projects/linux·blob master · refs

2919 lines · 72.0 KiB

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/kernel.h>
19#include <linux/kthread.h>
20#include <linux/mutex.h>
21#include <linux/percpu.h>
22#include <linux/rcupdate.h>
23#include <linux/tcp.h>
24#include <linux/udp.h>
25#include <linux/ethtool.h>
26#include <linux/wait.h>
27#include <asm/div64.h>
28#include <linux/highmem.h>
29#include <linux/netfilter_bridge.h>
30#include <linux/netfilter_ipv4.h>
31#include <linux/inetdevice.h>
32#include <linux/list.h>
33#include <linux/openvswitch.h>
34#include <linux/rculist.h>
35#include <linux/dmi.h>
36#include <net/genetlink.h>
37#include <net/gso.h>
38#include <net/net_namespace.h>
39#include <net/netns/generic.h>
40#include <net/pkt_cls.h>
41
42#include "datapath.h"
43#include "drop.h"
44#include "flow.h"
45#include "flow_table.h"
46#include "flow_netlink.h"
47#include "meter.h"
48#include "openvswitch_trace.h"
49#include "vport-internal_dev.h"
50#include "vport-netdev.h"
51
52unsigned int ovs_net_id __read_mostly;
53
54static struct genl_family dp_packet_genl_family;
55static struct genl_family dp_flow_genl_family;
56static struct genl_family dp_datapath_genl_family;
57
58static const struct nla_policy flow_policy[];
59
60static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
61 .name = OVS_FLOW_MCGROUP,
62};
63
64static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
65 .name = OVS_DATAPATH_MCGROUP,
66};
67
68static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
69 .name = OVS_VPORT_MCGROUP,
70};
71
72/* Check if need to build a reply message.
73 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
74static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
75 unsigned int group)
76{
77 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
78 genl_has_listeners(family, genl_info_net(info), group);
79}
80
81static void ovs_notify(struct genl_family *family,
82 struct sk_buff *skb, struct genl_info *info)
83{
84 genl_notify(family, skb, info, 0, GFP_KERNEL);
85}
86
87/**
88 * DOC: Locking:
89 *
90 * All writes e.g. Writes to device state (add/remove datapath, port, set
91 * operations on vports, etc.), Writes to other state (flow table
92 * modifications, set miscellaneous datapath parameters, etc.) are protected
93 * by ovs_lock.
94 *
95 * Reads are protected by RCU.
96 *
97 * There are a few special cases (mostly stats) that have their own
98 * synchronization but they nest under all of above and don't interact with
99 * each other.
100 *
101 * The RTNL lock nests inside ovs_mutex.
102 */
103
104static DEFINE_MUTEX(ovs_mutex);
105
106void ovs_lock(void)
107{
108 mutex_lock(&ovs_mutex);
109}
110
111void ovs_unlock(void)
112{
113 mutex_unlock(&ovs_mutex);
114}
115
116#ifdef CONFIG_LOCKDEP
117int lockdep_ovsl_is_held(void)
118{
119 if (debug_locks)
120 return lockdep_is_held(&ovs_mutex);
121 else
122 return 1;
123}
124#endif
125
126static struct vport *new_vport(const struct vport_parms *);
127static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
128 const struct sw_flow_key *,
129 const struct dp_upcall_info *,
130 uint32_t cutlen);
131static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
132 const struct sw_flow_key *,
133 const struct dp_upcall_info *,
134 uint32_t cutlen);
135
136static void ovs_dp_masks_rebalance(struct work_struct *work);
137
138static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
139
140/* Must be called with rcu_read_lock or ovs_mutex. */
141const char *ovs_dp_name(const struct datapath *dp)
142{
143 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
144 return ovs_vport_name(vport);
145}
146
147static int get_dpifindex(const struct datapath *dp)
148{
149 struct vport *local;
150 int ifindex;
151
152 rcu_read_lock();
153
154 local = ovs_vport_rcu(dp, OVSP_LOCAL);
155 if (local)
156 ifindex = local->dev->ifindex;
157 else
158 ifindex = 0;
159
160 rcu_read_unlock();
161
