1152 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1153 &acts, log);
1154 if (error)
1155 goto error;
1156
1157 if (acts) {
1158 /* Can allocate before locking if have acts. */
1159 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1160 ufid_flags);
1161 if (IS_ERR(reply)) {
1162 error = PTR_ERR(reply);
1163 goto err_kfree_acts;
1164 }
1165 }
1166
1167 ovs_lock();
1168 dp = get_dp(net, ovs_header->dp_ifindex);
1169 if (unlikely(!dp)) {
1170 error = -ENODEV;
1171 goto err_unlock_ovs;
1172 }
1173 /* Check that the flow exists. */
1174 if (ufid_present)
1175 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1176 else
1177 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1178 if (unlikely(!flow)) {
1179 error = -ENOENT;
1180 goto err_unlock_ovs;
1181 }
1182
1183 /* Update actions, if present. */
1184 if (likely(acts)) {
1185 old_acts = ovsl_dereference(flow->sf_acts);
1186 rcu_assign_pointer(flow->sf_acts, acts);
1187
1188 if (unlikely(reply)) {
1189 error = ovs_flow_cmd_fill_info(flow,
1190 ovs_header->dp_ifindex,
1191 reply, info->snd_portid,
1192 info->snd_seq, 0,
1193 OVS_FLOW_CMD_SET,
1194 ufid_flags);
1195 BUG_ON(error < 0);
1196 }
1197 } else {
1198 /* Could not alloc without acts before locking. */
1199 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1200 info, OVS_FLOW_CMD_SET, false,
1201 ufid_flags);
1202
1203 if (IS_ERR(reply)) {
1204 error = PTR_ERR(reply);
1205 goto err_unlock_ovs;
1206 }
1207 }
1208
1209 /* Clear stats. */
1210 if (a[OVS_FLOW_ATTR_CLEAR])
1211 ovs_flow_stats_clear(flow);
1212 ovs_unlock();
1213
1214 if (reply)
1215 ovs_notify(&dp_flow_genl_family, reply, info);
1216 if (old_acts)
1217 ovs_nla_free_flow_actions_rcu(old_acts);
1218
1219 return 0;
1220
1221err_unlock_ovs:
1222 ovs_unlock();
1223 kfree_skb(reply);
1224err_kfree_acts:
1225 ovs_nla_free_flow_actions(acts);
1226error:
1227 return error;
1228}
1229
1230static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1231{
1232 struct nlattr **a = info->attrs;
1233 struct ovs_header *ovs_header = info->userhdr;
1234 struct net *net = sock_net(skb->sk);
1235 struct sw_flow_key key;
1236 struct sk_buff *reply;
1237 struct sw_flow *flow;
1238 struct datapath *dp;
1239 struct sw_flow_match match;
1240 struct sw_flow_id ufid;
1241 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1242 int err = 0;
1243 bool log = !a[OVS_FLOW_ATTR_PROBE];
1244 bool ufid_present;
1245
1246 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1247 if (a[OVS_FLOW_ATTR_KEY]) {
1248 ovs_match_init(&match, &key, true, NULL);
1249 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1250 log);
1251 } else if (!ufid_present) {
1252 OVS_NLERR(log,
1253 "Flow get message rejected, Key attribute missing.");
1254 err = -EINVAL;
1255 }
1256 if (err)
1257 return err;
1258
1259 ovs_lock();
1260 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1261 if (!dp) {
1262 err = -ENODEV;
1263 goto unlock;
1264 }
1265
1266 if (ufid_present)
1267 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1268 else
1269 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1270 if (!flow) {
1271 err = -ENOENT;
1272 goto unlock;
1273 }
1274
1275 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1276 OVS_FLOW_CMD_GET, true, ufid_flags);
1277 if (IS_ERR(reply)) {
1278 err = PTR_ERR(reply);
1279 goto unlock;
1280 }
1281
1282 ovs_unlock();
1283 return genlmsg_reply(reply, info);
1284unlock:
1285 ovs_unlock();
1286 return err;
1287}
1288
1289static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1290{
1291 struct nlattr **a = info->attrs;
1292 struct ovs_header *ovs_header = info->userhdr;
1293 struct net *net = sock_net(skb->sk);
1294 struct sw_flow_key key;
1295 struct sk_buff *reply;
1296 struct sw_flow *flow = NULL;
1297 struct datapath *dp;
1298 struct sw_flow_match match;
1299 struct sw_flow_id ufid;
1300 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1301 int err;
1302 bool log = !a[OVS_FLOW_ATTR_PROBE];
1303 bool ufid_present;
1304
1305 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1306 if (a[OVS_FLOW_ATTR_KEY]) {
1307 ovs_match_init(&match, &key, true, NULL);
1308 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1309 NULL, log);
1310 if (unlikely(err))
1311 return err;
1312 }
1313
1314 ovs_lock();
1315 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1316 if (unlikely(!dp)) {
1317 err = -ENODEV;
1318 goto unlock;
1319 }
1320
1321 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1322 err = ovs_flow_tbl_flush(&dp->table);
1323 goto unlock;
1324 }
1325
1326 if (ufid_present)
1327 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1328 else
1329 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1330 if (unlikely(!flow)) {
1331 err = -ENOENT;
1332 goto unlock;
1333 }
1334
1335 ovs_flow_tbl_remove(&dp->table, flow);
1336 ovs_unlock();
1337
1338 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1339 &flow->id, info, false, ufid_flags);
1340 if (likely(reply)) {
1341 if (!IS_ERR(reply)) {
1342 rcu_read_lock(); /*To keep RCU checker happy. */
1343 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1344 reply, info->snd_portid,
1345 info->snd_seq, 0,
1346 OVS_FLOW_CMD_DEL,
1347 ufid_flags);
1348 rcu_read_unlock();
1349 BUG_ON(err < 0);
1350
1351 ovs_notify(&dp_flow_genl_family, reply, info);
1352 } else {
1353 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1354 }
1355 }
1356
1357 ovs_flow_free(flow, true);
1358 return 0;
1359unlock:
1360 ovs_unlock();
1361 return err;
1362}
1363
1364static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1365{
1366 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1367 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1368 struct table_instance *ti;
1369 struct datapath *dp;
1370 u32 ufid_flags;
1371 int err;
1372
1373 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1374 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1375 if (err)
1376 return err;
1377 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1378
1379 rcu_read_lock();
1380 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1381 if (!dp) {
1382 rcu_read_unlock();
1383 return -ENODEV;
1384 }
1385
1386 ti = rcu_dereference(dp->table.ti);
1387 for (;;) {
1388 struct sw_flow *flow;
1389 u32 bucket, obj;
1390
1391 bucket = cb->args[0];
1392 obj = cb->args[1];
1393 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1394 if (!flow)
1395 break;
1396
1397 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1398 NETLINK_CB(cb->skb).portid,
1399 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1400 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1401 break;
1402
1403 cb->args[0] = bucket;
1404 cb->args[1] = obj;
1405 }
1406 rcu_read_unlock();
1407 return skb->len;
1408}
1409
1410static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1411 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1412 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1413 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1414 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1415 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1416 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1417 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1418};
1419
1420static const struct genl_ops dp_flow_genl_ops[] = {
1421 { .cmd = OVS_FLOW_CMD_NEW,
1422 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1423 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1424 .doit = ovs_flow_cmd_new
1425 },
1426 { .cmd = OVS_FLOW_CMD_DEL,
1427 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1428 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429 .doit = ovs_flow_cmd_del
1430 },
1431 { .cmd = OVS_FLOW_CMD_GET,
1432 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1433 .flags = 0, /* OK for unprivileged users. */
1434 .doit = ovs_flow_cmd_get,
1435 .dumpit = ovs_flow_cmd_dump
1436 },
1437 { .cmd = OVS_FLOW_CMD_SET,
1438 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1439 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1440 .doit = ovs_flow_cmd_set,
1441 },
1442};
1443
1444static struct genl_family dp_flow_genl_family __ro_after_init = {
1445 .hdrsize = sizeof(struct ovs_header),
1446 .name = OVS_FLOW_FAMILY,
1447 .version = OVS_FLOW_VERSION,
1448 .maxattr = OVS_FLOW_ATTR_MAX,
1449 .policy = flow_policy,
1450 .netnsok = true,
1451 .parallel_ops = true,
1452 .ops = dp_flow_genl_ops,
1453 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1454 .mcgrps = &ovs_dp_flow_multicast_group,
1455 .n_mcgrps = 1,
1456 .module = THIS_MODULE,
1457};
1458
1459static size_t ovs_dp_cmd_msg_size(void)
1460{
1461 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1462
1463 msgsize += nla_total_size(IFNAMSIZ);
1464 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1465 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1466 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1467
1468 return msgsize;
1469}
1470
1471/* Called with ovs_mutex. */
1472static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1473 u32 portid, u32 seq, u32 flags, u8 cmd)
1474{
1475 struct ovs_header *ovs_header;
1476 struct ovs_dp_stats dp_stats;
1477 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1478 int err;
1479
1480 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1481 flags, cmd);
1482 if (!ovs_header)
1483 goto error;
1484
1485 ovs_header->dp_ifindex = get_dpifindex(dp);
1486
1487 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1488 if (err)
1489 goto nla_put_failure;
1490
1491 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1492 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1493 &dp_stats, OVS_DP_ATTR_PAD))
1494 goto nla_put_failure;
1495
1496 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1497 sizeof(struct ovs_dp_megaflow_stats),
1498 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1499 goto nla_put_failure;
1500
1501 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1502 goto nla_put_failure;
1503
1504 genlmsg_end(skb, ovs_header);
1505 return 0;
1506
1507nla_put_failure:
1508 genlmsg_cancel(skb, ovs_header);
1509error:
1510 return -EMSGSIZE;
1511}
1512
1513static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1514{
1515 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1516}
1517
1518/* Called with rcu_read_lock or ovs_mutex. */
1519static struct datapath *lookup_datapath(struct net *net,
1520 const struct ovs_header *ovs_header,
1521 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1522{
1523 struct datapath *dp;
1524
1525 if (!a[OVS_DP_ATTR_NAME])
1526 dp = get_dp(net, ovs_header->dp_ifindex);
1527 else {
1528 struct vport *vport;
1529
1530 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1531 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1532 }
1533 return dp ? dp : ERR_PTR(-ENODEV);
1534}
1535
1536static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1537{
1538 struct datapath *dp;
1539
1540 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1541 if (IS_ERR(dp))
1542 return;
1543
1544 WARN(dp->user_features, "Dropping previously announced user features\n");
1545 dp->user_features = 0;
1546}
1547
1548DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1549
1550static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1551{
1552 u32 user_features = 0;
1553
1554 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1555 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1556
1557 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1558 OVS_DP_F_UNALIGNED |
1559 OVS_DP_F_TC_RECIRC_SHARING))
1560 return -EOPNOTSUPP;
1561
1562#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1563 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1564 return -EOPNOTSUPP;
1565#endif
1566 }
1567
1568 dp->user_features = user_features;
1569
1570 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1571 static_branch_enable(&tc_recirc_sharing_support);
1572 else
1573 static_branch_disable(&tc_recirc_sharing_support);
1574
1575 return 0;
1576}
1577
1578static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1579{
1580 struct nlattr **a = info->attrs;
1581 struct vport_parms parms;
1582 struct sk_buff *reply;
1583 struct datapath *dp;
1584 struct vport *vport;
1585 struct ovs_net *ovs_net;
1586 int err, i;
1587
1588 err = -EINVAL;
1589 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1590 goto err;
1591
1592 reply = ovs_dp_cmd_alloc_info();
1593 if (!reply)
1594 return -ENOMEM;
1595
1596 err = -ENOMEM;
1597 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1598 if (dp == NULL)
1599 goto err_free_reply;
1600
1601 ovs_dp_set_net(dp, sock_net(skb->sk));
1602
1603 /* Allocate table. */
1604 err = ovs_flow_tbl_init(&dp->table);
1605 if (err)
1606 goto err_free_dp;
1607
1608 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1609 if (!dp->stats_percpu) {
1610 err = -ENOMEM;
1611 goto err_destroy_table;
1612 }
1613
1614 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1615 sizeof(struct hlist_head),
1616 GFP_KERNEL);
1617 if (!dp->ports) {
1618 err = -ENOMEM;
1619 goto err_destroy_percpu;
1620 }
1621
1622 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1623 INIT_HLIST_HEAD(&dp->ports[i]);
1624
1625 err = ovs_meters_init(dp);
1626 if (err)
1627 goto err_destroy_ports_array;
1628
1629 /* Set up our datapath device. */
1630 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1631 parms.type = OVS_VPORT_TYPE_INTERNAL;
1632 parms.options = NULL;
1633 parms.dp = dp;
1634 parms.port_no = OVSP_LOCAL;
1635 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1636
1637 err = ovs_dp_change(dp, a);
1638 if (err)
1639 goto err_destroy_meters;
1640
1641 /* So far only local changes have been made, now need the lock. */
1642 ovs_lock();
1643
1644 vport = new_vport(&parms);
1645 if (IS_ERR(vport)) {
1646 err = PTR_ERR(vport);
1647 if (err == -EBUSY)
1648 err = -EEXIST;
1649
1650 if (err == -EEXIST) {
1651 /* An outdated user space instance that does not understand
1652 * the concept of user_features has attempted to create a new
1653 * datapath and is likely to reuse it. Drop all user features.
1654 */
1655 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1656 ovs_dp_reset_user_features(skb, info);
1657 }
1658
1659 goto err_destroy_meters;
1660 }
1661
1662 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1663 info->snd_seq, 0, OVS_DP_CMD_NEW);
1664 BUG_ON(err < 0);
1665
1666 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1667 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1668
1669 ovs_unlock();
1670
1671 ovs_notify(&dp_datapath_genl_family, reply, info);
1672 return 0;
1673
1674err_destroy_meters:
1675 ovs_unlock();
1676 ovs_meters_exit(dp);
1677err_destroy_ports_array:
1678 kfree(dp->ports);
1679err_destroy_percpu:
1680 free_percpu(dp->stats_percpu);
1681err_destroy_table:
1682 ovs_flow_tbl_destroy(&dp->table);
1683err_free_dp:
1684 kfree(dp);
1685err_free_reply:
1686 kfree_skb(reply);
1687err:
1688 return err;
1689}
1690
1691/* Called with ovs_mutex. */
1692static void __dp_destroy(struct datapath *dp)
1693{
1694 int i;
1695
1696 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1697 struct vport *vport;
1698 struct hlist_node *n;
1699
1700 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1701 if (vport->port_no != OVSP_LOCAL)
1702 ovs_dp_detach_port(vport);
1703 }
1704
1705 list_del_rcu(&dp->list_node);
1706
1707 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1708 * all ports in datapath are destroyed first before freeing datapath.
