]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blob - net/openvswitch/datapath.c
projects / mirror_ubuntu-focal-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
UBUNTU: Ubuntu-5.4.0-117.132
[mirror_ubuntu-focal-kernel.git] / net / openvswitch / datapath.c
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
49 unsigned int ovs_net_id __read_mostly;
50
51 static struct genl_family dp_packet_genl_family;
52 static struct genl_family dp_flow_genl_family;
53 static struct genl_family dp_datapath_genl_family;
54
55 static const struct nla_policy flow_policy[];
56
57 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
58 .name = OVS_FLOW_MCGROUP,
59 };
60
61 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
62 .name = OVS_DATAPATH_MCGROUP,
63 };
64
65 static 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. */
71 static 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
78 static 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
101 static DEFINE_MUTEX(ovs_mutex);
102
103 void ovs_lock(void)
104 {
105 mutex_lock(&ovs_mutex);
106 }
107
108 void ovs_unlock(void)
109 {
110 mutex_unlock(&ovs_mutex);
111 }
112
113 #ifdef CONFIG_LOCKDEP
114 int 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
123 static struct vport *new_vport(const struct vport_parms *);
124 static 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);
128 static 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. */
134 const 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
140 static 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
158 static 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
169 static 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. */
176 struct 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. */
190 static 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
204 void 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. */
216 void 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
256 out:
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
264 int 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
286 err:
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
296 static 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
346 static 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
373 static 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
383 static 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;
525 out:
526 if (err)
527 skb_tx_error(skb);
528 kfree_skb(user_skb);
529 kfree_skb(nskb);
530 return err;
531 }
532
533 static 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
615 err_unlock:
616 rcu_read_unlock();
617 err_flow_free:
618 ovs_flow_free(flow, false);
619 err_kfree_skb:
620 kfree_skb(packet);
621 err:
622 return err;
623 }
624
625 static 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
633 static 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
641 static 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
654 static 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
685 static 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
691 static bool should_fill_mask(uint32_t ufid_flags)
692 {
693 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
694 }
695
696 static bool should_fill_actions(uint32_t ufid_flags)
697 {
698 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
699 }
700
701 static 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, or unmasked flow key as fallback
708 * see ovs_nla_put_identifier()
709 */
710 if (sfid && ovs_identifier_is_ufid(sfid))
711 len += nla_total_size(sfid->ufid_len);
712 else
713 len += nla_total_size(ovs_key_attr_size());
714
715 /* OVS_FLOW_ATTR_KEY */
716 if (!sfid || should_fill_key(sfid, ufid_flags))
717 len += nla_total_size(ovs_key_attr_size());
718
719 /* OVS_FLOW_ATTR_MASK */
720 if (should_fill_mask(ufid_flags))
721 len += nla_total_size(ovs_key_attr_size());
722
723 /* OVS_FLOW_ATTR_ACTIONS */
724 if (should_fill_actions(ufid_flags))
725 len += nla_total_size(acts->orig_len);
726
727 return len
728 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
729 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
730 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
731 }
732
733 /* Called with ovs_mutex or RCU read lock. */
734 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
735 struct sk_buff *skb)
736 {
737 struct ovs_flow_stats stats;
738 __be16 tcp_flags;
739 unsigned long used;
740
741 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
742
743 if (used &&
744 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
745 OVS_FLOW_ATTR_PAD))
746 return -EMSGSIZE;
747
748 if (stats.n_packets &&
749 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
750 sizeof(struct ovs_flow_stats), &stats,
751 OVS_FLOW_ATTR_PAD))
752 return -EMSGSIZE;
753
754 if ((u8)ntohs(tcp_flags) &&
755 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
756 return -EMSGSIZE;
757
758 return 0;
759 }
760
761 /* Called with ovs_mutex or RCU read lock. */
762 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
763 struct sk_buff *skb, int skb_orig_len)
764 {
765 struct nlattr *start;
766 int err;
767
768 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
769 * this is the first flow to be dumped into 'skb'. This is unusual for
770 * Netlink but individual action lists can be longer than
771 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
772 * The userspace caller can always fetch the actions separately if it
773 * really wants them. (Most userspace callers in fact don't care.)
774 *
775 * This can only fail for dump operations because the skb is always
776 * properly sized for single flows.
777 */
778 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
779 if (start) {
780 const struct sw_flow_actions *sf_acts;
781
782 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
783 err = ovs_nla_put_actions(sf_acts->actions,
784 sf_acts->actions_len, skb);
785
786 if (!err)
787 nla_nest_end(skb, start);
788 else {
789 if (skb_orig_len)
790 return err;
791
792 nla_nest_cancel(skb, start);
793 }
794 } else if (skb_orig_len) {
795 return -EMSGSIZE;
796 }
797
798 return 0;
799 }
800
801 /* Called with ovs_mutex or RCU read lock. */
802 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
803 struct sk_buff *skb, u32 portid,
804 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
805 {
806 const int skb_orig_len = skb->len;
807 struct ovs_header *ovs_header;
808 int err;
809
810 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
811 flags, cmd);
812 if (!ovs_header)
813 return -EMSGSIZE;
814
815 ovs_header->dp_ifindex = dp_ifindex;
816
817 err = ovs_nla_put_identifier(flow, skb);
818 if (err)
819 goto error;
820
821 if (should_fill_key(&flow->id, ufid_flags)) {
822 err = ovs_nla_put_masked_key(flow, skb);
823 if (err)
824 goto error;
825 }
826
827 if (should_fill_mask(ufid_flags)) {
828 err = ovs_nla_put_mask(flow, skb);
829 if (err)
830 goto error;
831 }
832
833 err = ovs_flow_cmd_fill_stats(flow, skb);
834 if (err)
835 goto error;
836
837 if (should_fill_actions(ufid_flags)) {
838 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
839 if (err)
840 goto error;
841 }
842
843 genlmsg_end(skb, ovs_header);
844 return 0;
845
846 error:
847 genlmsg_cancel(skb, ovs_header);
848 return err;
849 }
850
851 /* May not be called with RCU read lock. */
852 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
853 const struct sw_flow_id *sfid,
854 struct genl_info *info,
855 bool always,
856 uint32_t ufid_flags)
857 {
858 struct sk_buff *skb;
859 size_t len;
860
861 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
862 return NULL;
863
864 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
865 skb = genlmsg_new(len, GFP_KERNEL);
866 if (!skb)
867 return ERR_PTR(-ENOMEM);
868
869 return skb;
870 }
871
872 /* Called with ovs_mutex. */
873 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
874 int dp_ifindex,
875 struct genl_info *info, u8 cmd,
876 bool always, u32 ufid_flags)
877 {
878 struct sk_buff *skb;
879 int retval;
880
881 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
882 &flow->id, info, always, ufid_flags);
883 if (IS_ERR_OR_NULL(skb))
884 return skb;
885
886 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
887 info->snd_portid, info->snd_seq, 0,
888 cmd, ufid_flags);
889 if (WARN_ON_ONCE(retval < 0)) {
890 kfree_skb(skb);
891 skb = ERR_PTR(retval);
892 }
893 return skb;
894 }
895
896 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
897 {
898 struct net *net = sock_net(skb->sk);
899 struct nlattr **a = info->attrs;
900 struct ovs_header *ovs_header = info->userhdr;
901 struct sw_flow *flow = NULL, *new_flow;
902 struct sw_flow_mask mask;
903 struct sk_buff *reply;
904 struct datapath *dp;
905 struct sw_flow_actions *acts;
906 struct sw_flow_match match;
907 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
908 int error;
909 bool log = !a[OVS_FLOW_ATTR_PROBE];
910
911 /* Must have key and actions. */
912 error = -EINVAL;
913 if (!a[OVS_FLOW_ATTR_KEY]) {
914 OVS_NLERR(log, "Flow key attr not present in new flow.");
915 goto error;
916 }
917 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
918 OVS_NLERR(log, "Flow actions attr not present in new flow.");
919 goto error;
920 }
921
922 /* Most of the time we need to allocate a new flow, do it before
923 * locking.