162 return ifindex;
163}
164
165static void destroy_dp_rcu(struct rcu_head *rcu)
166{
167 struct datapath *dp = container_of(rcu, struct datapath, rcu);
168
169 ovs_flow_tbl_destroy(&dp->table);
170 free_percpu(dp->stats_percpu);
171 kfree(dp->ports);
172 ovs_meters_exit(dp);
173 kfree(rcu_dereference_raw(dp->upcall_portids));
174 kfree(dp);
175}
176
177static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
178 u16 port_no)
179{
180 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
181}
182
183/* Called with ovs_mutex or RCU read lock. */
184struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
185{
186 struct vport *vport;
187 struct hlist_head *head;
188
189 head = vport_hash_bucket(dp, port_no);
190 hlist_for_each_entry_rcu(vport, head, dp_hash_node,
191 lockdep_ovsl_is_held()) {
192 if (vport->port_no == port_no)
193 return vport;
194 }
195 return NULL;
196}
197
198/* Called with ovs_mutex. */
199static struct vport *new_vport(const struct vport_parms *parms)
200{
201 struct vport *vport;
202
203 vport = ovs_vport_add(parms);
204 if (!IS_ERR(vport)) {
205 struct datapath *dp = parms->dp;
206 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
207
208 hlist_add_head_rcu(&vport->dp_hash_node, head);
209 }
210 return vport;
211}
212
213static void ovs_vport_update_upcall_stats(struct sk_buff *skb,
214 const struct dp_upcall_info *upcall_info,
215 bool upcall_result)
216{
217 struct vport *p = OVS_CB(skb)->input_vport;
218 struct vport_upcall_stats_percpu *stats;
219
220 if (upcall_info->cmd != OVS_PACKET_CMD_MISS &&
221 upcall_info->cmd != OVS_PACKET_CMD_ACTION)
222 return;
223
224 stats = this_cpu_ptr(p->upcall_stats);
225 u64_stats_update_begin(&stats->syncp);
226 if (upcall_result)
227 u64_stats_inc(&stats->n_success);
228 else
229 u64_stats_inc(&stats->n_fail);
230 u64_stats_update_end(&stats->syncp);
231}
232
233void ovs_dp_detach_port(struct vport *p)
234{
235 ASSERT_OVSL();
236
237 /* First drop references to device. */
238 hlist_del_rcu(&p->dp_hash_node);
239
240 /* Then destroy it. */
241 ovs_vport_del(p);
242}
243
244/* Must be called with rcu_read_lock. */
245void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
246{
247 struct ovs_pcpu_storage *ovs_pcpu = this_cpu_ptr(ovs_pcpu_storage);
248 const struct vport *p = OVS_CB(skb)->input_vport;
249 struct datapath *dp = p->dp;
250 struct sw_flow *flow;
251 struct sw_flow_actions *sf_acts;
252 struct dp_stats_percpu *stats;
253 bool ovs_pcpu_locked = false;
254 u64 *stats_counter;
255 u32 n_mask_hit;
256 u32 n_cache_hit;
257 int error;
258
259 stats = this_cpu_ptr(dp->stats_percpu);
260
261 /* Look up flow. */
262 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
263 &n_mask_hit, &n_cache_hit);
264 if (unlikely(!flow)) {
265 struct dp_upcall_info upcall;
266
267 memset(&upcall, 0, sizeof(upcall));
268 upcall.cmd = OVS_PACKET_CMD_MISS;
269
270 if (OVS_CB(skb)->upcall_pid)
271 upcall.portid = OVS_CB(skb)->upcall_pid;
272 else if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
273 upcall.portid =
274 ovs_dp_get_upcall_portid(dp, smp_processor_id());
275 else
276 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277
278 upcall.mru = OVS_CB(skb)->mru;
279 error = ovs_dp_upcall(dp, skb, key, &upcall, U32_MAX);
280 switch (error) {
281 case 0:
282 case -EAGAIN:
283 case -ERESTARTSYS:
284 case -EINTR:
285 consume_skb(skb);
286 break;
287 default:
288 skb_tx_error(skb);
289 kfree_skb(skb);
290 break;
291 }
292 stats_counter = &stats->n_missed;
293 goto out;
294 }
295
296 ovs_flow_stats_update(flow, key->tp.flags, skb);
297 sf_acts = rcu_dereference(flow->sf_acts);
298 /* This path can be invoked recursively: Use the current task to
299 * identify recursive invocation - the lock must be acquired only once.
300 * Even with disabled bottom halves this can be preempted on PREEMPT_RT.
301 * Limit the locking to RT to avoid assigning `owner' if it can be
302 * avoided.