1709 */
1710 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1711
1712 /* RCU destroy the flow table */
1713 call_rcu(&dp->rcu, destroy_dp_rcu);
1714}
1715
1716static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1717{
1718 struct sk_buff *reply;
1719 struct datapath *dp;
1720 int err;
1721
1722 reply = ovs_dp_cmd_alloc_info();
1723 if (!reply)
1724 return -ENOMEM;
1725
1726 ovs_lock();
1727 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1728 err = PTR_ERR(dp);
1729 if (IS_ERR(dp))
1730 goto err_unlock_free;
1731
1732 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1733 info->snd_seq, 0, OVS_DP_CMD_DEL);
1734 BUG_ON(err < 0);
1735
1736 __dp_destroy(dp);
1737 ovs_unlock();
1738
1739 ovs_notify(&dp_datapath_genl_family, reply, info);
1740
1741 return 0;
1742
1743err_unlock_free:
1744 ovs_unlock();
1745 kfree_skb(reply);
1746 return err;
1747}
1748
1749static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1750{
1751 struct sk_buff *reply;
1752 struct datapath *dp;
1753 int err;
1754
1755 reply = ovs_dp_cmd_alloc_info();
1756 if (!reply)
1757 return -ENOMEM;
1758
1759 ovs_lock();
1760 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1761 err = PTR_ERR(dp);
1762 if (IS_ERR(dp))
1763 goto err_unlock_free;
1764
1765 err = ovs_dp_change(dp, info->attrs);
1766 if (err)
1767 goto err_unlock_free;
1768
1769 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1770 info->snd_seq, 0, OVS_DP_CMD_SET);
1771 BUG_ON(err < 0);
1772
1773 ovs_unlock();
1774 ovs_notify(&dp_datapath_genl_family, reply, info);
1775
1776 return 0;
1777
1778err_unlock_free:
1779 ovs_unlock();
1780 kfree_skb(reply);
1781 return err;
1782}
1783
1784static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1785{
1786 struct sk_buff *reply;
1787 struct datapath *dp;
1788 int err;
1789
1790 reply = ovs_dp_cmd_alloc_info();
1791 if (!reply)
1792 return -ENOMEM;
1793
1794 ovs_lock();
1795 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1796 if (IS_ERR(dp)) {
1797 err = PTR_ERR(dp);
1798 goto err_unlock_free;
1799 }
1800 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1801 info->snd_seq, 0, OVS_DP_CMD_GET);
1802 BUG_ON(err < 0);
1803 ovs_unlock();
1804
1805 return genlmsg_reply(reply, info);
1806
1807err_unlock_free:
1808 ovs_unlock();
1809 kfree_skb(reply);
1810 return err;
1811}
1812
1813static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1814{
1815 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1816 struct datapath *dp;
1817 int skip = cb->args[0];
1818 int i = 0;
1819
1820 ovs_lock();
1821 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1822 if (i >= skip &&
1823 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1824 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1825 OVS_DP_CMD_GET) < 0)
1826 break;
1827 i++;
1828 }
1829 ovs_unlock();
1830
1831 cb->args[0] = i;
1832
1833 return skb->len;
1834}
1835
1836static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1837 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1838 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1839 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1840};
1841
1842static const struct genl_ops dp_datapath_genl_ops[] = {
1843 { .cmd = OVS_DP_CMD_NEW,
1844 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1845 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1846 .doit = ovs_dp_cmd_new
1847 },
1848 { .cmd = OVS_DP_CMD_DEL,
1849 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1850 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1851 .doit = ovs_dp_cmd_del
1852 },
1853 { .cmd = OVS_DP_CMD_GET,
1854 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1855 .flags = 0, /* OK for unprivileged users. */
1856 .doit = ovs_dp_cmd_get,
1857 .dumpit = ovs_dp_cmd_dump
1858 },
1859 { .cmd = OVS_DP_CMD_SET,
1860 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1861 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1862 .doit = ovs_dp_cmd_set,
1863 },
1864};
1865
1866static struct genl_family dp_datapath_genl_family __ro_after_init = {
1867 .hdrsize = sizeof(struct ovs_header),
1868 .name = OVS_DATAPATH_FAMILY,
1869 .version = OVS_DATAPATH_VERSION,
1870 .maxattr = OVS_DP_ATTR_MAX,
1871 .policy = datapath_policy,
1872 .netnsok = true,
1873 .parallel_ops = true,
1874 .ops = dp_datapath_genl_ops,
1875 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1876 .mcgrps = &ovs_dp_datapath_multicast_group,
1877 .n_mcgrps = 1,
1878 .module = THIS_MODULE,
1879};
1880
1881/* Called with ovs_mutex or RCU read lock. */
1882static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1883 struct net *net, u32 portid, u32 seq,
1884 u32 flags, u8 cmd, gfp_t gfp)
1885{
1886 struct ovs_header *ovs_header;
1887 struct ovs_vport_stats vport_stats;
1888 int err;
1889
1890 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1891 flags, cmd);
1892 if (!ovs_header)
1893 return -EMSGSIZE;
1894
1895 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1896
1897 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1898 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1899 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1900 ovs_vport_name(vport)) ||
1901 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1902 goto nla_put_failure;
1903
1904 if (!net_eq(net, dev_net(vport->dev))) {
1905 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1906
1907 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1908 goto nla_put_failure;
1909 }
1910
1911 ovs_vport_get_stats(vport, &vport_stats);
1912 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1913 sizeof(struct ovs_vport_stats), &vport_stats,
1914 OVS_VPORT_ATTR_PAD))
1915 goto nla_put_failure;
1916
1917 if (ovs_vport_get_upcall_portids(vport, skb))
1918 goto nla_put_failure;
1919
1920 err = ovs_vport_get_options(vport, skb);
1921 if (err == -EMSGSIZE)
1922 goto error;
1923
1924 genlmsg_end(skb, ovs_header);
1925 return 0;
1926
1927nla_put_failure:
1928 err = -EMSGSIZE;
1929error:
1930 genlmsg_cancel(skb, ovs_header);
1931 return err;
1932}
1933
1934static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1935{
1936 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1937}
1938
1939/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1940struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1941 u32 portid, u32 seq, u8 cmd)
1942{
1943 struct sk_buff *skb;
1944 int retval;
1945
1946 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1947 if (!skb)
1948 return ERR_PTR(-ENOMEM);
1949
1950 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1951 GFP_KERNEL);
1952 BUG_ON(retval < 0);
1953
1954 return skb;
1955}
1956
1957/* Called with ovs_mutex or RCU read lock. */
1958static struct vport *lookup_vport(struct net *net,
1959 const struct ovs_header *ovs_header,
1960 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1961{
1962 struct datapath *dp;
1963 struct vport *vport;
1964
1965 if (a[OVS_VPORT_ATTR_IFINDEX])
1966 return ERR_PTR(-EOPNOTSUPP);
1967 if (a[OVS_VPORT_ATTR_NAME]) {
1968 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1969 if (!vport)
1970 return ERR_PTR(-ENODEV);
1971 if (ovs_header->dp_ifindex &&
1972 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1973 return ERR_PTR(-ENODEV);
1974 return vport;
1975 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1976 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1977
1978 if (port_no >= DP_MAX_PORTS)
1979 return ERR_PTR(-EFBIG);
1980
1981 dp = get_dp(net, ovs_header->dp_ifindex);
1982 if (!dp)
1983 return ERR_PTR(-ENODEV);
1984
1985 vport = ovs_vport_ovsl_rcu(dp, port_no);
1986 if (!vport)
1987 return ERR_PTR(-ENODEV);
1988 return vport;
1989 } else
1990 return ERR_PTR(-EINVAL);
1991
1992}
1993
1994static unsigned int ovs_get_max_headroom(struct datapath *dp)
1995{
1996 unsigned int dev_headroom, max_headroom = 0;
1997 struct net_device *dev;
1998 struct vport *vport;
1999 int i;
2000
2001 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2002 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2003 dev = vport->dev;
2004 dev_headroom = netdev_get_fwd_headroom(dev);
2005 if (dev_headroom > max_headroom)
2006 max_headroom = dev_headroom;
2007 }
2008 }
2009
2010 return max_headroom;
2011}
2012
2013/* Called with ovs_mutex */
2014static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2015{
2016 struct vport *vport;
2017 int i;
2018
2019 dp->max_headroom = new_headroom;
2020 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2021 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2022 netdev_set_rx_headroom(vport->dev, new_headroom);
2023}
2024
2025static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2026{
2027 struct nlattr **a = info->attrs;
2028 struct ovs_header *ovs_header = info->userhdr;
2029 struct vport_parms parms;
2030 struct sk_buff *reply;
2031 struct vport *vport;
2032 struct datapath *dp;
2033 unsigned int new_headroom;
2034 u32 port_no;
2035 int err;
2036
2037 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2038 !a[OVS_VPORT_ATTR_UPCALL_PID])
2039 return -EINVAL;
2040 if (a[OVS_VPORT_ATTR_IFINDEX])
2041 return -EOPNOTSUPP;
2042
2043 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2044 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2045 if (port_no >= DP_MAX_PORTS)
2046 return -EFBIG;
2047
2048 reply = ovs_vport_cmd_alloc_info();
2049 if (!reply)
2050 return -ENOMEM;
2051
2052 ovs_lock();
2053restart:
2054 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2055 err = -ENODEV;
2056 if (!dp)
2057 goto exit_unlock_free;
2058
2059 if (port_no) {
2060 vport = ovs_vport_ovsl(dp, port_no);
2061 err = -EBUSY;
2062 if (vport)
2063 goto exit_unlock_free;
2064 } else {
2065 for (port_no = 1; ; port_no++) {
2066 if (port_no >= DP_MAX_PORTS) {
2067 err = -EFBIG;
2068 goto exit_unlock_free;
2069 }
2070 vport = ovs_vport_ovsl(dp, port_no);
2071 if (!vport)
2072 break;
2073 }
2074 }
2075
2076 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2077 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2078 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2079 parms.dp = dp;
2080 parms.port_no = port_no;
2081 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2082
2083 vport = new_vport(&parms);
2084 err = PTR_ERR(vport);
2085 if (IS_ERR(vport)) {
2086 if (err == -EAGAIN)
2087 goto restart;
2088 goto exit_unlock_free;
2089 }
2090
2091 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2092 info->snd_portid, info->snd_seq, 0,
2093 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2094
2095 new_headroom = netdev_get_fwd_headroom(vport->dev);
2096
2097 if (new_headroom > dp->max_headroom)
2098 ovs_update_headroom(dp, new_headroom);
2099 else
2100 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2101
2102 BUG_ON(err < 0);
2103 ovs_unlock();
2104
2105 ovs_notify(&dp_vport_genl_family, reply, info);
2106 return 0;
2107
2108exit_unlock_free:
2109 ovs_unlock();
2110 kfree_skb(reply);
2111 return err;
2112}
2113
2114static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2115{
2116 struct nlattr **a = info->attrs;
2117 struct sk_buff *reply;
2118 struct vport *vport;
2119 int err;
2120
2121 reply = ovs_vport_cmd_alloc_info();
2122 if (!reply)
2123 return -ENOMEM;
2124
2125 ovs_lock();
2126 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2127 err = PTR_ERR(vport);
2128 if (IS_ERR(vport))
2129 goto exit_unlock_free;
2130
2131 if (a[OVS_VPORT_ATTR_TYPE] &&
2132 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2133 err = -EINVAL;
2134 goto exit_unlock_free;
2135 }
2136
2137 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2138 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2139 if (err)
2140 goto exit_unlock_free;
2141 }
2142
2143
2144 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2145 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2146
2147 err = ovs_vport_set_upcall_portids(vport, ids);
2148 if (err)
2149 goto exit_unlock_free;
2150 }
2151
2152 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2153 info->snd_portid, info->snd_seq, 0,
2154 OVS_VPORT_CMD_SET, GFP_KERNEL);
2155 BUG_ON(err < 0);
2156
2157 ovs_unlock();
2158 ovs_notify(&dp_vport_genl_family, reply, info);
2159 return 0;
2160
2161exit_unlock_free:
2162 ovs_unlock();
2163 kfree_skb(reply);
2164 return err;
2165}
2166
2167static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2168{
2169 bool update_headroom = false;
2170 struct nlattr **a = info->attrs;
2171 struct sk_buff *reply;
2172 struct datapath *dp;
2173 struct vport *vport;
2174 unsigned int new_headroom;