924 */
925 new_flow = ovs_flow_alloc();
926 if (IS_ERR(new_flow)) {
927 error = PTR_ERR(new_flow);
928 goto error;
929 }
930
931 /* Extract key. */
932 ovs_match_init(&match, &new_flow->key, false, &mask);
933 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
934 a[OVS_FLOW_ATTR_MASK], log);
935 if (error)
936 goto err_kfree_flow;
937
938 /* Extract flow identifier. */
939 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
940 &new_flow->key, log);
941 if (error)
942 goto err_kfree_flow;
943
944 /* unmasked key is needed to match when ufid is not used. */
945 if (ovs_identifier_is_key(&new_flow->id))
946 match.key = new_flow->id.unmasked_key;
947
948 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
949
950 /* Validate actions. */
951 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
952 &new_flow->key, &acts, log);
953 if (error) {
954 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
955 goto err_kfree_flow;
956 }
957
958 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
959 ufid_flags);
960 if (IS_ERR(reply)) {
961 error = PTR_ERR(reply);
962 goto err_kfree_acts;
963 }
964
965 ovs_lock();
966 dp = get_dp(net, ovs_header->dp_ifindex);
967 if (unlikely(!dp)) {
968 error = -ENODEV;
969 goto err_unlock_ovs;
970 }
971
972 /* Check if this is a duplicate flow */
973 if (ovs_identifier_is_ufid(&new_flow->id))
974 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
975 if (!flow)
976 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
977 if (likely(!flow)) {
978 rcu_assign_pointer(new_flow->sf_acts, acts);
979
980 /* Put flow in bucket. */
981 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
982 if (unlikely(error)) {
983 acts = NULL;
984 goto err_unlock_ovs;
985 }
986
987 if (unlikely(reply)) {
988 error = ovs_flow_cmd_fill_info(new_flow,
989 ovs_header->dp_ifindex,
990 reply, info->snd_portid,
991 info->snd_seq, 0,
992 OVS_FLOW_CMD_NEW,
993 ufid_flags);
994 BUG_ON(error < 0);
995 }
996 ovs_unlock();
997 } else {
998 struct sw_flow_actions *old_acts;
999
1000 /* Bail out if we're not allowed to modify an existing flow.
1001 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1002 * because Generic Netlink treats the latter as a dump
1003 * request. We also accept NLM_F_EXCL in case that bug ever
1004 * gets fixed.
1005 */
1006 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1007 | NLM_F_EXCL))) {
1008 error = -EEXIST;
1009 goto err_unlock_ovs;
1010 }
1011 /* The flow identifier has to be the same for flow updates.
1012 * Look for any overlapping flow.
1013 */
1014 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1015 if (ovs_identifier_is_key(&flow->id))
1016 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1017 &match);
1018 else /* UFID matches but key is different */
1019 flow = NULL;
1020 if (!flow) {
1021 error = -ENOENT;
1022 goto err_unlock_ovs;
1023 }
1024 }
1025 /* Update actions. */
1026 old_acts = ovsl_dereference(flow->sf_acts);
1027 rcu_assign_pointer(flow->sf_acts, acts);
1028
1029 if (unlikely(reply)) {
1030 error = ovs_flow_cmd_fill_info(flow,
1031 ovs_header->dp_ifindex,
1032 reply, info->snd_portid,
1033 info->snd_seq, 0,
1034 OVS_FLOW_CMD_NEW,
1035 ufid_flags);
1036 BUG_ON(error < 0);
1037 }
1038 ovs_unlock();
1039
1040 ovs_nla_free_flow_actions_rcu(old_acts);
1041 ovs_flow_free(new_flow, false);
1042 }
1043
1044 if (reply)
1045 ovs_notify(&dp_flow_genl_family, reply, info);
1046 return 0;
1047
1048 err_unlock_ovs:
1049 ovs_unlock();
1050 kfree_skb(reply);
1051 err_kfree_acts:
1052 ovs_nla_free_flow_actions(acts);
1053 err_kfree_flow:
1054 ovs_flow_free(new_flow, false);
1055 error:
1056 return error;
1057 }
1058
1059 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1060 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1061 const struct nlattr *a,
1062 const struct sw_flow_key *key,
1063 const struct sw_flow_mask *mask,
1064 bool log)
1065 {
1066 struct sw_flow_actions *acts;
1067 struct sw_flow_key masked_key;
1068 int error;
1069
1070 ovs_flow_mask_key(&masked_key, key, true, mask);
1071 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1072 if (error) {
1073 OVS_NLERR(log,
1074 "Actions may not be safe on all matching packets");
1075 return ERR_PTR(error);
1076 }
1077
1078 return acts;
1079 }
1080
1081 /* Factor out match-init and action-copy to avoid
1082 * "Wframe-larger-than=1024" warning. Because mask is only
1083 * used to get actions, we new a function to save some
1084 * stack space.
1085 *
1086 * If there are not key and action attrs, we return 0
1087 * directly. In the case, the caller will also not use the
1088 * match as before. If there is action attr, we try to get
1089 * actions and save them to *acts. Before returning from
1090 * the function, we reset the match->mask pointer. Because
1091 * we should not to return match object with dangling reference
1092 * to mask.