303 */
304 if (IS_ENABLED(CONFIG_PREEMPT_RT) && ovs_pcpu->owner != current) {
305 local_lock_nested_bh(&ovs_pcpu_storage->bh_lock);
306 ovs_pcpu->owner = current;
307 ovs_pcpu_locked = true;
308 }
309
310 error = ovs_execute_actions(dp, skb, sf_acts, key);
311 if (unlikely(error))
312 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
313 ovs_dp_name(dp), error);
314 if (ovs_pcpu_locked) {
315 ovs_pcpu->owner = NULL;
316 local_unlock_nested_bh(&ovs_pcpu_storage->bh_lock);
317 }
318
319 stats_counter = &stats->n_hit;
320
321out:
322 /* Update datapath statistics. */
323 u64_stats_update_begin(&stats->syncp);
324 (*stats_counter)++;
325 stats->n_mask_hit += n_mask_hit;
326 stats->n_cache_hit += n_cache_hit;
327 u64_stats_update_end(&stats->syncp);
328}
329
330int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
331 const struct sw_flow_key *key,
332 const struct dp_upcall_info *upcall_info,
333 uint32_t cutlen)
334{
335 struct dp_stats_percpu *stats;
336 int err;
337
338 if (trace_ovs_dp_upcall_enabled())
339 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
340
341 if (upcall_info->portid == 0) {
342 err = -ENOTCONN;
343 goto err;
344 }
345
346 if (!skb_is_gso(skb))
347 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
348 else
349 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
350
351 ovs_vport_update_upcall_stats(skb, upcall_info, !err);
352 if (err)
353 goto err;
354
355 return 0;
356
357err:
358 stats = this_cpu_ptr(dp->stats_percpu);
359
360 u64_stats_update_begin(&stats->syncp);
361 stats->n_lost++;
362 u64_stats_update_end(&stats->syncp);
363
364 return err;
365}
366
367static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
368 const struct sw_flow_key *key,
369 const struct dp_upcall_info *upcall_info,
370 uint32_t cutlen)
371{
372 unsigned int gso_type = skb_shinfo(skb)->gso_type;
373 struct sw_flow_key later_key;
374 struct sk_buff *segs, *nskb;
375 int err;
376
377 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
378 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
379 if (IS_ERR(segs))
380 return PTR_ERR(segs);
381 if (segs == NULL)
382 return -EINVAL;
383
384 if (gso_type & SKB_GSO_UDP) {
385 /* The initial flow key extracted by ovs_flow_key_extract()
386 * in this case is for a first fragment, so we need to
387 * properly mark later fragments.
388 */
389 later_key = *key;
390 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
391 }
392
393 /* Queue all of the segments. */
394 skb_list_walk_safe(segs, skb, nskb) {
395 if (gso_type & SKB_GSO_UDP && skb != segs)
396 key = &later_key;
397
398 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
399 if (err)
400 break;
401
402 }
403
404 /* Free all of the segments. */
405 skb_list_walk_safe(segs, skb, nskb) {
406 if (err)
407 kfree_skb(skb);
408 else
409 consume_skb(skb);
410 }
411 return err;
412}
413
414static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
415 unsigned int hdrlen, int actions_attrlen)
416{
417 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
418 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
419 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
420 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
421 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
422
423 /* OVS_PACKET_ATTR_USERDATA */
424 if (upcall_info->userdata)
425 size += NLA_ALIGN(upcall_info->userdata->nla_len);
426
427 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
428 if (upcall_info->egress_tun_info)
429 size += nla_total_size(ovs_tun_key_attr_size());
430
431 /* OVS_PACKET_ATTR_ACTIONS */
432 if (upcall_info->actions_len)
433 size += nla_total_size(actions_attrlen);
434
435 /* OVS_PACKET_ATTR_MRU */
436 if (upcall_info->mru)
437 size += nla_total_size(sizeof(upcall_info->mru));
438
439 return size;
440}
441
442static void pad_packet(struct datapath *dp, struct sk_buff *skb)
443{
444 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
445 size_t plen = NLA_ALIGN(skb->len) - skb->len;
446
447 if (plen > 0)
448 skb_put_zero(skb, plen);
449 }
450}
451
452static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
453 const struct sw_flow_key *key,
454 const struct dp_upcall_info *upcall_info,
455 uint32_t cutlen)
456{
457 struct ovs_header *upcall;
458 struct sk_buff *nskb = NULL;
459 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
460 struct nlattr *nla;
461 size_t msg_size;
462 size_t skb_len;
463 unsigned int hlen;
464 int err, dp_ifindex;
465 u64 hash;
466
467 dp_ifindex = get_dpifindex(dp);
468 if (!dp_ifindex)
469 return -ENODEV;
470
471 if (!skb_frags_readable(skb))
472 return -EFAULT;
473
474 if (skb_vlan_tag_present(skb)) {
475 nskb = skb_clone(skb, GFP_ATOMIC);
476 if (!nskb)
477 return -ENOMEM;
478
479 nskb = __vlan_hwaccel_push_inside(nskb);
480 if (!nskb)
481 return -ENOMEM;
482
483 skb = nskb;
484 }
485
486 skb_len = min(skb->len, cutlen);
487 if (nla_attr_size(skb_len) > USHRT_MAX) {
488 err = -EFBIG;
489 goto out;
490 }
491
492 /* Complete checksum if needed */
493 if (skb->ip_summed == CHECKSUM_PARTIAL &&
494 (err = skb_csum_hwoffload_help(skb, 0)))
495 goto out;
496
497 /* Older versions of OVS user space enforce alignment of the last
498 * Netlink attribute to NLA_ALIGNTO which would require extensive
499 * padding logic. Only perform zerocopy if padding is not required.