2175 int err;
2176
2177 reply = ovs_vport_cmd_alloc_info();
2178 if (!reply)
2179 return -ENOMEM;
2180
2181 ovs_lock();
2182 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2183 err = PTR_ERR(vport);
2184 if (IS_ERR(vport))
2185 goto exit_unlock_free;
2186
2187 if (vport->port_no == OVSP_LOCAL) {
2188 err = -EINVAL;
2189 goto exit_unlock_free;
2190 }
2191
2192 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2193 info->snd_portid, info->snd_seq, 0,
2194 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2195 BUG_ON(err < 0);
2196
2197 /* the vport deletion may trigger dp headroom update */
2198 dp = vport->dp;
2199 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2200 update_headroom = true;
2201
2202 netdev_reset_rx_headroom(vport->dev);
2203 ovs_dp_detach_port(vport);
2204
2205 if (update_headroom) {
2206 new_headroom = ovs_get_max_headroom(dp);
2207
2208 if (new_headroom < dp->max_headroom)
2209 ovs_update_headroom(dp, new_headroom);
2210 }
2211 ovs_unlock();
2212
2213 ovs_notify(&dp_vport_genl_family, reply, info);
2214 return 0;
2215
2216exit_unlock_free:
2217 ovs_unlock();
2218 kfree_skb(reply);
2219 return err;
2220}
2221
2222static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2223{
2224 struct nlattr **a = info->attrs;
2225 struct ovs_header *ovs_header = info->userhdr;
2226 struct sk_buff *reply;
2227 struct vport *vport;
2228 int err;
2229
2230 reply = ovs_vport_cmd_alloc_info();
2231 if (!reply)
2232 return -ENOMEM;
2233
2234 rcu_read_lock();
2235 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2236 err = PTR_ERR(vport);
2237 if (IS_ERR(vport))
2238 goto exit_unlock_free;
2239 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2240 info->snd_portid, info->snd_seq, 0,
2241 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2242 BUG_ON(err < 0);
2243 rcu_read_unlock();
2244
2245 return genlmsg_reply(reply, info);
2246
2247exit_unlock_free:
2248 rcu_read_unlock();
2249 kfree_skb(reply);
2250 return err;
2251}
2252
2253static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2254{
2255 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2256 struct datapath *dp;
2257 int bucket = cb->args[0], skip = cb->args[1];
2258 int i, j = 0;
2259
2260 rcu_read_lock();
2261 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2262 if (!dp) {
2263 rcu_read_unlock();
2264 return -ENODEV;
2265 }
2266 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2267 struct vport *vport;
2268
2269 j = 0;
2270 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2271 if (j >= skip &&
2272 ovs_vport_cmd_fill_info(vport, skb,
2273 sock_net(skb->sk),
2274 NETLINK_CB(cb->skb).portid,
2275 cb->nlh->nlmsg_seq,
2276 NLM_F_MULTI,
2277 OVS_VPORT_CMD_GET,
2278 GFP_ATOMIC) < 0)
2279 goto out;
2280
2281 j++;
2282 }
2283 skip = 0;
2284 }
2285out:
2286 rcu_read_unlock();
2287
2288 cb->args[0] = i;
2289 cb->args[1] = j;
2290
2291 return skb->len;
2292}
2293
2294static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2295 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2296 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2297 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2298 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2299 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2300 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2301 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2302 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2303};
2304
2305static const struct genl_ops dp_vport_genl_ops[] = {
2306 { .cmd = OVS_VPORT_CMD_NEW,
2307 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2308 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2309 .doit = ovs_vport_cmd_new
2310 },
2311 { .cmd = OVS_VPORT_CMD_DEL,
2312 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2313 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2314 .doit = ovs_vport_cmd_del
2315 },
2316 { .cmd = OVS_VPORT_CMD_GET,
2317 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2318 .flags = 0, /* OK for unprivileged users. */
2319 .doit = ovs_vport_cmd_get,
2320 .dumpit = ovs_vport_cmd_dump
2321 },
2322 { .cmd = OVS_VPORT_CMD_SET,
2323 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2324 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2325 .doit = ovs_vport_cmd_set,
2326 },
2327};
2328
2329struct genl_family dp_vport_genl_family __ro_after_init = {
2330 .hdrsize = sizeof(struct ovs_header),
2331 .name = OVS_VPORT_FAMILY,
2332 .version = OVS_VPORT_VERSION,
2333 .maxattr = OVS_VPORT_ATTR_MAX,
2334 .policy = vport_policy,
2335 .netnsok = true,
2336 .parallel_ops = true,
2337 .ops = dp_vport_genl_ops,
2338 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2339 .mcgrps = &ovs_dp_vport_multicast_group,
2340 .n_mcgrps = 1,
2341 .module = THIS_MODULE,
2342};
2343
2344static struct genl_family * const dp_genl_families[] = {
2345 &dp_datapath_genl_family,
2346 &dp_vport_genl_family,
2347 &dp_flow_genl_family,
2348 &dp_packet_genl_family,
2349 &dp_meter_genl_family,
2350#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2351 &dp_ct_limit_genl_family,
2352#endif
2353};
2354
2355static void dp_unregister_genl(int n_families)
2356{
2357 int i;
2358
2359 for (i = 0; i < n_families; i++)
2360 genl_unregister_family(dp_genl_families[i]);
2361}
2362
2363static int __init dp_register_genl(void)
2364{
2365 int err;
2366 int i;
2367
2368 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2369
2370 err = genl_register_family(dp_genl_families[i]);
2371 if (err)
2372 goto error;
2373 }
2374
2375 return 0;
2376
2377error:
2378 dp_unregister_genl(i);
2379 return err;
2380}
2381
2382static int __net_init ovs_init_net(struct net *net)
2383{
2384 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2385
2386 INIT_LIST_HEAD(&ovs_net->dps);
2387 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2388 return ovs_ct_init(net);
2389}
2390
2391static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2392 struct list_head *head)
2393{
2394 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2395 struct datapath *dp;
2396
2397 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2398 int i;
2399
2400 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2401 struct vport *vport;
2402
2403 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2404 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2405 continue;
2406
2407 if (dev_net(vport->dev) == dnet)
2408 list_add(&vport->detach_list, head);
2409 }
2410 }
2411 }
2412}
2413
2414static void __net_exit ovs_exit_net(struct net *dnet)
2415{
2416 struct datapath *dp, *dp_next;
2417 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2418 struct vport *vport, *vport_next;
2419 struct net *net;
2420 LIST_HEAD(head);
2421
2422 ovs_ct_exit(dnet);
2423 ovs_lock();
2424 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2425 __dp_destroy(dp);
2426
2427 down_read(&net_rwsem);
2428 for_each_net(net)
2429 list_vports_from_net(net, dnet, &head);
2430 up_read(&net_rwsem);
2431
2432 /* Detach all vports from given namespace. */
2433 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2434 list_del(&vport->detach_list);
2435 ovs_dp_detach_port(vport);
2436 }
2437
2438 ovs_unlock();
2439
2440 cancel_work_sync(&ovs_net->dp_notify_work);
2441}
2442
2443static struct pernet_operations ovs_net_ops = {
2444 .init = ovs_init_net,
2445 .exit = ovs_exit_net,
2446 .id = &ovs_net_id,
2447 .size = sizeof(struct ovs_net),
2448};
2449
2450static int __init dp_init(void)
2451{
2452 int err;
2453
2454 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2455
2456 pr_info("Open vSwitch switching datapath\n");
2457
2458 err = action_fifos_init();
2459 if (err)
2460 goto error;
2461
2462 err = ovs_internal_dev_rtnl_link_register();
2463 if (err)
2464 goto error_action_fifos_exit;
2465
2466 err = ovs_flow_init();
2467 if (err)
2468 goto error_unreg_rtnl_link;
2469
2470 err = ovs_vport_init();
2471 if (err)
2472 goto error_flow_exit;
2473
2474 err = register_pernet_device(&ovs_net_ops);
2475 if (err)
2476 goto error_vport_exit;
2477
2478 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2479 if (err)
2480 goto error_netns_exit;
2481
2482 err = ovs_netdev_init();
2483 if (err)
2484 goto error_unreg_notifier;
2485
2486 err = dp_register_genl();
2487 if (err < 0)
2488 goto error_unreg_netdev;
2489
2490 return 0;
2491
2492error_unreg_netdev:
2493 ovs_netdev_exit();
2494error_unreg_notifier:
2495 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2496error_netns_exit:
2497 unregister_pernet_device(&ovs_net_ops);
2498error_vport_exit:
2499 ovs_vport_exit();
2500error_flow_exit:
2501 ovs_flow_exit();
2502error_unreg_rtnl_link:
2503 ovs_internal_dev_rtnl_link_unregister();
2504error_action_fifos_exit:
2505 action_fifos_exit();
2506error:
2507 return err;
2508}
2509
2510static void dp_cleanup(void)
2511{
2512 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2513 ovs_netdev_exit();
2514 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2515 unregister_pernet_device(&ovs_net_ops);
2516 rcu_barrier();
2517 ovs_vport_exit();
2518 ovs_flow_exit();
2519 ovs_internal_dev_rtnl_link_unregister();
2520 action_fifos_exit();
2521}
2522
2523module_init(dp_init);
2524module_exit(dp_cleanup);
2525
2526MODULE_DESCRIPTION("Open vSwitch switching datapath");
2527MODULE_LICENSE("GPL");
2528MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2529MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2530MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2531MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2532MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2533MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2007-2014 Nicira, Inc.
4 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/init.h>
9#include <linux/module.h>
10#include <linux/if_arp.h>
11#include <linux/if_vlan.h>
12#include <linux/in.h>
13#include <linux/ip.h>
14#include <linux/jhash.h>
15#include <linux/delay.h>
16#include <linux/time.h>
17#include <linux/etherdevice.h>
18#include <linux/genetlink.h>
19#include <linux/kernel.h>
20#include <linux/kthread.h>
21#include <linux/mutex.h>
22#include <linux/percpu.h>
23#include <linux/rcupdate.h>
24#include <linux/tcp.h>
25#include <linux/udp.h>
26#include <linux/ethtool.h>
27#include <linux/wait.h>
28#include <asm/div64.h>
29#include <linux/highmem.h>
30#include <linux/netfilter_bridge.h>
31#include <linux/netfilter_ipv4.h>
32#include <linux/inetdevice.h>
33#include <linux/list.h>
34#include <linux/openvswitch.h>
35#include <linux/rculist.h>
36#include <linux/dmi.h>
37#include <net/genetlink.h>
38#include <net/gso.h>
39#include <net/net_namespace.h>
40#include <net/netns/generic.h>
41#include <net/pkt_cls.h>
42
43#include "datapath.h"
44#include "drop.h"
45#include "flow.h"
46#include "flow_table.h"
47#include "flow_netlink.h"
48#include "meter.h"
49#include "openvswitch_trace.h"
50#include "vport-internal_dev.h"
51#include "vport-netdev.h"
52
53unsigned int ovs_net_id __read_mostly;
54
55static struct genl_family dp_packet_genl_family;
56static struct genl_family dp_flow_genl_family;
57static struct genl_family dp_datapath_genl_family;
58
59static const struct nla_policy flow_policy[];
60
61static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
62 .name = OVS_FLOW_MCGROUP,
63};
64
65static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
66 .name = OVS_DATAPATH_MCGROUP,
67};
68
69static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
70 .name = OVS_VPORT_MCGROUP,
71};
72
73/* Check if need to build a reply message.
74 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
75static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
76 unsigned int group)
77{
78 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
79 genl_has_listeners(family, genl_info_net(info), group);
80}
81
82static void ovs_notify(struct genl_family *family,
83 struct sk_buff *skb, struct genl_info *info)
84{
85 genl_notify(family, skb, info, 0, GFP_KERNEL);
86}
87
88/**
89 * DOC: Locking:
90 *
91 * All writes e.g. Writes to device state (add/remove datapath, port, set
92 * operations on vports, etc.), Writes to other state (flow table
93 * modifications, set miscellaneous datapath parameters, etc.) are protected
94 * by ovs_lock.
95 *
96 * Reads are protected by RCU.
97 *
98 * There are a few special cases (mostly stats) that have their own
99 * synchronization but they nest under all of above and don't interact with
100 * each other.
101 *
102 * The RTNL lock nests inside ovs_mutex.