1093 * */
1094 static noinline_for_stack int
1095 ovs_nla_init_match_and_action(struct net *net,
1096 struct sw_flow_match *match,
1097 struct sw_flow_key *key,
1098 struct nlattr **a,
1099 struct sw_flow_actions **acts,
1100 bool log)
1101 {
1102 struct sw_flow_mask mask;
1103 int error = 0;
1104
1105 if (a[OVS_FLOW_ATTR_KEY]) {
1106 ovs_match_init(match, key, true, &mask);
1107 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1108 a[OVS_FLOW_ATTR_MASK], log);
1109 if (error)
1110 goto error;
1111 }
1112
1113 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1114 if (!a[OVS_FLOW_ATTR_KEY]) {
1115 OVS_NLERR(log,
1116 "Flow key attribute not present in set flow.");
1117 error = -EINVAL;
1118 goto error;
1119 }
1120
1121 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1122 &mask, log);
1123 if (IS_ERR(*acts)) {
1124 error = PTR_ERR(*acts);
1125 goto error;
1126 }
1127 }
1128
1129 /* On success, error is 0. */
1130 error:
1131 match->mask = NULL;
1132 return error;
1133 }
1134
1135 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1136 {
1137 struct net *net = sock_net(skb->sk);
1138 struct nlattr **a = info->attrs;
1139 struct ovs_header *ovs_header = info->userhdr;
1140 struct sw_flow_key key;
1141 struct sw_flow *flow;
1142 struct sk_buff *reply = NULL;
1143 struct datapath *dp;
1144 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1145 struct sw_flow_match match;
1146 struct sw_flow_id sfid;
1147 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1148 int error = 0;
1149 bool log = !a[OVS_FLOW_ATTR_PROBE];
1150 bool ufid_present;
1151
1152 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1153 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1154 OVS_NLERR(log,
1155 "Flow set message rejected, Key attribute missing.");
1156 return -EINVAL;
1157 }
1158
1159 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1160 &acts, log);
1161 if (error)
1162 goto error;
1163
1164 if (acts) {
1165 /* Can allocate before locking if have acts. */
1166 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1167 ufid_flags);
1168 if (IS_ERR(reply)) {
1169 error = PTR_ERR(reply);
1170 goto err_kfree_acts;
1171 }
1172 }
1173
1174 ovs_lock();
1175 dp = get_dp(net, ovs_header->dp_ifindex);
1176 if (unlikely(!dp)) {
1177 error = -ENODEV;
1178 goto err_unlock_ovs;
1179 }
1180 /* Check that the flow exists. */
1181 if (ufid_present)
1182 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1183 else
1184 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1185 if (unlikely(!flow)) {
1186 error = -ENOENT;
1187 goto err_unlock_ovs;
1188 }
1189
1190 /* Update actions, if present. */
1191 if (likely(acts)) {
1192 old_acts = ovsl_dereference(flow->sf_acts);
1193 rcu_assign_pointer(flow->sf_acts, acts);
1194
1195 if (unlikely(reply)) {
1196 error = ovs_flow_cmd_fill_info(flow,
1197 ovs_header->dp_ifindex,
1198 reply, info->snd_portid,
1199 info->snd_seq, 0,
1200 OVS_FLOW_CMD_SET,
1201 ufid_flags);
1202 BUG_ON(error < 0);
1203 }
1204 } else {
1205 /* Could not alloc without acts before locking. */
1206 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1207 info, OVS_FLOW_CMD_SET, false,
1208 ufid_flags);
1209
1210 if (IS_ERR(reply)) {
1211 error = PTR_ERR(reply);
1212 goto err_unlock_ovs;
1213 }
1214 }
1215
1216 /* Clear stats. */
1217 if (a[OVS_FLOW_ATTR_CLEAR])
1218 ovs_flow_stats_clear(flow);
1219 ovs_unlock();
1220
1221 if (reply)
1222 ovs_notify(&dp_flow_genl_family, reply, info);
1223 if (old_acts)
1224 ovs_nla_free_flow_actions_rcu(old_acts);
1225
1226 return 0;
1227
1228 err_unlock_ovs:
1229 ovs_unlock();
1230 kfree_skb(reply);
1231 err_kfree_acts:
1232 ovs_nla_free_flow_actions(acts);
1233 error:
1234 return error;
1235 }
1236
1237 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1238 {
1239 struct nlattr **a = info->attrs;
1240 struct ovs_header *ovs_header = info->userhdr;
1241 struct net *net = sock_net(skb->sk);
1242 struct sw_flow_key key;
1243 struct sk_buff *reply;
1244 struct sw_flow *flow;
1245 struct datapath *dp;
1246 struct sw_flow_match match;
1247 struct sw_flow_id ufid;
1248 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1249 int err = 0;
1250 bool log = !a[OVS_FLOW_ATTR_PROBE];
1251 bool ufid_present;
1252
1253 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1254 if (a[OVS_FLOW_ATTR_KEY]) {
1255 ovs_match_init(&match, &key, true, NULL);
1256 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1257 log);
1258 } else if (!ufid_present) {
1259 OVS_NLERR(log,
1260 "Flow get message rejected, Key attribute missing.");
1261 err = -EINVAL;
1262 }
1263 if (err)
1264 return err;
1265
1266 ovs_lock();
1267 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1268 if (!dp) {
1269 err = -ENODEV;
1270 goto unlock;
1271 }
1272
1273 if (ufid_present)
1274 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1275 else
1276 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1277 if (!flow) {
1278 err = -ENOENT;
1279 goto unlock;
1280 }
1281
1282 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1283 OVS_FLOW_CMD_GET, true, ufid_flags);
1284 if (IS_ERR(reply)) {
1285 err = PTR_ERR(reply);
1286 goto unlock;
1287 }
1288
1289 ovs_unlock();
1290 return genlmsg_reply(reply, info);
1291 unlock:
1292 ovs_unlock();
1293 return err;
1294 }
1295
1296 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1297 {
1298 struct nlattr **a = info->attrs;
1299 struct ovs_header *ovs_header = info->userhdr;
1300 struct net *net = sock_net(skb->sk);
1301 struct sw_flow_key key;
1302 struct sk_buff *reply;
1303 struct sw_flow *flow = NULL;
1304 struct datapath *dp;
1305 struct sw_flow_match match;
1306 struct sw_flow_id ufid;
1307 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1308 int err;
1309 bool log = !a[OVS_FLOW_ATTR_PROBE];
1310 bool ufid_present;
1311
1312 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1313 if (a[OVS_FLOW_ATTR_KEY]) {
1314 ovs_match_init(&match, &key, true, NULL);
1315 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1316 NULL, log);
1317 if (unlikely(err))
1318 return err;
1319 }
1320
1321 ovs_lock();
1322 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1323 if (unlikely(!dp)) {
1324 err = -ENODEV;
1325 goto unlock;
1326 }
1327
1328 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1329 err = ovs_flow_tbl_flush(&dp->table);
1330 goto unlock;
1331 }
1332
1333 if (ufid_present)
1334 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1335 else
1336 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1337 if (unlikely(!flow)) {
1338 err = -ENOENT;