500 */
501 if (dp->user_features & OVS_DP_F_UNALIGNED)
502 hlen = min(skb_zerocopy_headlen(skb), cutlen);
503 else
504 hlen = skb_len;
505
506 msg_size = upcall_msg_size(upcall_info, hlen,
507 OVS_CB(skb)->acts_origlen);
508 user_skb = genlmsg_new(msg_size, GFP_ATOMIC);
509 if (!user_skb) {
510 err = -ENOMEM;
511 goto out;
512 }
513
514 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
515 0, upcall_info->cmd);
516 if (!upcall) {
517 err = -EINVAL;
518 goto out;
519 }
520 upcall->dp_ifindex = dp_ifindex;
521
522 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
523 if (err)
524 goto out;
525
526 if (upcall_info->userdata)
527 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
528 nla_len(upcall_info->userdata),
529 nla_data(upcall_info->userdata));
530
531 if (upcall_info->egress_tun_info) {
532 nla = nla_nest_start_noflag(user_skb,
533 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
534 if (!nla) {
535 err = -EMSGSIZE;
536 goto out;
537 }
538 err = ovs_nla_put_tunnel_info(user_skb,
539 upcall_info->egress_tun_info);
540 if (err)
541 goto out;
542
543 nla_nest_end(user_skb, nla);
544 }
545
546 if (upcall_info->actions_len) {
547 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
548 if (!nla) {
549 err = -EMSGSIZE;
550 goto out;
551 }
552 err = ovs_nla_put_actions(upcall_info->actions,
553 upcall_info->actions_len,
554 user_skb);
555 if (!err)
556 nla_nest_end(user_skb, nla);
557 else
558 nla_nest_cancel(user_skb, nla);
559 }
560
561 /* Add OVS_PACKET_ATTR_MRU */
562 if (upcall_info->mru &&
563 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
564 err = -ENOBUFS;
565 goto out;
566 }
567
568 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
569 if (skb_len < skb->len &&
570 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
571 err = -ENOBUFS;
572 goto out;
573 }
574
575 /* Add OVS_PACKET_ATTR_HASH */
576 hash = skb_get_hash_raw(skb);
577 if (skb->sw_hash)
578 hash |= OVS_PACKET_HASH_SW_BIT;
579
580 if (skb->l4_hash)
581 hash |= OVS_PACKET_HASH_L4_BIT;
582
583 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
584 err = -ENOBUFS;
585 goto out;
586 }
587
588 /* Only reserve room for attribute header, packet data is added
589 * in skb_zerocopy() */
590 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
591 err = -ENOBUFS;
592 goto out;
593 }
594 nla->nla_len = nla_attr_size(skb_len);
595
596 err = skb_zerocopy(user_skb, skb, skb_len, hlen);
597 if (err)
598 goto out;
599
600 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
601 pad_packet(dp, user_skb);
602
603 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
604
605 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
606 user_skb = NULL;
607out:
608 consume_skb(user_skb);
609 consume_skb(nskb);
610
611 return err;
612}
613
614static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
615{
616 struct ovs_header *ovs_header = genl_info_userhdr(info);
617 struct net *net = sock_net(skb->sk);
618 struct nlattr **a = info->attrs;
619 struct sw_flow_actions *acts;
620 struct sk_buff *packet;
621 struct sw_flow *flow;
622 struct sw_flow_actions *sf_acts;
623 struct datapath *dp;
624 struct vport *input_vport;
625 u16 mru = 0;
626 u64 hash;
627 int len;
628 int err;
629 bool log = !a[OVS_PACKET_ATTR_PROBE];
630
631 err = -EINVAL;
632 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
633 !a[OVS_PACKET_ATTR_ACTIONS])
634 goto err;
635
636 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
637 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
638 err = -ENOMEM;
639 if (!packet)
640 goto err;
641 skb_reserve(packet, NET_IP_ALIGN);
642
643 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
644
645 /* Set packet's mru */
646 if (a[OVS_PACKET_ATTR_MRU]) {
647 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
648 packet->ignore_df = 1;
649 }
650 OVS_CB(packet)->mru = mru;
651 OVS_CB(packet)->cutlen = U32_MAX;
652
653 if (a[OVS_PACKET_ATTR_HASH]) {
654 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
655
656 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
657 !!(hash & OVS_PACKET_HASH_SW_BIT),
658 !!(hash & OVS_PACKET_HASH_L4_BIT));
659 }
660
661 OVS_CB(packet)->upcall_pid =
662 nla_get_u32_default(a[OVS_PACKET_ATTR_UPCALL_PID], 0);
663
664 /* Build an sw_flow for sending this packet. */
665 flow = ovs_flow_alloc();
666 err = PTR_ERR(flow);
667 if (IS_ERR(flow))
668 goto err_kfree_skb;
669
670 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
671 packet, &flow->key, log);
672 if (err)
673 goto err_flow_free;
674
675 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
676 &flow->key, &acts, log);
677 if (err)
678 goto err_flow_free;
679
680 rcu_assign_pointer(flow->sf_acts, acts);
681 packet->priority = flow->key.phy.priority;
682 packet->mark = flow->key.phy.skb_mark;
683
684 rcu_read_lock();
685 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
686 err = -ENODEV;
687 if (!dp)
688 goto err_unlock;
689
690 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
691 if (!input_vport)
692 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
693
694 if (!input_vport)
695 goto err_unlock;
696
697 packet->dev = input_vport->dev;
698 OVS_CB(packet)->input_vport = input_vport;
699 sf_acts = rcu_dereference(flow->sf_acts);
700
701 local_bh_disable();
702 local_lock_nested_bh(&ovs_pcpu_storage->bh_lock);
703 if (IS_ENABLED(CONFIG_PREEMPT_RT))
704 this_cpu_write(ovs_pcpu_storage->owner, current);