103 */
104
105static DEFINE_MUTEX(ovs_mutex);
106
107void ovs_lock(void)
108{
109 mutex_lock(&ovs_mutex);
110}
111
112void ovs_unlock(void)
113{
114 mutex_unlock(&ovs_mutex);
115}
116
117#ifdef CONFIG_LOCKDEP
118int lockdep_ovsl_is_held(void)
119{
120 if (debug_locks)
121 return lockdep_is_held(&ovs_mutex);
122 else
123 return 1;
124}
125#endif
126
127static struct vport *new_vport(const struct vport_parms *);
128static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
129 const struct sw_flow_key *,
130 const struct dp_upcall_info *,
131 uint32_t cutlen);
132static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
133 const struct sw_flow_key *,
134 const struct dp_upcall_info *,
135 uint32_t cutlen);
136
137static void ovs_dp_masks_rebalance(struct work_struct *work);
138
139static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
140
141/* Must be called with rcu_read_lock or ovs_mutex. */
142const char *ovs_dp_name(const struct datapath *dp)
143{
144 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
145 return ovs_vport_name(vport);
146}
147
148static int get_dpifindex(const struct datapath *dp)
149{
150 struct vport *local;
151 int ifindex;
152
153 rcu_read_lock();
154
155 local = ovs_vport_rcu(dp, OVSP_LOCAL);
156 if (local)
157 ifindex = local->dev->ifindex;
158 else
159 ifindex = 0;
160
161 rcu_read_unlock();
162
163 return ifindex;
164}
165
166static void destroy_dp_rcu(struct rcu_head *rcu)
167{
168 struct datapath *dp = container_of(rcu, struct datapath, rcu);
169
170 ovs_flow_tbl_destroy(&dp->table);
171 free_percpu(dp->stats_percpu);
172 kfree(dp->ports);
173 ovs_meters_exit(dp);
174 kfree(rcu_dereference_raw(dp->upcall_portids));
175 kfree(dp);
176}
177
178static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
179 u16 port_no)
180{
181 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
182}
183
184/* Called with ovs_mutex or RCU read lock. */
185struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
186{
187 struct vport *vport;
188 struct hlist_head *head;
189
190 head = vport_hash_bucket(dp, port_no);
191 hlist_for_each_entry_rcu(vport, head, dp_hash_node,
192 lockdep_ovsl_is_held()) {
193 if (vport->port_no == port_no)
194 return vport;
195 }
196 return NULL;
197}
198
199/* Called with ovs_mutex. */
200static struct vport *new_vport(const struct vport_parms *parms)
201{
202 struct vport *vport;
203
204 vport = ovs_vport_add(parms);
205 if (!IS_ERR(vport)) {
206 struct datapath *dp = parms->dp;
207 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
208
209 hlist_add_head_rcu(&vport->dp_hash_node, head);
210 }
211 return vport;
212}
213
214static void ovs_vport_update_upcall_stats(struct sk_buff *skb,
215 const struct dp_upcall_info *upcall_info,
216 bool upcall_result)
217{
218 struct vport *p = OVS_CB(skb)->input_vport;
219 struct vport_upcall_stats_percpu *stats;
220
221 if (upcall_info->cmd != OVS_PACKET_CMD_MISS &&
222 upcall_info->cmd != OVS_PACKET_CMD_ACTION)
223 return;
224
225 stats = this_cpu_ptr(p->upcall_stats);
226 u64_stats_update_begin(&stats->syncp);
227 if (upcall_result)
228 u64_stats_inc(&stats->n_success);
229 else
230 u64_stats_inc(&stats->n_fail);
231 u64_stats_update_end(&stats->syncp);
232}
233
234void ovs_dp_detach_port(struct vport *p)
235{
236 ASSERT_OVSL();
237
238 /* First drop references to device. */
239 hlist_del_rcu(&p->dp_hash_node);
240
241 /* Then destroy it. */
242 ovs_vport_del(p);
243}
244
245/* Must be called with rcu_read_lock. */
246void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
247{
248 const struct vport *p = OVS_CB(skb)->input_vport;
249 struct datapath *dp = p->dp;
250 struct sw_flow *flow;
251 struct sw_flow_actions *sf_acts;
252 struct dp_stats_percpu *stats;
253 u64 *stats_counter;
254 u32 n_mask_hit;
255 u32 n_cache_hit;
256 int error;
257
258 stats = this_cpu_ptr(dp->stats_percpu);
259
260 /* Look up flow. */
261 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
262 &n_mask_hit, &n_cache_hit);
263 if (unlikely(!flow)) {
264 struct dp_upcall_info upcall;
265
266 memset(&upcall, 0, sizeof(upcall));
267 upcall.cmd = OVS_PACKET_CMD_MISS;
268
269 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
270 upcall.portid =
271 ovs_dp_get_upcall_portid(dp, smp_processor_id());
272 else
273 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
274
275 upcall.mru = OVS_CB(skb)->mru;
276 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
277 switch (error) {
278 case 0:
279 case -EAGAIN:
280 case -ERESTARTSYS:
281 case -EINTR:
282 consume_skb(skb);
283 break;
284 default:
285 kfree_skb(skb);
286 break;
287 }
288 stats_counter = &stats->n_missed;
289 goto out;
290 }
291
292 ovs_flow_stats_update(flow, key->tp.flags, skb);
293 sf_acts = rcu_dereference(flow->sf_acts);
294 error = ovs_execute_actions(dp, skb, sf_acts, key);
295 if (unlikely(error))
296 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
297 ovs_dp_name(dp), error);
298
299 stats_counter = &stats->n_hit;
300
301out:
302 /* Update datapath statistics. */
303 u64_stats_update_begin(&stats->syncp);
304 (*stats_counter)++;
305 stats->n_mask_hit += n_mask_hit;
306 stats->n_cache_hit += n_cache_hit;
307 u64_stats_update_end(&stats->syncp);
308}
309
310int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
311 const struct sw_flow_key *key,
312 const struct dp_upcall_info *upcall_info,
313 uint32_t cutlen)
314{
315 struct dp_stats_percpu *stats;
316 int err;
317
318 if (trace_ovs_dp_upcall_enabled())
319 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
320
321 if (upcall_info->portid == 0) {
322 err = -ENOTCONN;
323 goto err;
324 }
325
326 if (!skb_is_gso(skb))
327 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
328 else
329 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
330
331 ovs_vport_update_upcall_stats(skb, upcall_info, !err);
332 if (err)
333 goto err;
334
335 return 0;
336
337err:
338 stats = this_cpu_ptr(dp->stats_percpu);
339
340 u64_stats_update_begin(&stats->syncp);
341 stats->n_lost++;
342 u64_stats_update_end(&stats->syncp);
343
344 return err;
345}
346
347static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
348 const struct sw_flow_key *key,
349 const struct dp_upcall_info *upcall_info,
350 uint32_t cutlen)
351{
352 unsigned int gso_type = skb_shinfo(skb)->gso_type;
353 struct sw_flow_key later_key;
354 struct sk_buff *segs, *nskb;
355 int err;
356
357 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
358 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
359 if (IS_ERR(segs))
360 return PTR_ERR(segs);
361 if (segs == NULL)
362 return -EINVAL;
363
364 if (gso_type & SKB_GSO_UDP) {
365 /* The initial flow key extracted by ovs_flow_key_extract()
366 * in this case is for a first fragment, so we need to
367 * properly mark later fragments.
368 */
369 later_key = *key;
370 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
371 }
372
373 /* Queue all of the segments. */
374 skb_list_walk_safe(segs, skb, nskb) {
375 if (gso_type & SKB_GSO_UDP && skb != segs)
376 key = &later_key;
377
378 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
379 if (err)
380 break;
381
382 }
383
384 /* Free all of the segments. */
385 skb_list_walk_safe(segs, skb, nskb) {
386 if (err)
387 kfree_skb(skb);
388 else
389 consume_skb(skb);
390 }
391 return err;
392}
393
394static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
395 unsigned int hdrlen, int actions_attrlen)
396{
397 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
398 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
399 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
400 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
401 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
402
403 /* OVS_PACKET_ATTR_USERDATA */
404 if (upcall_info->userdata)
405 size += NLA_ALIGN(upcall_info->userdata->nla_len);
406
407 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
408 if (upcall_info->egress_tun_info)
409 size += nla_total_size(ovs_tun_key_attr_size());
410
411 /* OVS_PACKET_ATTR_ACTIONS */
412 if (upcall_info->actions_len)
413 size += nla_total_size(actions_attrlen);
414
415 /* OVS_PACKET_ATTR_MRU */
416 if (upcall_info->mru)
417 size += nla_total_size(sizeof(upcall_info->mru));
418
419 return size;
420}
421
422static void pad_packet(struct datapath *dp, struct sk_buff *skb)
423{
424 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
425 size_t plen = NLA_ALIGN(skb->len) - skb->len;
426
427 if (plen > 0)
428 skb_put_zero(skb, plen);
429 }
430}
431
432static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
433 const struct sw_flow_key *key,
434 const struct dp_upcall_info *upcall_info,
435 uint32_t cutlen)
436{
437 struct ovs_header *upcall;
438 struct sk_buff *nskb = NULL;
439 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
440 struct nlattr *nla;
441 size_t len;
442 unsigned int hlen;
443 int err, dp_ifindex;
444 u64 hash;
445
446 dp_ifindex = get_dpifindex(dp);
447 if (!dp_ifindex)
448 return -ENODEV;
449
450 if (skb_vlan_tag_present(skb)) {
451 nskb = skb_clone(skb, GFP_ATOMIC);
452 if (!nskb)
453 return -ENOMEM;
454
455 nskb = __vlan_hwaccel_push_inside(nskb);
456 if (!nskb)
457 return -ENOMEM;
458
459 skb = nskb;
460 }
461
462 if (nla_attr_size(skb->len) > USHRT_MAX) {
463 err = -EFBIG;
464 goto out;
465 }
466
467 /* Complete checksum if needed */
468 if (skb->ip_summed == CHECKSUM_PARTIAL &&
469 (err = skb_csum_hwoffload_help(skb, 0)))
470 goto out;
471
472 /* Older versions of OVS user space enforce alignment of the last
473 * Netlink attribute to NLA_ALIGNTO which would require extensive
474 * padding logic. Only perform zerocopy if padding is not required.
475 */
476 if (dp->user_features & OVS_DP_F_UNALIGNED)
477 hlen = skb_zerocopy_headlen(skb);
478 else
479 hlen = skb->len;
480
481 len = upcall_msg_size(upcall_info, hlen - cutlen,
482 OVS_CB(skb)->acts_origlen);
483 user_skb = genlmsg_new(len, GFP_ATOMIC);
484 if (!user_skb) {
485 err = -ENOMEM;
486 goto out;
487 }
488
489 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
490 0, upcall_info->cmd);
491 if (!upcall) {
492 err = -EINVAL;
493 goto out;
494 }
495 upcall->dp_ifindex = dp_ifindex;
496
497 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
498 if (err)
499 goto out;
500
501 if (upcall_info->userdata)
502 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
503 nla_len(upcall_info->userdata),
504 nla_data(upcall_info->userdata));
505
506 if (upcall_info->egress_tun_info) {
507 nla = nla_nest_start_noflag(user_skb,
508 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
509 if (!nla) {
510 err = -EMSGSIZE;
511 goto out;
512 }
513 err = ovs_nla_put_tunnel_info(user_skb,
514 upcall_info->egress_tun_info);
515 if (err)
516 goto out;
517
518 nla_nest_end(user_skb, nla);
519 }
520
521 if (upcall_info->actions_len) {
522 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
523 if (!nla) {
524 err = -EMSGSIZE;
525 goto out;
526 }
527 err = ovs_nla_put_actions(upcall_info->actions,
528 upcall_info->actions_len,
529 user_skb);
530 if (!err)
531 nla_nest_end(user_skb, nla);
532 else
533 nla_nest_cancel(user_skb, nla);
534 }
535
536 /* Add OVS_PACKET_ATTR_MRU */
537 if (upcall_info->mru &&
538 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
539 err = -ENOBUFS;
540 goto out;
541 }
542
543 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
544 if (cutlen > 0 &&
545 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
546 err = -ENOBUFS;
547 goto out;
548 }
549
550 /* Add OVS_PACKET_ATTR_HASH */
551 hash = skb_get_hash_raw(skb);
552 if (skb->sw_hash)
553 hash |= OVS_PACKET_HASH_SW_BIT;
554
555 if (skb->l4_hash)
556 hash |= OVS_PACKET_HASH_L4_BIT;
557
558 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
559 err = -ENOBUFS;
560 goto out;
561 }
562
563 /* Only reserve room for attribute header, packet data is added
564 * in skb_zerocopy() */
565 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
566 err = -ENOBUFS;
567 goto out;
568 }
569 nla->nla_len = nla_attr_size(skb->len - cutlen);
570
571 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
572 if (err)
573 goto out;
574
575 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
576 pad_packet(dp, user_skb);
577
578 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
579
580 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
581 user_skb = NULL;
582out:
583 if (err)
584 skb_tx_error(skb);
585 consume_skb(user_skb);
586 consume_skb(nskb);
587
588 return err;
589}
590
591static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
592{
593 struct ovs_header *ovs_header = genl_info_userhdr(info);
594 struct net *net = sock_net(skb->sk);
595 struct nlattr **a = info->attrs;
596 struct sw_flow_actions *acts;
597 struct sk_buff *packet;
598 struct sw_flow *flow;
599 struct sw_flow_actions *sf_acts;
600 struct datapath *dp;
601 struct vport *input_vport;
602 u16 mru = 0;
603 u64 hash;
604 int len;
605 int err;
606 bool log = !a[OVS_PACKET_ATTR_PROBE];
607
608 err = -EINVAL;
609 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
610 !a[OVS_PACKET_ATTR_ACTIONS])
611 goto err;
612
613 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
614 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
615 err = -ENOMEM;
616 if (!packet)
617 goto err;
618 skb_reserve(packet, NET_IP_ALIGN);
619
620 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
621
622 /* Set packet's mru */
623 if (a[OVS_PACKET_ATTR_MRU]) {
624 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
625 packet->ignore_df = 1;
626 }
627 OVS_CB(packet)->mru = mru;
628
629 if (a[OVS_PACKET_ATTR_HASH]) {
630 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
631
632 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
633 !!(hash & OVS_PACKET_HASH_SW_BIT),
634 !!(hash & OVS_PACKET_HASH_L4_BIT));
635 }
636
637 /* Build an sw_flow for sending this packet. */
638 flow = ovs_flow_alloc();
639 err = PTR_ERR(flow);
640 if (IS_ERR(flow))
641 goto err_kfree_skb;
642
643 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
644 packet, &flow->key, log);
645 if (err)
646 goto err_flow_free;
647
648 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
649 &flow->key, &acts, log);
650 if (err)
651 goto err_flow_free;
652
653 rcu_assign_pointer(flow->sf_acts, acts);
654 packet->priority = flow->key.phy.priority;
655 packet->mark = flow->key.phy.skb_mark;
656
657 rcu_read_lock();
658 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
659 err = -ENODEV;
660 if (!dp)
661 goto err_unlock;
662
663 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
664 if (!input_vport)
665 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
666
667 if (!input_vport)
668 goto err_unlock;
669
670 packet->dev = input_vport->dev;
671 OVS_CB(packet)->input_vport = input_vport;
672 sf_acts = rcu_dereference(flow->sf_acts);
673
674 local_bh_disable();
675 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
676 local_bh_enable();
677 rcu_read_unlock();
678
679 ovs_flow_free(flow, false);
680 return err;
681
682err_unlock:
683 rcu_read_unlock();
684err_flow_free:
685 ovs_flow_free(flow, false);
686err_kfree_skb:
687 kfree_skb(packet);
688err:
689 return err;
690}
691
692static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
693 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
694 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
695 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
696 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
697 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
698 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
699};
700
701static const struct genl_small_ops dp_packet_genl_ops[] = {
702 { .cmd = OVS_PACKET_CMD_EXECUTE,
703 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
704 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
705 .doit = ovs_packet_cmd_execute
706 }
707};
708
709static struct genl_family dp_packet_genl_family __ro_after_init = {
710 .hdrsize = sizeof(struct ovs_header),
711 .name = OVS_PACKET_FAMILY,
712 .version = OVS_PACKET_VERSION,
713 .maxattr = OVS_PACKET_ATTR_MAX,
714 .policy = packet_policy,
715 .netnsok = true,
716 .parallel_ops = true,
717 .small_ops = dp_packet_genl_ops,
718 .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
719 .resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
720 .module = THIS_MODULE,
721};
722
723static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
724 struct ovs_dp_megaflow_stats *mega_stats)
725{
726 int i;
727
728 memset(mega_stats, 0, sizeof(*mega_stats));
729
730 stats->n_flows = ovs_flow_tbl_count(&dp->table);
731 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
732
733 stats->n_hit = stats->n_missed = stats->n_lost = 0;
734
735 for_each_possible_cpu(i) {
736 const struct dp_stats_percpu *percpu_stats;
737 struct dp_stats_percpu local_stats;
738 unsigned int start;
739
740 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
741
742 do {
743 start = u64_stats_fetch_begin(&percpu_stats->syncp);
744 local_stats = *percpu_stats;
745 } while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
746
747 stats->n_hit += local_stats.n_hit;
748 stats->n_missed += local_stats.n_missed;
749 stats->n_lost += local_stats.n_lost;
750 mega_stats->n_mask_hit += local_stats.n_mask_hit;
751 mega_stats->n_cache_hit += local_stats.n_cache_hit;
752 }
753}
754
755static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
756{
757 return ovs_identifier_is_ufid(sfid) &&
758 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
759}
760
761static bool should_fill_mask(uint32_t ufid_flags)
762{
763 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
764}
765
766static bool should_fill_actions(uint32_t ufid_flags)
767{
768 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
769}
770
771static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
772 const struct sw_flow_id *sfid,
773 uint32_t ufid_flags)
774{
775 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
776
777 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
778 * see ovs_nla_put_identifier()
779 */
780 if (sfid && ovs_identifier_is_ufid(sfid))
781 len += nla_total_size(sfid->ufid_len);
782 else
783 len += nla_total_size(ovs_key_attr_size());
784
785 /* OVS_FLOW_ATTR_KEY */
786 if (!sfid || should_fill_key(sfid, ufid_flags))
787 len += nla_total_size(ovs_key_attr_size());
788
789 /* OVS_FLOW_ATTR_MASK */
790 if (should_fill_mask(ufid_flags))
791 len += nla_total_size(ovs_key_attr_size());
792
793 /* OVS_FLOW_ATTR_ACTIONS */
794 if (should_fill_actions(ufid_flags))
795 len += nla_total_size(acts->orig_len);
796
797 return len
798 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
799 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
800 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
801}
802
803/* Called with ovs_mutex or RCU read lock. */
804static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
805 struct sk_buff *skb)
806{
807 struct ovs_flow_stats stats;
808 __be16 tcp_flags;
809 unsigned long used;
810
811 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
812
813 if (used &&
814 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
815 OVS_FLOW_ATTR_PAD))
816 return -EMSGSIZE;
817
818 if (stats.n_packets &&
819 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
820 sizeof(struct ovs_flow_stats), &stats,
821 OVS_FLOW_ATTR_PAD))
822 return -EMSGSIZE;
823
824 if ((u8)ntohs(tcp_flags) &&
825 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
826 return -EMSGSIZE;
827
828 return 0;
829}
830
831/* Called with ovs_mutex or RCU read lock. */
832static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
833 struct sk_buff *skb, int skb_orig_len)
834{
835 struct nlattr *start;
836 int err;
837
838 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
839 * this is the first flow to be dumped into 'skb'. This is unusual for
840 * Netlink but individual action lists can be longer than
841 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
842 * The userspace caller can always fetch the actions separately if it
843 * really wants them. (Most userspace callers in fact don't care.)