1339 goto unlock;
1340 }
1341
1342 ovs_flow_tbl_remove(&dp->table, flow);
1343 ovs_unlock();
1344
1345 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1346 &flow->id, info, false, ufid_flags);
1347 if (likely(reply)) {
1348 if (!IS_ERR(reply)) {
1349 rcu_read_lock(); /*To keep RCU checker happy. */
1350 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1351 reply, info->snd_portid,
1352 info->snd_seq, 0,
1353 OVS_FLOW_CMD_DEL,
1354 ufid_flags);
1355 rcu_read_unlock();
1356 if (WARN_ON_ONCE(err < 0)) {
1357 kfree_skb(reply);
1358 goto out_free;
1359 }
1360
1361 ovs_notify(&dp_flow_genl_family, reply, info);
1362 } else {
1363 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1364 }
1365 }
1366
1367 out_free:
1368 ovs_flow_free(flow, true);
1369 return 0;
1370 unlock:
1371 ovs_unlock();
1372 return err;
1373 }
1374
1375 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1376 {
1377 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1378 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1379 struct table_instance *ti;
1380 struct datapath *dp;
1381 u32 ufid_flags;
1382 int err;
1383
1384 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1385 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1386 if (err)
1387 return err;
1388 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1389
1390 rcu_read_lock();
1391 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1392 if (!dp) {
1393 rcu_read_unlock();
1394 return -ENODEV;
1395 }
1396
1397 ti = rcu_dereference(dp->table.ti);
1398 for (;;) {
1399 struct sw_flow *flow;
1400 u32 bucket, obj;
1401
1402 bucket = cb->args[0];
1403 obj = cb->args[1];
1404 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1405 if (!flow)
1406 break;
1407
1408 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1409 NETLINK_CB(cb->skb).portid,
1410 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1411 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1412 break;
1413
1414 cb->args[0] = bucket;
1415 cb->args[1] = obj;
1416 }
1417 rcu_read_unlock();
1418 return skb->len;
1419 }
1420
1421 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1422 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1423 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1424 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1425 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1426 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1427 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1428 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1429 };
1430
1431 static const struct genl_ops dp_flow_genl_ops[] = {
1432 { .cmd = OVS_FLOW_CMD_NEW,
1433 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1434 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1435 .doit = ovs_flow_cmd_new
1436 },
1437 { .cmd = OVS_FLOW_CMD_DEL,
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_del
1441 },
1442 { .cmd = OVS_FLOW_CMD_GET,
1443 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1444 .flags = 0, /* OK for unprivileged users. */
1445 .doit = ovs_flow_cmd_get,
1446 .dumpit = ovs_flow_cmd_dump
1447 },
1448 { .cmd = OVS_FLOW_CMD_SET,
1449 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1450 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1451 .doit = ovs_flow_cmd_set,
1452 },
1453 };
1454
1455 static struct genl_family dp_flow_genl_family __ro_after_init = {
1456 .hdrsize = sizeof(struct ovs_header),
1457 .name = OVS_FLOW_FAMILY,
1458 .version = OVS_FLOW_VERSION,
1459 .maxattr = OVS_FLOW_ATTR_MAX,
1460 .policy = flow_policy,
1461 .netnsok = true,
1462 .parallel_ops = true,
1463 .ops = dp_flow_genl_ops,
1464 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1465 .mcgrps = &ovs_dp_flow_multicast_group,
1466 .n_mcgrps = 1,
1467 .module = THIS_MODULE,
1468 };
1469
1470 static size_t ovs_dp_cmd_msg_size(void)
1471 {
1472 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1473
1474 msgsize += nla_total_size(IFNAMSIZ);
1475 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1476 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1477 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1478
1479 return msgsize;
1480 }
1481
1482 /* Called with ovs_mutex. */
1483 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1484 u32 portid, u32 seq, u32 flags, u8 cmd)
1485 {
1486 struct ovs_header *ovs_header;
1487 struct ovs_dp_stats dp_stats;
1488 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1489 int err;
1490
1491 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1492 flags, cmd);
1493 if (!ovs_header)
1494 goto error;
1495
1496 ovs_header->dp_ifindex = get_dpifindex(dp);
1497
1498 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1499 if (err)
1500 goto nla_put_failure;
1501
1502 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1503 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1504 &dp_stats, OVS_DP_ATTR_PAD))
1505 goto nla_put_failure;
1506
1507 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1508 sizeof(struct ovs_dp_megaflow_stats),
1509 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1510 goto nla_put_failure;
1511
1512 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1513 goto nla_put_failure;
1514
1515 genlmsg_end(skb, ovs_header);
1516 return 0;
1517
1518 nla_put_failure:
1519 genlmsg_cancel(skb, ovs_header);
1520 error:
1521 return -EMSGSIZE;
1522 }
1523
1524 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1525 {
1526 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1527 }
1528
1529 /* Called with rcu_read_lock or ovs_mutex. */
1530 static struct datapath *lookup_datapath(struct net *net,
1531 const struct ovs_header *ovs_header,
1532 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1533 {
1534 struct datapath *dp;
1535
1536 if (!a[OVS_DP_ATTR_NAME])
1537 dp = get_dp(net, ovs_header->dp_ifindex);
1538 else {
1539 struct vport *vport;
1540
1541 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1542 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1543 }
1544 return dp ? dp : ERR_PTR(-ENODEV);
1545 }
1546
1547 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1548 {
1549 struct datapath *dp;
1550
1551 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1552 if (IS_ERR(dp))
1553 return;
1554
1555 WARN(dp->user_features, "Dropping previously announced user features\n");
1556 dp->user_features = 0;
1557 }
1558
1559 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1560
1561 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1562 {
1563 u32 user_features = 0;
1564