705 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
706 if (IS_ENABLED(CONFIG_PREEMPT_RT))
707 this_cpu_write(ovs_pcpu_storage->owner, NULL);
708 local_unlock_nested_bh(&ovs_pcpu_storage->bh_lock);
709 local_bh_enable();
710 rcu_read_unlock();
711
712 ovs_flow_free(flow, false);
713 return err;
714
715err_unlock:
716 rcu_read_unlock();
717err_flow_free:
718 ovs_flow_free(flow, false);
719err_kfree_skb:
720 kfree_skb(packet);
721err:
722 return err;
723}
724
725static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
726 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
727 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
728 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
729 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
730 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
731 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
732 [OVS_PACKET_ATTR_UPCALL_PID] = { .type = NLA_U32 },
733};
734
735static const struct genl_small_ops dp_packet_genl_ops[] = {
736 { .cmd = OVS_PACKET_CMD_EXECUTE,
737 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
738 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
739 .doit = ovs_packet_cmd_execute
740 }
741};
742
743static struct genl_family dp_packet_genl_family __ro_after_init = {
744 .hdrsize = sizeof(struct ovs_header),
745 .name = OVS_PACKET_FAMILY,
746 .version = OVS_PACKET_VERSION,
747 .maxattr = OVS_PACKET_ATTR_MAX,
748 .policy = packet_policy,
749 .netnsok = true,
750 .parallel_ops = true,
751 .small_ops = dp_packet_genl_ops,
752 .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
753 .resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
754 .module = THIS_MODULE,
755};
756
757static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
758 struct ovs_dp_megaflow_stats *mega_stats)
759{
760 int i;
761
762 memset(mega_stats, 0, sizeof(*mega_stats));
763
764 stats->n_flows = ovs_flow_tbl_count(&dp->table);
765 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
766
767 stats->n_hit = stats->n_missed = stats->n_lost = 0;
768
769 for_each_possible_cpu(i) {
770 const struct dp_stats_percpu *percpu_stats;
771 struct dp_stats_percpu local_stats;
772 unsigned int start;
773
774 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
775
776 do {
777 start = u64_stats_fetch_begin(&percpu_stats->syncp);
778 local_stats = *percpu_stats;
779 } while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
780
781 stats->n_hit += local_stats.n_hit;
782 stats->n_missed += local_stats.n_missed;
783 stats->n_lost += local_stats.n_lost;
784 mega_stats->n_mask_hit += local_stats.n_mask_hit;
785 mega_stats->n_cache_hit += local_stats.n_cache_hit;
786 }
787}
788
789static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
790{
791 return ovs_identifier_is_ufid(sfid) &&
792 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
793}
794
795static bool should_fill_mask(uint32_t ufid_flags)
796{
797 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
798}
799
800static bool should_fill_actions(uint32_t ufid_flags)
801{
802 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
803}
804
805static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
806 const struct sw_flow_id *sfid,
807 uint32_t ufid_flags)
808{
809 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
810
811 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
812 * see ovs_nla_put_identifier()
813 */
814 if (sfid && ovs_identifier_is_ufid(sfid))
815 len += nla_total_size(sfid->ufid_len);
816 else
817 len += nla_total_size(ovs_key_attr_size());
818
819 /* OVS_FLOW_ATTR_KEY */
820 if (!sfid || should_fill_key(sfid, ufid_flags))
821 len += nla_total_size(ovs_key_attr_size());
822
823 /* OVS_FLOW_ATTR_MASK */
824 if (should_fill_mask(ufid_flags))
825 len += nla_total_size(ovs_key_attr_size());
826
827 /* OVS_FLOW_ATTR_ACTIONS */
828 if (should_fill_actions(ufid_flags))
829 len += nla_total_size(acts->orig_len);
830
831 return len
832 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
833 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
834 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
835}
836
837/* Called with ovs_mutex or RCU read lock. */
838static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
839 struct sk_buff *skb)
840{
841 struct ovs_flow_stats stats;
842 __be16 tcp_flags;
843 unsigned long used;
844
845 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
846
847 if (used &&
848 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
849 OVS_FLOW_ATTR_PAD))
850 return -EMSGSIZE;
851
852 if (stats.n_packets &&
853 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
854 sizeof(struct ovs_flow_stats), &stats,
855 OVS_FLOW_ATTR_PAD))
856 return -EMSGSIZE;
857
858 if ((u8)ntohs(tcp_flags) &&
859 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
860 return -EMSGSIZE;
861
862 return 0;
863}
864
865/* Called with ovs_mutex or RCU read lock. */
866static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
867 struct sk_buff *skb, int skb_orig_len)
868{
869 struct nlattr *start;
870 int err;
871
872 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
873 * this is the first flow to be dumped into 'skb'. This is unusual for
874 * Netlink but individual action lists can be longer than
875 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
876 * The userspace caller can always fetch the actions separately if it
877 * really wants them. (Most userspace callers in fact don't care.)