844 *
845 * This can only fail for dump operations because the skb is always
846 * properly sized for single flows.
847 */
848 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
849 if (start) {
850 const struct sw_flow_actions *sf_acts;
851
852 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
853 err = ovs_nla_put_actions(sf_acts->actions,
854 sf_acts->actions_len, skb);
855
856 if (!err)
857 nla_nest_end(skb, start);
858 else {
859 if (skb_orig_len)
860 return err;
861
862 nla_nest_cancel(skb, start);
863 }
864 } else if (skb_orig_len) {
865 return -EMSGSIZE;
866 }
867
868 return 0;
869}
870
871/* Called with ovs_mutex or RCU read lock. */
872static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
873 struct sk_buff *skb, u32 portid,
874 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
875{
876 const int skb_orig_len = skb->len;
877 struct ovs_header *ovs_header;
878 int err;
879
880 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
881 flags, cmd);
882 if (!ovs_header)
883 return -EMSGSIZE;
884
885 ovs_header->dp_ifindex = dp_ifindex;
886
887 err = ovs_nla_put_identifier(flow, skb);
888 if (err)
889 goto error;
890
891 if (should_fill_key(&flow->id, ufid_flags)) {
892 err = ovs_nla_put_masked_key(flow, skb);
893 if (err)
894 goto error;
895 }
896
897 if (should_fill_mask(ufid_flags)) {
898 err = ovs_nla_put_mask(flow, skb);
899 if (err)
900 goto error;
901 }
902
903 err = ovs_flow_cmd_fill_stats(flow, skb);
904 if (err)
905 goto error;
906
907 if (should_fill_actions(ufid_flags)) {
908 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
909 if (err)
910 goto error;
911 }
912
913 genlmsg_end(skb, ovs_header);
914 return 0;
915
916error:
917 genlmsg_cancel(skb, ovs_header);
918 return err;
919}
920
921/* May not be called with RCU read lock. */
922static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
923 const struct sw_flow_id *sfid,
924 struct genl_info *info,
925 bool always,
926 uint32_t ufid_flags)
927{
928 struct sk_buff *skb;
929 size_t len;
930
931 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
932 return NULL;
933
934 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
935 skb = genlmsg_new(len, GFP_KERNEL);
936 if (!skb)
937 return ERR_PTR(-ENOMEM);
938
939 return skb;
940}
941
942/* Called with ovs_mutex. */
943static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
944 int dp_ifindex,
945 struct genl_info *info, u8 cmd,
946 bool always, u32 ufid_flags)
947{
948 struct sk_buff *skb;
949 int retval;
950
951 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
952 &flow->id, info, always, ufid_flags);
953 if (IS_ERR_OR_NULL(skb))
954 return skb;
955
956 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
957 info->snd_portid, info->snd_seq, 0,
958 cmd, ufid_flags);
959 if (WARN_ON_ONCE(retval < 0)) {
960 kfree_skb(skb);
961 skb = ERR_PTR(retval);
962 }
963 return skb;
964}
965
966static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
967{
968 struct net *net = sock_net(skb->sk);
969 struct nlattr **a = info->attrs;
970 struct ovs_header *ovs_header = genl_info_userhdr(info);
971 struct sw_flow *flow = NULL, *new_flow;
972 struct sw_flow_mask mask;
973 struct sk_buff *reply;
974 struct datapath *dp;
975 struct sw_flow_key *key;
976 struct sw_flow_actions *acts;
977 struct sw_flow_match match;
978 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
979 int error;
980 bool log = !a[OVS_FLOW_ATTR_PROBE];
981
982 /* Must have key and actions. */
983 error = -EINVAL;
984 if (!a[OVS_FLOW_ATTR_KEY]) {
985 OVS_NLERR(log, "Flow key attr not present in new flow.");
986 goto error;
987 }
988 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
989 OVS_NLERR(log, "Flow actions attr not present in new flow.");
990 goto error;
991 }
992
993 /* Most of the time we need to allocate a new flow, do it before
994 * locking.
995 */
996 new_flow = ovs_flow_alloc();
997 if (IS_ERR(new_flow)) {
998 error = PTR_ERR(new_flow);
999 goto error;
1000 }
1001
1002 /* Extract key. */
1003 key = kzalloc(sizeof(*key), GFP_KERNEL);
1004 if (!key) {
1005 error = -ENOMEM;
1006 goto err_kfree_flow;
1007 }
1008
1009 ovs_match_init(&match, key, false, &mask);
1010 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1011 a[OVS_FLOW_ATTR_MASK], log);
1012 if (error)
1013 goto err_kfree_key;
1014
1015 ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1016
1017 /* Extract flow identifier. */
1018 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
1019 key, log);
1020 if (error)
1021 goto err_kfree_key;
1022
1023 /* Validate actions. */
1024 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1025 &new_flow->key, &acts, log);
1026 if (error) {
1027 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1028 goto err_kfree_key;
1029 }
1030
1031 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1032 ufid_flags);
1033 if (IS_ERR(reply)) {
1034 error = PTR_ERR(reply);
1035 goto err_kfree_acts;
1036 }
1037
1038 ovs_lock();
1039 dp = get_dp(net, ovs_header->dp_ifindex);
1040 if (unlikely(!dp)) {
1041 error = -ENODEV;
1042 goto err_unlock_ovs;
1043 }
1044
1045 /* Check if this is a duplicate flow */
1046 if (ovs_identifier_is_ufid(&new_flow->id))
1047 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1048 if (!flow)
1049 flow = ovs_flow_tbl_lookup(&dp->table, key);
1050 if (likely(!flow)) {
1051 rcu_assign_pointer(new_flow->sf_acts, acts);
1052
1053 /* Put flow in bucket. */
1054 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1055 if (unlikely(error)) {
1056 acts = NULL;
1057 goto err_unlock_ovs;
1058 }
1059
1060 if (unlikely(reply)) {
1061 error = ovs_flow_cmd_fill_info(new_flow,
1062 ovs_header->dp_ifindex,
1063 reply, info->snd_portid,
1064 info->snd_seq, 0,
1065 OVS_FLOW_CMD_NEW,
1066 ufid_flags);
1067 BUG_ON(error < 0);
1068 }
1069 ovs_unlock();
1070 } else {
1071 struct sw_flow_actions *old_acts;
1072
1073 /* Bail out if we're not allowed to modify an existing flow.
1074 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1075 * because Generic Netlink treats the latter as a dump
1076 * request. We also accept NLM_F_EXCL in case that bug ever
1077 * gets fixed.
1078 */
1079 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1080 | NLM_F_EXCL))) {
1081 error = -EEXIST;
1082 goto err_unlock_ovs;
1083 }
1084 /* The flow identifier has to be the same for flow updates.
1085 * Look for any overlapping flow.
1086 */
1087 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1088 if (ovs_identifier_is_key(&flow->id))
1089 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1090 &match);
1091 else /* UFID matches but key is different */
1092 flow = NULL;
1093 if (!flow) {
1094 error = -ENOENT;
1095 goto err_unlock_ovs;
1096 }
1097 }
1098 /* Update actions. */
1099 old_acts = ovsl_dereference(flow->sf_acts);
1100 rcu_assign_pointer(flow->sf_acts, acts);
1101
1102 if (unlikely(reply)) {
1103 error = ovs_flow_cmd_fill_info(flow,
1104 ovs_header->dp_ifindex,
1105 reply, info->snd_portid,
1106 info->snd_seq, 0,
1107 OVS_FLOW_CMD_NEW,
1108 ufid_flags);
1109 BUG_ON(error < 0);
1110 }
1111 ovs_unlock();
1112
1113 ovs_nla_free_flow_actions_rcu(old_acts);
1114 ovs_flow_free(new_flow, false);
1115 }
1116
1117 if (reply)
1118 ovs_notify(&dp_flow_genl_family, reply, info);
1119
1120 kfree(key);
1121 return 0;
1122
1123err_unlock_ovs:
1124 ovs_unlock();
1125 kfree_skb(reply);
1126err_kfree_acts:
1127 ovs_nla_free_flow_actions(acts);
1128err_kfree_key:
1129 kfree(key);
1130err_kfree_flow:
1131 ovs_flow_free(new_flow, false);
1132error:
1133 return error;
1134}
1135
1136/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1137static noinline_for_stack
1138struct sw_flow_actions *get_flow_actions(struct net *net,
1139 const struct nlattr *a,
1140 const struct sw_flow_key *key,
1141 const struct sw_flow_mask *mask,
1142 bool log)
1143{
1144 struct sw_flow_actions *acts;
1145 struct sw_flow_key masked_key;
1146 int error;
1147
1148 ovs_flow_mask_key(&masked_key, key, true, mask);
1149 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1150 if (error) {
1151 OVS_NLERR(log,
1152 "Actions may not be safe on all matching packets");
1153 return ERR_PTR(error);
1154 }
1155
1156 return acts;
1157}
1158
1159/* Factor out match-init and action-copy to avoid
1160 * "Wframe-larger-than=1024" warning. Because mask is only
1161 * used to get actions, we new a function to save some
1162 * stack space.
1163 *
1164 * If there are not key and action attrs, we return 0
1165 * directly. In the case, the caller will also not use the
1166 * match as before. If there is action attr, we try to get
1167 * actions and save them to *acts. Before returning from
1168 * the function, we reset the match->mask pointer. Because
1169 * we should not to return match object with dangling reference
1170 * to mask.