1565 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1566 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1567
1568 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1569 OVS_DP_F_UNALIGNED |
1570 OVS_DP_F_TC_RECIRC_SHARING))
1571 return -EOPNOTSUPP;
1572
1573 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1574 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1575 return -EOPNOTSUPP;
1576 #endif
1577 }
1578
1579 dp->user_features = user_features;
1580
1581 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1582 static_branch_enable(&tc_recirc_sharing_support);
1583 else
1584 static_branch_disable(&tc_recirc_sharing_support);
1585
1586 return 0;
1587 }
1588
1589 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1590 {
1591 struct nlattr **a = info->attrs;
1592 struct vport_parms parms;
1593 struct sk_buff *reply;
1594 struct datapath *dp;
1595 struct vport *vport;
1596 struct ovs_net *ovs_net;
1597 int err, i;
1598
1599 err = -EINVAL;
1600 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1601 goto err;
1602
1603 reply = ovs_dp_cmd_alloc_info();
1604 if (!reply)
1605 return -ENOMEM;
1606
1607 err = -ENOMEM;
1608 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1609 if (dp == NULL)
1610 goto err_free_reply;
1611
1612 ovs_dp_set_net(dp, sock_net(skb->sk));
1613
1614 /* Allocate table. */
1615 err = ovs_flow_tbl_init(&dp->table);
1616 if (err)
1617 goto err_free_dp;
1618
1619 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1620 if (!dp->stats_percpu) {
1621 err = -ENOMEM;
1622 goto err_destroy_table;
1623 }
1624
1625 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1626 sizeof(struct hlist_head),
1627 GFP_KERNEL);
1628 if (!dp->ports) {
1629 err = -ENOMEM;
1630 goto err_destroy_percpu;
1631 }
1632
1633 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1634 INIT_HLIST_HEAD(&dp->ports[i]);
1635
1636 err = ovs_meters_init(dp);
1637 if (err)
1638 goto err_destroy_ports_array;
1639
1640 /* Set up our datapath device. */
1641 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1642 parms.type = OVS_VPORT_TYPE_INTERNAL;
1643 parms.options = NULL;
1644 parms.dp = dp;
1645 parms.port_no = OVSP_LOCAL;
1646 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1647
1648 err = ovs_dp_change(dp, a);
1649 if (err)
1650 goto err_destroy_meters;
1651
1652 /* So far only local changes have been made, now need the lock. */
1653 ovs_lock();
1654
1655 vport = new_vport(&parms);
1656 if (IS_ERR(vport)) {
1657 err = PTR_ERR(vport);
1658 if (err == -EBUSY)
1659 err = -EEXIST;
1660
1661 if (err == -EEXIST) {
1662 /* An outdated user space instance that does not understand
1663 * the concept of user_features has attempted to create a new
1664 * datapath and is likely to reuse it. Drop all user features.
1665 */
1666 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1667 ovs_dp_reset_user_features(skb, info);
1668 }
1669
1670 ovs_unlock();
1671 goto err_destroy_meters;
1672 }
1673
1674 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1675 info->snd_seq, 0, OVS_DP_CMD_NEW);
1676 BUG_ON(err < 0);
1677
1678 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1679 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1680
1681 ovs_unlock();
1682
1683 ovs_notify(&dp_datapath_genl_family, reply, info);
1684 return 0;
1685
1686 err_destroy_meters:
1687 ovs_meters_exit(dp);
1688 err_destroy_ports_array:
1689 kfree(dp->ports);
1690 err_destroy_percpu:
1691 free_percpu(dp->stats_percpu);
1692 err_destroy_table:
1693 ovs_flow_tbl_destroy(&dp->table);
1694 err_free_dp:
1695 kfree(dp);
1696 err_free_reply:
1697 kfree_skb(reply);
1698 err:
1699 return err;
1700 }
1701
1702 /* Called with ovs_mutex. */
1703 static void __dp_destroy(struct datapath *dp)
1704 {
1705 int i;
1706
1707 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1708 struct vport *vport;
1709 struct hlist_node *n;
1710
1711 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1712 if (vport->port_no != OVSP_LOCAL)
1713 ovs_dp_detach_port(vport);
1714 }
1715
1716 list_del_rcu(&dp->list_node);
1717
1718 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1719 * all ports in datapath are destroyed first before freeing datapath.
1720 */
1721 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1722
1723 /* RCU destroy the flow table */
1724 call_rcu(&dp->rcu, destroy_dp_rcu);
1725 }
1726
1727 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1728 {
1729 struct sk_buff *reply;
1730 struct datapath *dp;
1731 int err;
1732
1733 reply = ovs_dp_cmd_alloc_info();
1734 if (!reply)
1735 return -ENOMEM;
1736
1737 ovs_lock();
1738 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1739 err = PTR_ERR(dp);
1740 if (IS_ERR(dp))
1741 goto err_unlock_free;
1742
1743 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1744 info->snd_seq, 0, OVS_DP_CMD_DEL);
1745 BUG_ON(err < 0);
1746
1747 __dp_destroy(dp);
1748 ovs_unlock();
1749
1750 ovs_notify(&dp_datapath_genl_family, reply, info);
1751
1752 return 0;
1753
1754 err_unlock_free:
1755 ovs_unlock();
1756 kfree_skb(reply);
1757 return err;
1758 }
1759
1760 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1761 {
1762 struct sk_buff *reply;
1763 struct datapath *dp;
1764 int err;
1765
1766 reply = ovs_dp_cmd_alloc_info();
1767 if (!reply)
1768 return -ENOMEM;
1769
1770 ovs_lock();
1771 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1772 err = PTR_ERR(dp);
1773 if (IS_ERR(dp))
1774 goto err_unlock_free;
1775
1776 err = ovs_dp_change(dp, info->attrs);
1777 if (err)
1778 goto err_unlock_free;
1779
1780 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1781 info->snd_seq, 0, OVS_DP_CMD_SET);
1782 BUG_ON(err < 0);
1783
1784 ovs_unlock();
1785 ovs_notify(&dp_datapath_genl_family, reply, info);
1786
1787 return 0;
1788
1789 err_unlock_free:
1790 ovs_unlock();
1791 kfree_skb(reply);
1792 return err;
1793 }
1794
1795 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1796 {
1797 struct sk_buff *reply;
1798 struct datapath *dp;
1799 int err;
1800
1801 reply = ovs_dp_cmd_alloc_info();
1802 if (!reply)
1803 return -ENOMEM;
1804
1805 ovs_lock();
1806 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1807 if (IS_ERR(dp)) {
1808 err = PTR_ERR(dp);
1809 goto err_unlock_free;
1810 }
1811 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1812 info->snd_seq, 0, OVS_DP_CMD_GET);
1813 BUG_ON(err < 0);
1814 ovs_unlock();
1815
1816 return genlmsg_reply(reply, info);
1817
1818 err_unlock_free:
1819 ovs_unlock();
1820 kfree_skb(reply);
1821 return err;
1822 }
1823
1824 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1825 {