878 *
879 * This can only fail for dump operations because the skb is always
880 * properly sized for single flows.
881 */
882 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
883 if (start) {
884 const struct sw_flow_actions *sf_acts;
885
886 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
887 err = ovs_nla_put_actions(sf_acts->actions,
888 sf_acts->actions_len, skb);
889
890 if (!err)
891 nla_nest_end(skb, start);
892 else {
893 if (skb_orig_len)
894 return err;
895
896 nla_nest_cancel(skb, start);
897 }
898 } else if (skb_orig_len) {
899 return -EMSGSIZE;
900 }
901
902 return 0;
903}
904
905/* Called with ovs_mutex or RCU read lock. */
906static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
907 struct sk_buff *skb, u32 portid,
908 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
909{
910 const int skb_orig_len = skb->len;
911 struct ovs_header *ovs_header;
912 int err;
913
914 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
915 flags, cmd);
916 if (!ovs_header)
917 return -EMSGSIZE;
918
919 ovs_header->dp_ifindex = dp_ifindex;
920
921 err = ovs_nla_put_identifier(flow, skb);
922 if (err)
923 goto error;
924
925 if (should_fill_key(&flow->id, ufid_flags)) {
926 err = ovs_nla_put_masked_key(flow, skb);
927 if (err)
928 goto error;
929 }
930
931 if (should_fill_mask(ufid_flags)) {
932 err = ovs_nla_put_mask(flow, skb);
933 if (err)
934 goto error;
935 }
936
937 err = ovs_flow_cmd_fill_stats(flow, skb);
938 if (err)
939 goto error;
940
941 if (should_fill_actions(ufid_flags)) {
942 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
943 if (err)
944 goto error;
945 }
946
947 genlmsg_end(skb, ovs_header);
948 return 0;
949
950error:
951 genlmsg_cancel(skb, ovs_header);
952 return err;
953}
954
955/* May not be called with RCU read lock. */
956static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
957 const struct sw_flow_id *sfid,
958 struct genl_info *info,
959 bool always,
960 uint32_t ufid_flags)
961{
962 struct sk_buff *skb;
963 size_t len;
964
965 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
966 return NULL;
967
968 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
969 skb = genlmsg_new(len, GFP_KERNEL);
970 if (!skb)
971 return ERR_PTR(-ENOMEM);
972
973 return skb;
974}
975
976/* Called with ovs_mutex. */
977static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
978 int dp_ifindex,
979 struct genl_info *info, u8 cmd,
980 bool always, u32 ufid_flags)
981{
982 struct sk_buff *skb;
983 int retval;
984
985 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
986 &flow->id, info, always, ufid_flags);
987 if (IS_ERR_OR_NULL(skb))
988 return skb;
989
990 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
991 info->snd_portid, info->snd_seq, 0,
992 cmd, ufid_flags);
993 if (WARN_ON_ONCE(retval < 0)) {
994 kfree_skb(skb);
995 skb = ERR_PTR(retval);
996 }
997 return skb;
998}
999
1000static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
1001{
1002 struct net *net = sock_net(skb->sk);
1003 struct nlattr **a = info->attrs;
1004 struct ovs_header *ovs_header = genl_info_userhdr(info);
1005 struct sw_flow *flow = NULL, *new_flow;
1006 struct sw_flow_mask mask;
1007 struct sk_buff *reply;
1008 struct datapath *dp;
1009 struct sw_flow_key *key;
1010 struct sw_flow_actions *acts;
1011 struct sw_flow_match match;
1012 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1013 int error;
1014 bool log = !a[OVS_FLOW_ATTR_PROBE];
1015
1016 /* Must have key and actions. */
1017 error = -EINVAL;
1018 if (!a[OVS_FLOW_ATTR_KEY]) {
1019 OVS_NLERR(log, "Flow key attr not present in new flow.");
1020 goto error;
1021 }
1022 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
1023 OVS_NLERR(log, "Flow actions attr not present in new flow.");
1024 goto error;
1025 }
1026
1027 /* Most of the time we need to allocate a new flow, do it before
1028 * locking.