1171 * */
1172static noinline_for_stack int
1173ovs_nla_init_match_and_action(struct net *net,
1174 struct sw_flow_match *match,
1175 struct sw_flow_key *key,
1176 struct nlattr **a,
1177 struct sw_flow_actions **acts,
1178 bool log)
1179{
1180 struct sw_flow_mask mask;
1181 int error = 0;
1182
1183 if (a[OVS_FLOW_ATTR_KEY]) {
1184 ovs_match_init(match, key, true, &mask);
1185 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1186 a[OVS_FLOW_ATTR_MASK], log);
1187 if (error)
1188 goto error;
1189 }
1190
1191 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1192 if (!a[OVS_FLOW_ATTR_KEY]) {
1193 OVS_NLERR(log,
1194 "Flow key attribute not present in set flow.");
1195 error = -EINVAL;
1196 goto error;
1197 }
1198
1199 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1200 &mask, log);
1201 if (IS_ERR(*acts)) {
1202 error = PTR_ERR(*acts);
1203 goto error;
1204 }
1205 }
1206
1207 /* On success, error is 0. */
1208error:
1209 match->mask = NULL;
1210 return error;
1211}
1212
1213static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1214{
1215 struct net *net = sock_net(skb->sk);
1216 struct nlattr **a = info->attrs;
1217 struct ovs_header *ovs_header = genl_info_userhdr(info);
1218 struct sw_flow_key key;
1219 struct sw_flow *flow;
1220 struct sk_buff *reply = NULL;
1221 struct datapath *dp;
1222 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1223 struct sw_flow_match match;
1224 struct sw_flow_id sfid;
1225 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1226 int error = 0;
1227 bool log = !a[OVS_FLOW_ATTR_PROBE];
1228 bool ufid_present;
1229
1230 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1231 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1232 OVS_NLERR(log,
1233 "Flow set message rejected, Key attribute missing.");
1234 return -EINVAL;
1235 }
1236
1237 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1238 &acts, log);
1239 if (error)
1240 goto error;
1241
1242 if (acts) {
1243 /* Can allocate before locking if have acts. */
1244 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1245 ufid_flags);
1246 if (IS_ERR(reply)) {
1247 error = PTR_ERR(reply);
1248 goto err_kfree_acts;
1249 }
1250 }
1251
1252 ovs_lock();
1253 dp = get_dp(net, ovs_header->dp_ifindex);
1254 if (unlikely(!dp)) {
1255 error = -ENODEV;
1256 goto err_unlock_ovs;
1257 }
1258 /* Check that the flow exists. */
1259 if (ufid_present)
1260 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1261 else
1262 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1263 if (unlikely(!flow)) {
1264 error = -ENOENT;
1265 goto err_unlock_ovs;
1266 }
1267
1268 /* Update actions, if present. */
1269 if (likely(acts)) {
1270 old_acts = ovsl_dereference(flow->sf_acts);
1271 rcu_assign_pointer(flow->sf_acts, acts);
1272
1273 if (unlikely(reply)) {
1274 error = ovs_flow_cmd_fill_info(flow,
1275 ovs_header->dp_ifindex,
1276 reply, info->snd_portid,
1277 info->snd_seq, 0,
1278 OVS_FLOW_CMD_SET,
1279 ufid_flags);
1280 BUG_ON(error < 0);
1281 }
1282 } else {
1283 /* Could not alloc without acts before locking. */
1284 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1285 info, OVS_FLOW_CMD_SET, false,
1286 ufid_flags);
1287
1288 if (IS_ERR(reply)) {
1289 error = PTR_ERR(reply);
1290 goto err_unlock_ovs;
1291 }
1292 }
1293
1294 /* Clear stats. */
1295 if (a[OVS_FLOW_ATTR_CLEAR])
1296 ovs_flow_stats_clear(flow);
1297 ovs_unlock();
1298
1299 if (reply)
1300 ovs_notify(&dp_flow_genl_family, reply, info);
1301 if (old_acts)
1302 ovs_nla_free_flow_actions_rcu(old_acts);
1303
1304 return 0;
1305
1306err_unlock_ovs:
1307 ovs_unlock();
1308 kfree_skb(reply);
1309err_kfree_acts:
1310 ovs_nla_free_flow_actions(acts);
1311error:
1312 return error;
1313}
1314
1315static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1316{
1317 struct nlattr **a = info->attrs;
1318 struct ovs_header *ovs_header = genl_info_userhdr(info);
1319 struct net *net = sock_net(skb->sk);
1320 struct sw_flow_key key;
1321 struct sk_buff *reply;
1322 struct sw_flow *flow;
1323 struct datapath *dp;
1324 struct sw_flow_match match;
1325 struct sw_flow_id ufid;
1326 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1327 int err = 0;
1328 bool log = !a[OVS_FLOW_ATTR_PROBE];
1329 bool ufid_present;
1330
1331 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1332 if (a[OVS_FLOW_ATTR_KEY]) {
1333 ovs_match_init(&match, &key, true, NULL);
1334 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1335 log);
1336 } else if (!ufid_present) {
1337 OVS_NLERR(log,
1338 "Flow get message rejected, Key attribute missing.");
1339 err = -EINVAL;
1340 }
1341 if (err)
1342 return err;
1343
1344 ovs_lock();
1345 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1346 if (!dp) {
1347 err = -ENODEV;
1348 goto unlock;
1349 }
1350
1351 if (ufid_present)
1352 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1353 else
1354 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1355 if (!flow) {
1356 err = -ENOENT;
1357 goto unlock;
1358 }
1359
1360 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1361 OVS_FLOW_CMD_GET, true, ufid_flags);
1362 if (IS_ERR(reply)) {
1363 err = PTR_ERR(reply);
1364 goto unlock;
1365 }
1366
1367 ovs_unlock();
1368 return genlmsg_reply(reply, info);
1369unlock:
1370 ovs_unlock();
1371 return err;
1372}
1373
1374static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1375{
1376 struct nlattr **a = info->attrs;
1377 struct ovs_header *ovs_header = genl_info_userhdr(info);
1378 struct net *net = sock_net(skb->sk);
1379 struct sw_flow_key key;
1380 struct sk_buff *reply;
1381 struct sw_flow *flow = NULL;
1382 struct datapath *dp;
1383 struct sw_flow_match match;
1384 struct sw_flow_id ufid;
1385 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1386 int err;
1387 bool log = !a[OVS_FLOW_ATTR_PROBE];
1388 bool ufid_present;
1389
1390 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1391 if (a[OVS_FLOW_ATTR_KEY]) {
1392 ovs_match_init(&match, &key, true, NULL);
1393 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1394 NULL, log);
1395 if (unlikely(err))
1396 return err;
1397 }
1398
1399 ovs_lock();
1400 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1401 if (unlikely(!dp)) {
1402 err = -ENODEV;
1403 goto unlock;
1404 }
1405
1406 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1407 err = ovs_flow_tbl_flush(&dp->table);
1408 goto unlock;
1409 }
1410
1411 if (ufid_present)
1412 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1413 else
1414 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1415 if (unlikely(!flow)) {
1416 err = -ENOENT;
1417 goto unlock;
1418 }
1419
1420 ovs_flow_tbl_remove(&dp->table, flow);
1421 ovs_unlock();
1422
1423 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1424 &flow->id, info, false, ufid_flags);
1425 if (likely(reply)) {
1426 if (!IS_ERR(reply)) {
1427 rcu_read_lock(); /*To keep RCU checker happy. */
1428 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1429 reply, info->snd_portid,
1430 info->snd_seq, 0,
1431 OVS_FLOW_CMD_DEL,
1432 ufid_flags);
1433 rcu_read_unlock();
1434 if (WARN_ON_ONCE(err < 0)) {
1435 kfree_skb(reply);
1436 goto out_free;
1437 }
1438
1439 ovs_notify(&dp_flow_genl_family, reply, info);
1440 } else {
1441 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1442 PTR_ERR(reply));
1443 }
1444 }
1445
1446out_free:
1447 ovs_flow_free(flow, true);
1448 return 0;
1449unlock:
1450 ovs_unlock();
1451 return err;
1452}
1453
1454static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1455{
1456 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1457 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1458 struct table_instance *ti;
1459 struct datapath *dp;
1460 u32 ufid_flags;
1461 int err;
1462
1463 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1464 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1465 if (err)
1466 return err;
1467 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1468
1469 rcu_read_lock();
1470 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1471 if (!dp) {
1472 rcu_read_unlock();
1473 return -ENODEV;
1474 }
1475
1476 ti = rcu_dereference(dp->table.ti);
1477 for (;;) {
1478 struct sw_flow *flow;
1479 u32 bucket, obj;
1480
1481 bucket = cb->args[0];
1482 obj = cb->args[1];
1483 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1484 if (!flow)
1485 break;
1486
1487 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1488 NETLINK_CB(cb->skb).portid,
1489 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1490 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1491 break;
1492
1493 cb->args[0] = bucket;
1494 cb->args[1] = obj;
1495 }
1496 rcu_read_unlock();
1497 return skb->len;
1498}
1499
1500static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1501 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1502 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1503 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1504 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1505 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1506 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1507 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1508};
1509
1510static const struct genl_small_ops dp_flow_genl_ops[] = {
1511 { .cmd = OVS_FLOW_CMD_NEW,
1512 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1513 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1514 .doit = ovs_flow_cmd_new
1515 },
1516 { .cmd = OVS_FLOW_CMD_DEL,
1517 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1518 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1519 .doit = ovs_flow_cmd_del
1520 },
1521 { .cmd = OVS_FLOW_CMD_GET,
1522 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1523 .flags = 0, /* OK for unprivileged users. */
1524 .doit = ovs_flow_cmd_get,
1525 .dumpit = ovs_flow_cmd_dump
1526 },
1527 { .cmd = OVS_FLOW_CMD_SET,
1528 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1529 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1530 .doit = ovs_flow_cmd_set,
1531 },
1532};
1533
1534static struct genl_family dp_flow_genl_family __ro_after_init = {
1535 .hdrsize = sizeof(struct ovs_header),
1536 .name = OVS_FLOW_FAMILY,
1537 .version = OVS_FLOW_VERSION,
1538 .maxattr = OVS_FLOW_ATTR_MAX,
1539 .policy = flow_policy,
1540 .netnsok = true,
1541 .parallel_ops = true,
1542 .small_ops = dp_flow_genl_ops,
1543 .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1544 .resv_start_op = OVS_FLOW_CMD_SET + 1,
1545 .mcgrps = &ovs_dp_flow_multicast_group,
1546 .n_mcgrps = 1,
1547 .module = THIS_MODULE,
1548};
1549
1550static size_t ovs_dp_cmd_msg_size(void)
1551{
1552 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1553
1554 msgsize += nla_total_size(IFNAMSIZ);
1555 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1556 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1557 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1558 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1559 msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1560
1561 return msgsize;
1562}
1563
1564/* Called with ovs_mutex. */
1565static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1566 u32 portid, u32 seq, u32 flags, u8 cmd)
1567{
1568 struct ovs_header *ovs_header;
1569 struct ovs_dp_stats dp_stats;
1570 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1571 struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1572 int err, pids_len;
1573
1574 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1575 flags, cmd);
1576 if (!ovs_header)
1577 goto error;
1578
1579 ovs_header->dp_ifindex = get_dpifindex(dp);
1580
1581 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1582 if (err)
1583 goto nla_put_failure;
1584
1585 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1586 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1587 &dp_stats, OVS_DP_ATTR_PAD))
1588 goto nla_put_failure;
1589
1590 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1591 sizeof(struct ovs_dp_megaflow_stats),
1592 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1593 goto nla_put_failure;
1594
1595 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1596 goto nla_put_failure;
1597
1598 if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1599 ovs_flow_tbl_masks_cache_size(&dp->table)))
1600 goto nla_put_failure;
1601
1602 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1603 pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1604 if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1605 goto nla_put_failure;
1606 }
1607
1608 genlmsg_end(skb, ovs_header);
1609 return 0;
1610
1611nla_put_failure:
1612 genlmsg_cancel(skb, ovs_header);
1613error:
1614 return -EMSGSIZE;
1615}
1616
1617static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1618{
1619 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1620}
1621
1622/* Called with rcu_read_lock or ovs_mutex. */
1623static struct datapath *lookup_datapath(struct net *net,
1624 const struct ovs_header *ovs_header,
1625 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1626{
1627 struct datapath *dp;
1628
1629 if (!a[OVS_DP_ATTR_NAME])
1630 dp = get_dp(net, ovs_header->dp_ifindex);
1631 else {
1632 struct vport *vport;
1633
1634 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1635 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1636 }
1637 return dp ? dp : ERR_PTR(-ENODEV);
1638}
1639
1640static void ovs_dp_reset_user_features(struct sk_buff *skb,
1641 struct genl_info *info)
1642{
1643 struct datapath *dp;
1644
1645 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1646 info->attrs);
1647 if (IS_ERR(dp))
1648 return;
1649
1650 pr_warn("%s: Dropping previously announced user features\n",
1651 ovs_dp_name(dp));
1652 dp->user_features = 0;
1653}
1654
1655static int ovs_dp_set_upcall_portids(struct datapath *dp,
1656 const struct nlattr *ids)
1657{
1658 struct dp_nlsk_pids *old, *dp_nlsk_pids;
1659
1660 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1661 return -EINVAL;
1662
1663 old = ovsl_dereference(dp->upcall_portids);
1664
1665 dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1666 GFP_KERNEL);
1667 if (!dp_nlsk_pids)
1668 return -ENOMEM;
1669
1670 dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1671 nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1672
1673 rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1674
1675 kfree_rcu(old, rcu);
1676
1677 return 0;
1678}
1679
1680u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1681{
1682 struct dp_nlsk_pids *dp_nlsk_pids;
1683
1684 dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1685
1686 if (dp_nlsk_pids) {
1687 if (cpu_id < dp_nlsk_pids->n_pids) {
1688 return dp_nlsk_pids->pids[cpu_id];
1689 } else if (dp_nlsk_pids->n_pids > 0 &&
1690 cpu_id >= dp_nlsk_pids->n_pids) {
1691 /* If the number of netlink PIDs is mismatched with
1692 * the number of CPUs as seen by the kernel, log this
1693 * and send the upcall to an arbitrary socket (0) in
1694 * order to not drop packets
1695 */
1696 pr_info_ratelimited("cpu_id mismatch with handler threads");
1697 return dp_nlsk_pids->pids[cpu_id %
1698 dp_nlsk_pids->n_pids];
1699 } else {
1700 return 0;
1701 }
1702 } else {
1703 return 0;
1704 }
1705}
1706
1707static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1708{
1709 u32 user_features = 0, old_features = dp->user_features;
1710 int err;
1711
1712 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1713 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1714
1715 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1716 OVS_DP_F_UNALIGNED |
1717 OVS_DP_F_TC_RECIRC_SHARING |
1718 OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1719 return -EOPNOTSUPP;
1720
1721#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1722 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1723 return -EOPNOTSUPP;
1724#endif
1725 }
1726
1727 if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1728 int err;
1729 u32 cache_size;
1730
1731 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1732 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1733 if (err)
1734 return err;
1735 }
1736
1737 dp->user_features = user_features;
1738
1739 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1740 a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1741 /* Upcall Netlink Port IDs have been updated */
1742 err = ovs_dp_set_upcall_portids(dp,
1743 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1744 if (err)
1745 return err;
1746 }
1747
1748 if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1749 !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1750 tc_skb_ext_tc_enable();
1751 else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1752 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1753 tc_skb_ext_tc_disable();
1754
1755 return 0;
1756}
1757
1758static int ovs_dp_stats_init(struct datapath *dp)
1759{
1760 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1761 if (!dp->stats_percpu)
1762 return -ENOMEM;
1763
1764 return 0;
1765}
1766
1767static int ovs_dp_vport_init(struct datapath *dp)
1768{
1769 int i;
1770
1771 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1772 sizeof(struct hlist_head),
1773 GFP_KERNEL);
1774 if (!dp->ports)
1775 return -ENOMEM;
1776
1777 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1778 INIT_HLIST_HEAD(&dp->ports[i]);
1779
1780 return 0;
1781}
1782
1783static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1784{
1785 struct nlattr **a = info->attrs;
1786 struct vport_parms parms;
1787 struct sk_buff *reply;
1788 struct datapath *dp;
1789 struct vport *vport;
1790 struct ovs_net *ovs_net;
1791 int err;
1792
1793 err = -EINVAL;
1794 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1795 goto err;
1796
1797 reply = ovs_dp_cmd_alloc_info();
1798 if (!reply)
1799 return -ENOMEM;
1800
1801 err = -ENOMEM;
1802 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1803 if (dp == NULL)
1804 goto err_destroy_reply;
1805
1806 ovs_dp_set_net(dp, sock_net(skb->sk));
1807
1808 /* Allocate table. */
1809 err = ovs_flow_tbl_init(&dp->table);
1810 if (err)
1811 goto err_destroy_dp;
1812
1813 err = ovs_dp_stats_init(dp);
1814 if (err)
1815 goto err_destroy_table;
1816
1817 err = ovs_dp_vport_init(dp);
1818 if (err)
1819 goto err_destroy_stats;
1820
1821 err = ovs_meters_init(dp);
1822 if (err)
1823 goto err_destroy_ports;
1824
1825 /* Set up our datapath device. */
1826 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1827 parms.type = OVS_VPORT_TYPE_INTERNAL;
1828 parms.options = NULL;
1829 parms.dp = dp;
1830 parms.port_no = OVSP_LOCAL;
1831 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1832 parms.desired_ifindex = a[OVS_DP_ATTR_IFINDEX]
1833 ? nla_get_s32(a[OVS_DP_ATTR_IFINDEX]) : 0;
1834
1835 /* So far only local changes have been made, now need the lock. */
1836 ovs_lock();
1837
1838 err = ovs_dp_change(dp, a);
1839 if (err)
1840 goto err_unlock_and_destroy_meters;
1841
1842 vport = new_vport(&parms);
1843 if (IS_ERR(vport)) {
1844 err = PTR_ERR(vport);
1845 if (err == -EBUSY)
1846 err = -EEXIST;
1847
1848 if (err == -EEXIST) {
1849 /* An outdated user space instance that does not understand
1850 * the concept of user_features has attempted to create a new
1851 * datapath and is likely to reuse it. Drop all user features.