1826 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1827 struct datapath *dp;
1828 int skip = cb->args[0];
1829 int i = 0;
1830
1831 ovs_lock();
1832 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1833 if (i >= skip &&
1834 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1835 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1836 OVS_DP_CMD_GET) < 0)
1837 break;
1838 i++;
1839 }
1840 ovs_unlock();
1841
1842 cb->args[0] = i;
1843
1844 return skb->len;
1845 }
1846
1847 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1848 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1849 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1850 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1851 };
1852
1853 static const struct genl_ops dp_datapath_genl_ops[] = {
1854 { .cmd = OVS_DP_CMD_NEW,
1855 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1856 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1857 .doit = ovs_dp_cmd_new
1858 },
1859 { .cmd = OVS_DP_CMD_DEL,
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_del
1863 },
1864 { .cmd = OVS_DP_CMD_GET,
1865 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1866 .flags = 0, /* OK for unprivileged users. */
1867 .doit = ovs_dp_cmd_get,
1868 .dumpit = ovs_dp_cmd_dump
1869 },
1870 { .cmd = OVS_DP_CMD_SET,
1871 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1872 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1873 .doit = ovs_dp_cmd_set,
1874 },
1875 };
1876
1877 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1878 .hdrsize = sizeof(struct ovs_header),
1879 .name = OVS_DATAPATH_FAMILY,
1880 .version = OVS_DATAPATH_VERSION,
1881 .maxattr = OVS_DP_ATTR_MAX,
1882 .policy = datapath_policy,
1883 .netnsok = true,
1884 .parallel_ops = true,
1885 .ops = dp_datapath_genl_ops,
1886 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1887 .mcgrps = &ovs_dp_datapath_multicast_group,
1888 .n_mcgrps = 1,
1889 .module = THIS_MODULE,
1890 };
1891
1892 /* Called with ovs_mutex or RCU read lock. */
1893 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1894 struct net *net, u32 portid, u32 seq,
1895 u32 flags, u8 cmd, gfp_t gfp)
1896 {
1897 struct ovs_header *ovs_header;
1898 struct ovs_vport_stats vport_stats;
1899 int err;
1900
1901 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1902 flags, cmd);
1903 if (!ovs_header)
1904 return -EMSGSIZE;
1905
1906 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1907
1908 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1909 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1910 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1911 ovs_vport_name(vport)) ||
1912 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1913 goto nla_put_failure;
1914
1915 if (!net_eq(net, dev_net(vport->dev))) {
1916 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1917
1918 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1919 goto nla_put_failure;
1920 }
1921
1922 ovs_vport_get_stats(vport, &vport_stats);
1923 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1924 sizeof(struct ovs_vport_stats), &vport_stats,
1925 OVS_VPORT_ATTR_PAD))
1926 goto nla_put_failure;
1927
1928 if (ovs_vport_get_upcall_portids(vport, skb))
1929 goto nla_put_failure;
1930
1931 err = ovs_vport_get_options(vport, skb);
1932 if (err == -EMSGSIZE)
1933 goto error;
1934
1935 genlmsg_end(skb, ovs_header);
1936 return 0;
1937
1938 nla_put_failure:
1939 err = -EMSGSIZE;
1940 error:
1941 genlmsg_cancel(skb, ovs_header);
1942 return err;
1943 }
1944
1945 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1946 {
1947 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1948 }
1949
1950 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1951 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1952 u32 portid, u32 seq, u8 cmd)
1953 {
1954 struct sk_buff *skb;
1955 int retval;
1956
1957 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1958 if (!skb)
1959 return ERR_PTR(-ENOMEM);
1960
1961 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1962 GFP_KERNEL);
1963 BUG_ON(retval < 0);
1964
1965 return skb;
1966 }
1967
1968 /* Called with ovs_mutex or RCU read lock. */
1969 static struct vport *lookup_vport(struct net *net,
1970 const struct ovs_header *ovs_header,
1971 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1972 {
1973 struct datapath *dp;
1974 struct vport *vport;
1975
1976 if (a[OVS_VPORT_ATTR_IFINDEX])
1977 return ERR_PTR(-EOPNOTSUPP);
1978 if (a[OVS_VPORT_ATTR_NAME]) {
1979 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1980 if (!vport)
1981 return ERR_PTR(-ENODEV);
1982 if (ovs_header->dp_ifindex &&
1983 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1984 return ERR_PTR(-ENODEV);
1985 return vport;
1986 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1987 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1988
1989 if (port_no >= DP_MAX_PORTS)
1990 return ERR_PTR(-EFBIG);
1991
1992 dp = get_dp(net, ovs_header->dp_ifindex);
1993 if (!dp)
1994 return ERR_PTR(-ENODEV);
1995
1996 vport = ovs_vport_ovsl_rcu(dp, port_no);
1997 if (!vport)
1998 return ERR_PTR(-ENODEV);
1999 return vport;
2000 } else
2001 return ERR_PTR(-EINVAL);
2002
2003 }
2004
2005 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2006 {
2007 unsigned int dev_headroom, max_headroom = 0;
2008 struct net_device *dev;
2009 struct vport *vport;
2010 int i;
2011
2012 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2013 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2014 dev = vport->dev;
2015 dev_headroom = netdev_get_fwd_headroom(dev);
2016 if (dev_headroom > max_headroom)
2017 max_headroom = dev_headroom;
2018 }
2019 }
2020
2021 return max_headroom;
2022 }
2023
2024 /* Called with ovs_mutex */
2025 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2026 {
2027 struct vport *vport;
2028 int i;
2029
2030 dp->max_headroom = new_headroom;
2031 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2032 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2033 netdev_set_rx_headroom(vport->dev, new_headroom);
2034 }
2035
2036 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2037 {
2038 struct nlattr **a = info->attrs;
2039 struct ovs_header *ovs_header = info->userhdr;
2040 struct vport_parms parms;
2041 struct sk_buff *reply;
2042 struct vport *vport;
2043 struct datapath *dp;
2044 unsigned int new_headroom;
2045 u32 port_no;
2046 int err;
2047
2048 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2049 !a[OVS_VPORT_ATTR_UPCALL_PID])
2050 return -EINVAL;
2051 if (a[OVS_VPORT_ATTR_IFINDEX])