1029 */
1030 new_flow = ovs_flow_alloc();
1031 if (IS_ERR(new_flow)) {
1032 error = PTR_ERR(new_flow);
1033 goto error;
1034 }
1035
1036 /* Extract key. */
1037 key = kzalloc_obj(*key);
1038 if (!key) {
1039 error = -ENOMEM;
1040 goto err_kfree_flow;
1041 }
1042
1043 ovs_match_init(&match, key, false, &mask);
1044 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1045 a[OVS_FLOW_ATTR_MASK], log);
1046 if (error)
1047 goto err_kfree_key;
1048
1049 ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1050
1051 /* Extract flow identifier. */
1052 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
1053 key, log);
1054 if (error)
1055 goto err_kfree_key;
1056
1057 /* Validate actions. */
1058 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1059 &new_flow->key, &acts, log);
1060 if (error) {
1061 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1062 goto err_kfree_key;
1063 }
1064
1065 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1066 ufid_flags);
1067 if (IS_ERR(reply)) {
1068 error = PTR_ERR(reply);
1069 goto err_kfree_acts;
1070 }
1071
1072 ovs_lock();
1073 dp = get_dp(net, ovs_header->dp_ifindex);
1074 if (unlikely(!dp)) {
1075 error = -ENODEV;
1076 goto err_unlock_ovs;
1077 }
1078
1079 /* Check if this is a duplicate flow */
1080 if (ovs_identifier_is_ufid(&new_flow->id))
1081 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1082 if (!flow)
1083 flow = ovs_flow_tbl_lookup(&dp->table, key);
1084 if (likely(!flow)) {
1085 rcu_assign_pointer(new_flow->sf_acts, acts);
1086
1087 /* Put flow in bucket. */
1088 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1089 if (unlikely(error)) {
1090 acts = NULL;
1091 goto err_unlock_ovs;
1092 }
1093
1094 if (unlikely(reply)) {
1095 error = ovs_flow_cmd_fill_info(new_flow,
1096 ovs_header->dp_ifindex,
1097 reply, info->snd_portid,
1098 info->snd_seq, 0,
1099 OVS_FLOW_CMD_NEW,
1100 ufid_flags);
1101 BUG_ON(error < 0);
1102 }
1103 ovs_unlock();
1104 } else {
1105 struct sw_flow_actions *old_acts;
1106
1107 /* Bail out if we're not allowed to modify an existing flow.
1108 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1109 * because Generic Netlink treats the latter as a dump
1110 * request. We also accept NLM_F_EXCL in case that bug ever
1111 * gets fixed.
1112 */
1113 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1114 | NLM_F_EXCL))) {
1115 error = -EEXIST;
1116 goto err_unlock_ovs;
1117 }
1118
1119 /* Look for any overlapping flow. */
1120 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1121 if (ovs_identifier_is_key(&flow->id))
1122 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1123 &match);
1124 else /* UFID matches but key is different */
1125 flow = NULL;
1126 if (!flow) {
1127 error = -ENOENT;
1128 goto err_unlock_ovs;
1129 }
1130 }
1131
1132 if (unlikely(reply)) {
1133 size_t cur, req;
1134
1135 cur = ovs_flow_cmd_msg_size(acts, &new_flow->id,
1136 ufid_flags);
1137 req = ovs_flow_cmd_msg_size(acts, &flow->id,
1138 ufid_flags);
1139 if (cur < req) {
1140 struct sk_buff *resized;
1141
1142 resized = ovs_flow_cmd_alloc_info(acts,
1143 &flow->id,
1144 info, false,
1145 ufid_flags);
1146 if (IS_ERR(resized)) {
1147 error = PTR_ERR(resized);
1148 goto err_unlock_ovs;
1149 }
1150 kfree_skb(reply);
1151 reply = resized;
1152 }
1153 }
1154
1155 /* Update actions. */
1156 old_acts = ovsl_dereference(flow->sf_acts);
1157 rcu_assign_pointer(flow->sf_acts, acts);
1158
1159 if (unlikely(reply)) {
1160 error = ovs_flow_cmd_fill_info(flow,
1161 ovs_header->dp_ifindex,
1162 reply, info->snd_portid,
1163 info->snd_seq, 0,
1164 OVS_FLOW_CMD_NEW,
1165 ufid_flags);
1166 BUG_ON(error < 0);
1167 }
1168 ovs_unlock();
1169
1170 ovs_nla_free_flow_actions_rcu(old_acts);
1171 ovs_flow_free(new_flow, false);
1172 }
1173
1174 if (reply)
1175 ovs_notify(&dp_flow_genl_family, reply, info);
1176
1177 kfree(key);
1178 return 0;
1179
1180err_unlock_ovs:
1181 ovs_unlock();
1182 kfree_skb(reply);
1183err_kfree_acts:
1184 ovs_nla_free_flow_actions(acts);
1185err_kfree_key:
1186 kfree(key);
1187err_kfree_flow:
1188 ovs_flow_free(new_flow, false);