1852 */
1853 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1854 ovs_dp_reset_user_features(skb, info);
1855 }
1856
1857 goto err_destroy_portids;
1858 }
1859
1860 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1861 info->snd_seq, 0, OVS_DP_CMD_NEW);
1862 BUG_ON(err < 0);
1863
1864 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1865 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1866
1867 ovs_unlock();
1868
1869 ovs_notify(&dp_datapath_genl_family, reply, info);
1870 return 0;
1871
1872err_destroy_portids:
1873 kfree(rcu_dereference_raw(dp->upcall_portids));
1874err_unlock_and_destroy_meters:
1875 ovs_unlock();
1876 ovs_meters_exit(dp);
1877err_destroy_ports:
1878 kfree(dp->ports);
1879err_destroy_stats:
1880 free_percpu(dp->stats_percpu);
1881err_destroy_table:
1882 ovs_flow_tbl_destroy(&dp->table);
1883err_destroy_dp:
1884 kfree(dp);
1885err_destroy_reply:
1886 kfree_skb(reply);
1887err:
1888 return err;
1889}
1890
1891/* Called with ovs_mutex. */
1892static void __dp_destroy(struct datapath *dp)
1893{
1894 struct flow_table *table = &dp->table;
1895 int i;
1896
1897 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1898 tc_skb_ext_tc_disable();
1899
1900 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1901 struct vport *vport;
1902 struct hlist_node *n;
1903
1904 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1905 if (vport->port_no != OVSP_LOCAL)
1906 ovs_dp_detach_port(vport);
1907 }
1908
1909 list_del_rcu(&dp->list_node);
1910
1911 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1912 * all ports in datapath are destroyed first before freeing datapath.
1913 */
1914 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1915
1916 /* Flush sw_flow in the tables. RCU cb only releases resource
1917 * such as dp, ports and tables. That may avoid some issues
1918 * such as RCU usage warning.
1919 */
1920 table_instance_flow_flush(table, ovsl_dereference(table->ti),
1921 ovsl_dereference(table->ufid_ti));
1922
1923 /* RCU destroy the ports, meters and flow tables. */
1924 call_rcu(&dp->rcu, destroy_dp_rcu);
1925}
1926
1927static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1928{
1929 struct sk_buff *reply;
1930 struct datapath *dp;
1931 int err;
1932
1933 reply = ovs_dp_cmd_alloc_info();
1934 if (!reply)
1935 return -ENOMEM;
1936
1937 ovs_lock();
1938 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1939 info->attrs);
1940 err = PTR_ERR(dp);
1941 if (IS_ERR(dp))
1942 goto err_unlock_free;
1943
1944 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1945 info->snd_seq, 0, OVS_DP_CMD_DEL);
1946 BUG_ON(err < 0);
1947
1948 __dp_destroy(dp);
1949 ovs_unlock();
1950
1951 ovs_notify(&dp_datapath_genl_family, reply, info);
1952
1953 return 0;
1954
1955err_unlock_free:
1956 ovs_unlock();
1957 kfree_skb(reply);
1958 return err;
1959}
1960
1961static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1962{
1963 struct sk_buff *reply;
1964 struct datapath *dp;
1965 int err;
1966
1967 reply = ovs_dp_cmd_alloc_info();
1968 if (!reply)
1969 return -ENOMEM;
1970
1971 ovs_lock();
1972 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1973 info->attrs);
1974 err = PTR_ERR(dp);
1975 if (IS_ERR(dp))
1976 goto err_unlock_free;
1977
1978 err = ovs_dp_change(dp, info->attrs);
1979 if (err)
1980 goto err_unlock_free;
1981
1982 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1983 info->snd_seq, 0, OVS_DP_CMD_SET);
1984 BUG_ON(err < 0);
1985
1986 ovs_unlock();
1987 ovs_notify(&dp_datapath_genl_family, reply, info);
1988
1989 return 0;
1990
1991err_unlock_free:
1992 ovs_unlock();
1993 kfree_skb(reply);
1994 return err;
1995}
1996
1997static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1998{
1999 struct sk_buff *reply;
2000 struct datapath *dp;
2001 int err;
2002
2003 reply = ovs_dp_cmd_alloc_info();
2004 if (!reply)
2005 return -ENOMEM;
2006
2007 ovs_lock();
2008 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2009 info->attrs);
2010 if (IS_ERR(dp)) {
2011 err = PTR_ERR(dp);
2012 goto err_unlock_free;
2013 }
2014 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2015 info->snd_seq, 0, OVS_DP_CMD_GET);
2016 BUG_ON(err < 0);
2017 ovs_unlock();
2018
2019 return genlmsg_reply(reply, info);
2020
2021err_unlock_free:
2022 ovs_unlock();
2023 kfree_skb(reply);
2024 return err;
2025}
2026
2027static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2028{
2029 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2030 struct datapath *dp;
2031 int skip = cb->args[0];
2032 int i = 0;
2033
2034 ovs_lock();
2035 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2036 if (i >= skip &&
2037 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2038 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2039 OVS_DP_CMD_GET) < 0)
2040 break;
2041 i++;
2042 }
2043 ovs_unlock();
2044
2045 cb->args[0] = i;
2046
2047 return skb->len;
2048}
2049
2050static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2051 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2052 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2053 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2054 [OVS_DP_ATTR_MASKS_CACHE_SIZE] = NLA_POLICY_RANGE(NLA_U32, 0,
2055 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2056 [OVS_DP_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2057};
2058
2059static const struct genl_small_ops dp_datapath_genl_ops[] = {
2060 { .cmd = OVS_DP_CMD_NEW,
2061 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2062 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2063 .doit = ovs_dp_cmd_new
2064 },
2065 { .cmd = OVS_DP_CMD_DEL,
2066 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2067 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2068 .doit = ovs_dp_cmd_del
2069 },
2070 { .cmd = OVS_DP_CMD_GET,
2071 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2072 .flags = 0, /* OK for unprivileged users. */
2073 .doit = ovs_dp_cmd_get,
2074 .dumpit = ovs_dp_cmd_dump
2075 },
2076 { .cmd = OVS_DP_CMD_SET,
2077 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2078 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2079 .doit = ovs_dp_cmd_set,
2080 },
2081};
2082
2083static struct genl_family dp_datapath_genl_family __ro_after_init = {
2084 .hdrsize = sizeof(struct ovs_header),
2085 .name = OVS_DATAPATH_FAMILY,
2086 .version = OVS_DATAPATH_VERSION,
2087 .maxattr = OVS_DP_ATTR_MAX,
2088 .policy = datapath_policy,
2089 .netnsok = true,
2090 .parallel_ops = true,
2091 .small_ops = dp_datapath_genl_ops,
2092 .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2093 .resv_start_op = OVS_DP_CMD_SET + 1,
2094 .mcgrps = &ovs_dp_datapath_multicast_group,
2095 .n_mcgrps = 1,
2096 .module = THIS_MODULE,
2097};
2098
2099/* Called with ovs_mutex or RCU read lock. */
2100static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2101 struct net *net, u32 portid, u32 seq,
2102 u32 flags, u8 cmd, gfp_t gfp)
2103{
2104 struct ovs_header *ovs_header;
2105 struct ovs_vport_stats vport_stats;
2106 int err;
2107
2108 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2109 flags, cmd);
2110 if (!ovs_header)
2111 return -EMSGSIZE;
2112
2113 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2114
2115 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2116 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2117 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2118 ovs_vport_name(vport)) ||
2119 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2120 goto nla_put_failure;
2121
2122 if (!net_eq(net, dev_net(vport->dev))) {
2123 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2124
2125 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2126 goto nla_put_failure;
2127 }
2128
2129 ovs_vport_get_stats(vport, &vport_stats);
2130 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2131 sizeof(struct ovs_vport_stats), &vport_stats,
2132 OVS_VPORT_ATTR_PAD))
2133 goto nla_put_failure;
2134
2135 if (ovs_vport_get_upcall_stats(vport, skb))
2136 goto nla_put_failure;
2137
2138 if (ovs_vport_get_upcall_portids(vport, skb))
2139 goto nla_put_failure;
2140
2141 err = ovs_vport_get_options(vport, skb);
2142 if (err == -EMSGSIZE)
2143 goto error;
2144
2145 genlmsg_end(skb, ovs_header);
2146 return 0;
2147
2148nla_put_failure:
2149 err = -EMSGSIZE;
2150error:
2151 genlmsg_cancel(skb, ovs_header);
2152 return err;
2153}
2154
2155static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2156{
2157 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2158}
2159
2160/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2161struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2162 u32 portid, u32 seq, u8 cmd)
2163{
2164 struct sk_buff *skb;
2165 int retval;
2166
2167 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2168 if (!skb)
2169 return ERR_PTR(-ENOMEM);
2170
2171 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2172 GFP_KERNEL);
2173 BUG_ON(retval < 0);
2174
2175 return skb;
2176}
2177
2178/* Called with ovs_mutex or RCU read lock. */
2179static struct vport *lookup_vport(struct net *net,
2180 const struct ovs_header *ovs_header,
2181 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2182{
2183 struct datapath *dp;
2184 struct vport *vport;
2185
2186 if (a[OVS_VPORT_ATTR_IFINDEX])
2187 return ERR_PTR(-EOPNOTSUPP);
2188 if (a[OVS_VPORT_ATTR_NAME]) {
2189 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2190 if (!vport)
2191 return ERR_PTR(-ENODEV);
2192 if (ovs_header->dp_ifindex &&
2193 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2194 return ERR_PTR(-ENODEV);
2195 return vport;
2196 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2197 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2198
2199 if (port_no >= DP_MAX_PORTS)
2200 return ERR_PTR(-EFBIG);
2201
2202 dp = get_dp(net, ovs_header->dp_ifindex);
2203 if (!dp)
2204 return ERR_PTR(-ENODEV);
2205
2206 vport = ovs_vport_ovsl_rcu(dp, port_no);
2207 if (!vport)
2208 return ERR_PTR(-ENODEV);
2209 return vport;
2210 } else
2211 return ERR_PTR(-EINVAL);
2212
2213}
2214
2215static unsigned int ovs_get_max_headroom(struct datapath *dp)
2216{
2217 unsigned int dev_headroom, max_headroom = 0;
2218 struct net_device *dev;
2219 struct vport *vport;
2220 int i;
2221
2222 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2223 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2224 lockdep_ovsl_is_held()) {
2225 dev = vport->dev;
2226 dev_headroom = netdev_get_fwd_headroom(dev);
2227 if (dev_headroom > max_headroom)
2228 max_headroom = dev_headroom;
2229 }
2230 }
2231
2232 return max_headroom;
2233}
2234
2235/* Called with ovs_mutex */
2236static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2237{
2238 struct vport *vport;
2239 int i;
2240
2241 dp->max_headroom = new_headroom;
2242 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2243 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2244 lockdep_ovsl_is_held())
2245 netdev_set_rx_headroom(vport->dev, new_headroom);
2246 }
2247}
2248
2249static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2250{
2251 struct nlattr **a = info->attrs;
2252 struct ovs_header *ovs_header = genl_info_userhdr(info);
2253 struct vport_parms parms;
2254 struct sk_buff *reply;
2255 struct vport *vport;
2256 struct datapath *dp;
2257 unsigned int new_headroom;
2258 u32 port_no;
2259 int err;
2260
2261 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2262 !a[OVS_VPORT_ATTR_UPCALL_PID])
2263 return -EINVAL;
2264
2265 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2266
2267 if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2268 return -EOPNOTSUPP;
2269
2270 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2271 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2272 if (port_no >= DP_MAX_PORTS)
2273 return -EFBIG;
2274
2275 reply = ovs_vport_cmd_alloc_info();
2276 if (!reply)
2277 return -ENOMEM;
2278
2279 ovs_lock();
2280restart:
2281 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2282 err = -ENODEV;
2283 if (!dp)
2284 goto exit_unlock_free;
2285
2286 if (port_no) {
2287 vport = ovs_vport_ovsl(dp, port_no);
2288 err = -EBUSY;
2289 if (vport)
2290 goto exit_unlock_free;
2291 } else {
2292 for (port_no = 1; ; port_no++) {
2293 if (port_no >= DP_MAX_PORTS) {
2294 err = -EFBIG;
2295 goto exit_unlock_free;
2296 }
2297 vport = ovs_vport_ovsl(dp, port_no);
2298 if (!vport)
2299 break;
2300 }
2301 }
2302
2303 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2304 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2305 parms.dp = dp;
2306 parms.port_no = port_no;
2307 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2308 parms.desired_ifindex = a[OVS_VPORT_ATTR_IFINDEX]
2309 ? nla_get_s32(a[OVS_VPORT_ATTR_IFINDEX]) : 0;
2310
2311 vport = new_vport(&parms);
2312 err = PTR_ERR(vport);
2313 if (IS_ERR(vport)) {
2314 if (err == -EAGAIN)
2315 goto restart;
2316 goto exit_unlock_free;
2317 }
2318
2319 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2320 info->snd_portid, info->snd_seq, 0,
2321 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2322
2323 new_headroom = netdev_get_fwd_headroom(vport->dev);