2052 return -EOPNOTSUPP;
2053
2054 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2055 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2056 if (port_no >= DP_MAX_PORTS)
2057 return -EFBIG;
2058
2059 reply = ovs_vport_cmd_alloc_info();
2060 if (!reply)
2061 return -ENOMEM;
2062
2063 ovs_lock();
2064 restart:
2065 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2066 err = -ENODEV;
2067 if (!dp)
2068 goto exit_unlock_free;
2069
2070 if (port_no) {
2071 vport = ovs_vport_ovsl(dp, port_no);
2072 err = -EBUSY;
2073 if (vport)
2074 goto exit_unlock_free;
2075 } else {
2076 for (port_no = 1; ; port_no++) {
2077 if (port_no >= DP_MAX_PORTS) {
2078 err = -EFBIG;
2079 goto exit_unlock_free;
2080 }
2081 vport = ovs_vport_ovsl(dp, port_no);
2082 if (!vport)
2083 break;
2084 }
2085 }
2086
2087 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2088 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2089 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2090 parms.dp = dp;
2091 parms.port_no = port_no;
2092 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2093
2094 vport = new_vport(&parms);
2095 err = PTR_ERR(vport);
2096 if (IS_ERR(vport)) {
2097 if (err == -EAGAIN)
2098 goto restart;
2099 goto exit_unlock_free;
2100 }
2101
2102 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2103 info->snd_portid, info->snd_seq, 0,
2104 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2105
2106 new_headroom = netdev_get_fwd_headroom(vport->dev);
2107
2108 if (new_headroom > dp->max_headroom)
2109 ovs_update_headroom(dp, new_headroom);
2110 else
2111 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2112
2113 BUG_ON(err < 0);
2114 ovs_unlock();
2115
2116 ovs_notify(&dp_vport_genl_family, reply, info);
2117 return 0;
2118
2119 exit_unlock_free:
2120 ovs_unlock();
2121 kfree_skb(reply);
2122 return err;
2123 }
2124
2125 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2126 {
2127 struct nlattr **a = info->attrs;
2128 struct sk_buff *reply;
2129 struct vport *vport;
2130 int err;
2131
2132 reply = ovs_vport_cmd_alloc_info();
2133 if (!reply)
2134 return -ENOMEM;
2135
2136 ovs_lock();
2137 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2138 err = PTR_ERR(vport);
2139 if (IS_ERR(vport))
2140 goto exit_unlock_free;
2141
2142 if (a[OVS_VPORT_ATTR_TYPE] &&
2143 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2144 err = -EINVAL;
2145 goto exit_unlock_free;
2146 }
2147
2148 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2149 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2150 if (err)
2151 goto exit_unlock_free;
2152 }
2153
2154
2155 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2156 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2157
2158 err = ovs_vport_set_upcall_portids(vport, ids);
2159 if (err)
2160 goto exit_unlock_free;
2161 }
2162
2163 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2164 info->snd_portid, info->snd_seq, 0,
2165 OVS_VPORT_CMD_SET, GFP_KERNEL);
2166 BUG_ON(err < 0);
2167
2168 ovs_unlock();
2169 ovs_notify(&dp_vport_genl_family, reply, info);
2170 return 0;
2171
2172 exit_unlock_free:
2173 ovs_unlock();
2174 kfree_skb(reply);
2175 return err;
2176 }
2177
2178 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2179 {
2180 bool update_headroom = false;
2181 struct nlattr **a = info->attrs;
2182 struct sk_buff *reply;
2183 struct datapath *dp;
2184 struct vport *vport;
2185 unsigned int new_headroom;
2186 int err;
2187
2188 reply = ovs_vport_cmd_alloc_info();
2189 if (!reply)
2190 return -ENOMEM;
2191
2192 ovs_lock();
2193 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2194 err = PTR_ERR(vport);
2195 if (IS_ERR(vport))
2196 goto exit_unlock_free;
2197
2198 if (vport->port_no == OVSP_LOCAL) {
2199 err = -EINVAL;
2200 goto exit_unlock_free;
2201 }
2202
2203 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2204 info->snd_portid, info->snd_seq, 0,
2205 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2206 BUG_ON(err < 0);
2207
2208 /* the vport deletion may trigger dp headroom update */
2209 dp = vport->dp;
2210 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2211 update_headroom = true;
2212
2213 netdev_reset_rx_headroom(vport->dev);
2214 ovs_dp_detach_port(vport);
2215
2216 if (update_headroom) {
2217 new_headroom = ovs_get_max_headroom(dp);
2218
2219 if (new_headroom < dp->max_headroom)
2220 ovs_update_headroom(dp, new_headroom);
2221 }
2222 ovs_unlock();
2223
2224 ovs_notify(&dp_vport_genl_family, reply, info);
2225 return 0;
2226
2227 exit_unlock_free:
2228 ovs_unlock();
2229 kfree_skb(reply);
2230 return err;
2231 }
2232
2233 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2234 {
2235 struct nlattr **a = info->attrs;
2236 struct ovs_header *ovs_header = info->userhdr;
2237 struct sk_buff *reply;
2238 struct vport *vport;
2239 int err;
2240
2241 reply = ovs_vport_cmd_alloc_info();
2242 if (!reply)
2243 return -ENOMEM;
2244
2245 rcu_read_lock();
2246 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2247 err = PTR_ERR(vport);
2248 if (IS_ERR(vport))
2249 goto exit_unlock_free;
2250 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2251 info->snd_portid, info->snd_seq, 0,
2252 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2253 BUG_ON(err < 0);
2254 rcu_read_unlock();
2255
2256 return genlmsg_reply(reply, info);
2257
2258 exit_unlock_free:
2259 rcu_read_unlock();
2260 kfree_skb(reply);
2261 return err;
2262 }
2263
2264 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2265 {
2266 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2267 struct datapath *dp;
2268 int bucket = cb->args[0], skip = cb->args[1];
2269 int i, j = 0;
2270
2271 rcu_read_lock();
2272 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2273 if (!dp) {
2274 rcu_read_unlock();
2275 return -ENODEV;
2276 }
2277 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2278 struct vport *vport;
2279
2280 j = 0;
2281 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2282 if (j >= skip &&
2283 ovs_vport_cmd_fill_info(vport, skb,
2284 sock_net(skb->sk),
2285 NETLINK_CB(cb->skb).portid,
2286 cb->nlh->nlmsg_seq,
2287 NLM_F_MULTI,
2288 OVS_VPORT_CMD_GET,
2289 GFP_ATOMIC) < 0)
2290 goto out;
2291
2292 j++;
2293 }
2294 skip = 0;
2295 }
2296 out:
2297 rcu_read_unlock();
2298
2299 cb->args[0] = i;
2300 cb->args[1] = j;
2301
2302 return skb->len;
2303 }
2304
