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