2324
2325 if (new_headroom > dp->max_headroom)
2326 ovs_update_headroom(dp, new_headroom);
2327 else
2328 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2329
2330 BUG_ON(err < 0);
2331 ovs_unlock();
2332
2333 ovs_notify(&dp_vport_genl_family, reply, info);
2334 return 0;
2335
2336exit_unlock_free:
2337 ovs_unlock();
2338 kfree_skb(reply);
2339 return err;
2340}
2341
2342static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2343{
2344 struct nlattr **a = info->attrs;
2345 struct sk_buff *reply;
2346 struct vport *vport;
2347 int err;
2348
2349 reply = ovs_vport_cmd_alloc_info();
2350 if (!reply)
2351 return -ENOMEM;
2352
2353 ovs_lock();
2354 vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2355 err = PTR_ERR(vport);
2356 if (IS_ERR(vport))
2357 goto exit_unlock_free;
2358
2359 if (a[OVS_VPORT_ATTR_TYPE] &&
2360 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2361 err = -EINVAL;
2362 goto exit_unlock_free;
2363 }
2364
2365 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2366 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2367 if (err)
2368 goto exit_unlock_free;
2369 }
2370
2371
2372 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2373 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2374
2375 err = ovs_vport_set_upcall_portids(vport, ids);
2376 if (err)
2377 goto exit_unlock_free;
2378 }
2379
2380 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2381 info->snd_portid, info->snd_seq, 0,
2382 OVS_VPORT_CMD_SET, GFP_KERNEL);
2383 BUG_ON(err < 0);
2384
2385 ovs_unlock();
2386 ovs_notify(&dp_vport_genl_family, reply, info);
2387 return 0;
2388
2389exit_unlock_free:
2390 ovs_unlock();
2391 kfree_skb(reply);
2392 return err;
2393}
2394
2395static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2396{
2397 bool update_headroom = false;
2398 struct nlattr **a = info->attrs;
2399 struct sk_buff *reply;
2400 struct datapath *dp;
2401 struct vport *vport;
2402 unsigned int new_headroom;
2403 int err;
2404
2405 reply = ovs_vport_cmd_alloc_info();
2406 if (!reply)
2407 return -ENOMEM;
2408
2409 ovs_lock();
2410 vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2411 err = PTR_ERR(vport);
2412 if (IS_ERR(vport))
2413 goto exit_unlock_free;
2414
2415 if (vport->port_no == OVSP_LOCAL) {
2416 err = -EINVAL;
2417 goto exit_unlock_free;
2418 }
2419
2420 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2421 info->snd_portid, info->snd_seq, 0,
2422 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2423 BUG_ON(err < 0);
2424
2425 /* the vport deletion may trigger dp headroom update */
2426 dp = vport->dp;
2427 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2428 update_headroom = true;
2429
2430 netdev_reset_rx_headroom(vport->dev);
2431 ovs_dp_detach_port(vport);
2432
2433 if (update_headroom) {
2434 new_headroom = ovs_get_max_headroom(dp);
2435
2436 if (new_headroom < dp->max_headroom)
2437 ovs_update_headroom(dp, new_headroom);
2438 }
2439 ovs_unlock();
2440
2441 ovs_notify(&dp_vport_genl_family, reply, info);
2442 return 0;
2443
2444exit_unlock_free:
2445 ovs_unlock();
2446 kfree_skb(reply);
2447 return err;
2448}
2449
2450static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2451{
2452 struct nlattr **a = info->attrs;
2453 struct ovs_header *ovs_header = genl_info_userhdr(info);
2454 struct sk_buff *reply;
2455 struct vport *vport;
2456 int err;
2457
2458 reply = ovs_vport_cmd_alloc_info();
2459 if (!reply)
2460 return -ENOMEM;
2461
2462 rcu_read_lock();
2463 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2464 err = PTR_ERR(vport);
2465 if (IS_ERR(vport))
2466 goto exit_unlock_free;
2467 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2468 info->snd_portid, info->snd_seq, 0,
2469 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2470 BUG_ON(err < 0);
2471 rcu_read_unlock();
2472
2473 return genlmsg_reply(reply, info);
2474
2475exit_unlock_free:
2476 rcu_read_unlock();
2477 kfree_skb(reply);
2478 return err;
2479}
2480
2481static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2482{
2483 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2484 struct datapath *dp;
2485 int bucket = cb->args[0], skip = cb->args[1];
2486 int i, j = 0;
2487
2488 rcu_read_lock();
2489 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2490 if (!dp) {
2491 rcu_read_unlock();
2492 return -ENODEV;
2493 }
2494 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2495 struct vport *vport;
2496
2497 j = 0;
2498 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2499 if (j >= skip &&
2500 ovs_vport_cmd_fill_info(vport, skb,
2501 sock_net(skb->sk),
2502 NETLINK_CB(cb->skb).portid,
2503 cb->nlh->nlmsg_seq,
2504 NLM_F_MULTI,
2505 OVS_VPORT_CMD_GET,
2506 GFP_ATOMIC) < 0)
2507 goto out;
2508
2509 j++;
2510 }
2511 skip = 0;
2512 }
2513out:
2514 rcu_read_unlock();
2515
2516 cb->args[0] = i;
2517 cb->args[1] = j;
2518
2519 return skb->len;
2520}
2521
2522static void ovs_dp_masks_rebalance(struct work_struct *work)
2523{
2524 struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2525 masks_rebalance.work);
2526 struct datapath *dp;
2527
2528 ovs_lock();
2529
2530 list_for_each_entry(dp, &ovs_net->dps, list_node)
2531 ovs_flow_masks_rebalance(&dp->table);
2532
2533 ovs_unlock();
2534
2535 schedule_delayed_work(&ovs_net->masks_rebalance,
2536 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2537}
2538
2539static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2540 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2541 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2542 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2543 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2544 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2545 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2546 [OVS_VPORT_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2547 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2548 [OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
2549};
2550
2551static const struct genl_small_ops dp_vport_genl_ops[] = {
2552 { .cmd = OVS_VPORT_CMD_NEW,
2553 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2554 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2555 .doit = ovs_vport_cmd_new
2556 },
2557 { .cmd = OVS_VPORT_CMD_DEL,
2558 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2559 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2560 .doit = ovs_vport_cmd_del
2561 },
2562 { .cmd = OVS_VPORT_CMD_GET,
2563 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2564 .flags = 0, /* OK for unprivileged users. */
2565 .doit = ovs_vport_cmd_get,
2566 .dumpit = ovs_vport_cmd_dump
2567 },
2568 { .cmd = OVS_VPORT_CMD_SET,
2569 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2570 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2571 .doit = ovs_vport_cmd_set,
2572 },
2573};
2574
2575struct genl_family dp_vport_genl_family __ro_after_init = {
2576 .hdrsize = sizeof(struct ovs_header),
2577 .name = OVS_VPORT_FAMILY,
2578 .version = OVS_VPORT_VERSION,
2579 .maxattr = OVS_VPORT_ATTR_MAX,
2580 .policy = vport_policy,
2581 .netnsok = true,
2582 .parallel_ops = true,
2583 .small_ops = dp_vport_genl_ops,
2584 .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2585 .resv_start_op = OVS_VPORT_CMD_SET + 1,
2586 .mcgrps = &ovs_dp_vport_multicast_group,
2587 .n_mcgrps = 1,
2588 .module = THIS_MODULE,
2589};
2590
2591static struct genl_family * const dp_genl_families[] = {
2592 &dp_datapath_genl_family,
2593 &dp_vport_genl_family,
2594 &dp_flow_genl_family,
2595 &dp_packet_genl_family,
2596 &dp_meter_genl_family,
2597#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2598 &dp_ct_limit_genl_family,
2599#endif
2600};
2601
2602static void dp_unregister_genl(int n_families)
2603{
2604 int i;
2605
2606 for (i = 0; i < n_families; i++)
2607 genl_unregister_family(dp_genl_families[i]);
2608}
2609
2610static int __init dp_register_genl(void)
2611{
2612 int err;
2613 int i;
2614
2615 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2616
2617 err = genl_register_family(dp_genl_families[i]);
2618 if (err)
2619 goto error;
2620 }
2621
2622 return 0;
2623
2624error:
2625 dp_unregister_genl(i);
2626 return err;
2627}
2628
2629static int __net_init ovs_init_net(struct net *net)
2630{
2631 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2632 int err;
2633
2634 INIT_LIST_HEAD(&ovs_net->dps);
2635 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2636 INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2637
2638 err = ovs_ct_init(net);
2639 if (err)
2640 return err;
2641
2642 schedule_delayed_work(&ovs_net->masks_rebalance,
2643 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2644 return 0;
2645}
2646
2647static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2648 struct list_head *head)
2649{
2650 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2651 struct datapath *dp;
2652
2653 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2654 int i;
2655
2656 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2657 struct vport *vport;
2658
2659 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2660 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2661 continue;
2662
2663 if (dev_net(vport->dev) == dnet)
2664 list_add(&vport->detach_list, head);
2665 }
2666 }
2667 }
2668}
2669
2670static void __net_exit ovs_exit_net(struct net *dnet)
2671{
2672 struct datapath *dp, *dp_next;
2673 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2674 struct vport *vport, *vport_next;
2675 struct net *net;
2676 LIST_HEAD(head);
2677
2678 ovs_lock();
2679
2680 ovs_ct_exit(dnet);
2681
2682 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2683 __dp_destroy(dp);
2684
2685 down_read(&net_rwsem);
2686 for_each_net(net)
2687 list_vports_from_net(net, dnet, &head);
2688 up_read(&net_rwsem);
2689
2690 /* Detach all vports from given namespace. */
2691 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2692 list_del(&vport->detach_list);
2693 ovs_dp_detach_port(vport);
2694 }
2695
2696 ovs_unlock();
2697
2698 cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2699 cancel_work_sync(&ovs_net->dp_notify_work);
2700}
2701
2702static struct pernet_operations ovs_net_ops = {
2703 .init = ovs_init_net,
2704 .exit = ovs_exit_net,
2705 .id = &ovs_net_id,
2706 .size = sizeof(struct ovs_net),
2707};
2708
2709static const char * const ovs_drop_reasons[] = {
2710#define S(x) (#x),
2711 OVS_DROP_REASONS(S)
2712#undef S
2713};
2714
2715static struct drop_reason_list drop_reason_list_ovs = {
2716 .reasons = ovs_drop_reasons,
2717 .n_reasons = ARRAY_SIZE(ovs_drop_reasons),
2718};
2719
2720static int __init dp_init(void)
2721{
2722 int err;
2723
2724 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2725 sizeof_field(struct sk_buff, cb));
2726
2727 pr_info("Open vSwitch switching datapath\n");
2728
2729 err = action_fifos_init();
2730 if (err)
2731 goto error;
2732
2733 err = ovs_internal_dev_rtnl_link_register();
2734 if (err)
2735 goto error_action_fifos_exit;
2736
2737 err = ovs_flow_init();
2738 if (err)
2739 goto error_unreg_rtnl_link;
2740
2741 err = ovs_vport_init();
2742 if (err)
2743 goto error_flow_exit;
2744
2745 err = register_pernet_device(&ovs_net_ops);
2746 if (err)
2747 goto error_vport_exit;
2748
2749 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2750 if (err)
2751 goto error_netns_exit;
2752
2753 err = ovs_netdev_init();
2754 if (err)
2755 goto error_unreg_notifier;
2756
2757 err = dp_register_genl();
2758 if (err < 0)
2759 goto error_unreg_netdev;
2760
2761 drop_reasons_register_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH,
2762 &drop_reason_list_ovs);
2763
2764 return 0;
2765
2766error_unreg_netdev:
2767 ovs_netdev_exit();
2768error_unreg_notifier:
2769 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2770error_netns_exit:
2771 unregister_pernet_device(&ovs_net_ops);
2772error_vport_exit:
2773 ovs_vport_exit();
2774error_flow_exit:
2775 ovs_flow_exit();
2776error_unreg_rtnl_link:
2777 ovs_internal_dev_rtnl_link_unregister();
2778error_action_fifos_exit:
2779 action_fifos_exit();
2780error:
2781 return err;
2782}
2783
2784static void dp_cleanup(void)
2785{
2786 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2787 ovs_netdev_exit();
2788 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2789 unregister_pernet_device(&ovs_net_ops);
2790 drop_reasons_unregister_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH);
2791 rcu_barrier();
2792 ovs_vport_exit();
2793 ovs_flow_exit();
2794 ovs_internal_dev_rtnl_link_unregister();
2795 action_fifos_exit();
2796}
2797
2798module_init(dp_init);
2799module_exit(dp_cleanup);
2800
2801MODULE_DESCRIPTION("Open vSwitch switching datapath");
2802MODULE_LICENSE("GPL");
2803MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2804MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2805MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2806MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2807MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2808MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);