2305 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2306 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2307 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2308 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2309 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2310 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2311 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2312 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2313 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2314 };
2315
2316 static const struct genl_ops dp_vport_genl_ops[] = {
2317 { .cmd = OVS_VPORT_CMD_NEW,
2318 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2319 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2320 .doit = ovs_vport_cmd_new
2321 },
2322 { .cmd = OVS_VPORT_CMD_DEL,
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_del
2326 },
2327 { .cmd = OVS_VPORT_CMD_GET,
2328 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2329 .flags = 0, /* OK for unprivileged users. */
2330 .doit = ovs_vport_cmd_get,
2331 .dumpit = ovs_vport_cmd_dump
2332 },
2333 { .cmd = OVS_VPORT_CMD_SET,
2334 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2335 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2336 .doit = ovs_vport_cmd_set,
2337 },
2338 };
2339
2340 struct genl_family dp_vport_genl_family __ro_after_init = {
2341 .hdrsize = sizeof(struct ovs_header),
2342 .name = OVS_VPORT_FAMILY,
2343 .version = OVS_VPORT_VERSION,
2344 .maxattr = OVS_VPORT_ATTR_MAX,
2345 .policy = vport_policy,
2346 .netnsok = true,
2347 .parallel_ops = true,
2348 .ops = dp_vport_genl_ops,
2349 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2350 .mcgrps = &ovs_dp_vport_multicast_group,
2351 .n_mcgrps = 1,
2352 .module = THIS_MODULE,
2353 };
2354
2355 static struct genl_family * const dp_genl_families[] = {
2356 &dp_datapath_genl_family,
2357 &dp_vport_genl_family,
2358 &dp_flow_genl_family,
2359 &dp_packet_genl_family,
2360 &dp_meter_genl_family,
2361 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2362 &dp_ct_limit_genl_family,
2363 #endif
2364 };
2365
2366 static void dp_unregister_genl(int n_families)
2367 {
2368 int i;
2369
2370 for (i = 0; i < n_families; i++)
2371 genl_unregister_family(dp_genl_families[i]);
2372 }
2373
2374 static int __init dp_register_genl(void)
2375 {
2376 int err;
2377 int i;
2378
2379 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2380
2381 err = genl_register_family(dp_genl_families[i]);
2382 if (err)
2383 goto error;
2384 }
2385
2386 return 0;
2387
2388 error:
2389 dp_unregister_genl(i);
2390 return err;
2391 }
2392
2393 static int __net_init ovs_init_net(struct net *net)
2394 {
2395 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2396
2397 INIT_LIST_HEAD(&ovs_net->dps);
2398 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2399 return ovs_ct_init(net);
2400 }
2401
2402 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2403 struct list_head *head)
2404 {
2405 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2406 struct datapath *dp;
2407
2408 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2409 int i;
2410
2411 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2412 struct vport *vport;
2413
2414 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2415 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2416 continue;
2417
2418 if (dev_net(vport->dev) == dnet)
2419 list_add(&vport->detach_list, head);
2420 }
2421 }
2422 }
2423 }
2424
2425 static void __net_exit ovs_exit_net(struct net *dnet)
2426 {
2427 struct datapath *dp, *dp_next;
2428 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2429 struct vport *vport, *vport_next;
2430 struct net *net;
2431 LIST_HEAD(head);
2432
2433 ovs_lock();
2434
2435 ovs_ct_exit(dnet);
2436
2437 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2438 __dp_destroy(dp);
2439
2440 down_read(&net_rwsem);
2441 for_each_net(net)
2442 list_vports_from_net(net, dnet, &head);
2443 up_read(&net_rwsem);
2444
2445 /* Detach all vports from given namespace. */
2446 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2447 list_del(&vport->detach_list);
2448 ovs_dp_detach_port(vport);
2449 }
2450
2451 ovs_unlock();
2452
2453 cancel_work_sync(&ovs_net->dp_notify_work);
2454 }
2455
2456 static struct pernet_operations ovs_net_ops = {
2457 .init = ovs_init_net,
2458 .exit = ovs_exit_net,
2459 .id = &ovs_net_id,
2460 .size = sizeof(struct ovs_net),
2461 };
2462
2463 static int __init dp_init(void)
2464 {
2465 int err;
2466
2467 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2468
2469 pr_info("Open vSwitch switching datapath\n");
2470
2471 err = action_fifos_init();
2472 if (err)
2473 goto error;
2474
2475 err = ovs_internal_dev_rtnl_link_register();
2476 if (err)
2477 goto error_action_fifos_exit;
2478
2479 err = ovs_flow_init();
2480 if (err)
2481 goto error_unreg_rtnl_link;
2482
2483 err = ovs_vport_init();
2484 if (err)
2485 goto error_flow_exit;
2486
2487 err = register_pernet_device(&ovs_net_ops);
2488 if (err)
2489 goto error_vport_exit;
2490
2491 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2492 if (err)
2493 goto error_netns_exit;
2494
2495 err = ovs_netdev_init();
2496 if (err)
2497 goto error_unreg_notifier;
2498
2499 err = dp_register_genl();
2500 if (err < 0)
2501 goto error_unreg_netdev;
2502
2503 return 0;
2504
2505 error_unreg_netdev:
2506 ovs_netdev_exit();
2507 error_unreg_notifier:
2508 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2509 error_netns_exit:
2510 unregister_pernet_device(&ovs_net_ops);
2511 error_vport_exit:
2512 ovs_vport_exit();
2513 error_flow_exit:
2514 ovs_flow_exit();
2515 error_unreg_rtnl_link:
2516 ovs_internal_dev_rtnl_link_unregister();
2517 error_action_fifos_exit:
2518 action_fifos_exit();
2519 error:
2520 return err;
2521 }
2522
2523 static void dp_cleanup(void)
2524 {
2525 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2526 ovs_netdev_exit();
2527 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2528 unregister_pernet_device(&ovs_net_ops);
2529 rcu_barrier();
2530 ovs_vport_exit();
2531 ovs_flow_exit();
2532 ovs_internal_dev_rtnl_link_unregister();
2533 action_fifos_exit();
2534 }
2535
2536 module_init(dp_init);
2537 module_exit(dp_cleanup);
2538
2539 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2540 MODULE_LICENSE("GPL");
2541 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2542 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2543 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2544 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2545 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2546 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
Ubuntu Focal Linux kernel mirror