1189error:
1190 return error;
1191}
1192
1193/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1194static noinline_for_stack
1195struct sw_flow_actions *get_flow_actions(struct net *net,
1196 const struct nlattr *a,
1197 const struct sw_flow_key *key,
1198 const struct sw_flow_mask *mask,
1199 bool log)
1200{
1201 struct sw_flow_actions *acts;
1202 struct sw_flow_key masked_key;
1203 int error;
1204
1205 ovs_flow_mask_key(&masked_key, key, true, mask);
1206 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1207 if (error) {
1208 OVS_NLERR(log,
1209 "Actions may not be safe on all matching packets");
1210 return ERR_PTR(error);
1211 }
1212
1213 return acts;
1214}
1215
1216/* Factor out match-init and action-copy to avoid
1217 * "Wframe-larger-than=1024" warning. Because mask is only
1218 * used to get actions, we new a function to save some
1219 * stack space.
1220 *
1221 * If there are not key and action attrs, we return 0
1222 * directly. In the case, the caller will also not use the
1223 * match as before. If there is action attr, we try to get
1224 * actions and save them to *acts. Before returning from
1225 * the function, we reset the match->mask pointer. Because
1226 * we should not to return match object with dangling reference
1227 * to mask.
1228 * */
1229static noinline_for_stack int
1230ovs_nla_init_match_and_action(struct net *net,
1231 struct sw_flow_match *match,
1232 struct sw_flow_key *key,
1233 struct nlattr **a,
1234 struct sw_flow_actions **acts,
1235 bool log)
1236{
1237 struct sw_flow_mask mask;
1238 int error = 0;
1239
1240 if (a[OVS_FLOW_ATTR_KEY]) {
1241 ovs_match_init(match, key, true, &mask);
1242 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1243 a[OVS_FLOW_ATTR_MASK], log);
1244 if (error)
1245 goto error;
1246 }
1247
1248 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1249 if (!a[OVS_FLOW_ATTR_KEY]) {
1250 OVS_NLERR(log,
1251 "Flow key attribute not present in set flow.");
1252 error = -EINVAL;
1253 goto error;
1254 }
1255
1256 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1257 &mask, log);
1258 if (IS_ERR(*acts)) {
1259 error = PTR_ERR(*acts);
1260 goto error;
1261 }
1262 }
1263
1264 /* On success, error is 0. */
1265error:
1266 match->mask = NULL;
1267 return error;
1268}
1269
1270static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1271{
1272 struct net *net = sock_net(skb->sk);
1273 struct nlattr **a = info->attrs;
1274 struct ovs_header *ovs_header = genl_info_userhdr(info);
1275 struct sw_flow_key key;
1276 struct sw_flow *flow;
1277 struct sk_buff *reply = NULL;
1278 struct datapath *dp;
1279 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1280 struct sw_flow_match match;
1281 struct sw_flow_id sfid;
1282 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1283 int error = 0;
1284 bool log = !a[OVS_FLOW_ATTR_PROBE];
1285 bool ufid_present;
1286
1287 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1288 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1289 OVS_NLERR(log,
1290 "Flow set message rejected, Key attribute missing.");
1291 return -EINVAL;
1292 }
1293
1294 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1295 &acts, log);
1296 if (error)
1297 goto error;
1298
1299 if (acts) {
1300 /* Can allocate before locking if have acts. */
1301 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1302 ufid_flags);
1303 if (IS_ERR(reply)) {
1304 error = PTR_ERR(reply);
1305 goto err_kfree_acts;
1306 }
1307 }
1308
1309 ovs_lock();
1310 dp = get_dp(net, ovs_header->dp_ifindex);
1311 if (unlikely(!dp)) {
1312 error = -ENODEV;
1313 goto err_unlock_ovs;
1314 }
1315 /* Check that the flow exists. */
1316 if (ufid_present)
1317 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1318 else
1319 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1320 if (unlikely(!flow)) {
1321 error = -ENOENT;
1322 goto err_unlock_ovs;
1323 }
1324
1325 /* Update actions, if present. */
1326 if (likely(acts)) {
1327 old_acts = ovsl_dereference(flow->sf_acts);
1328 rcu_assign_pointer(flow->sf_acts, acts);
1329
1330 if (unlikely(reply)) {
1331 error = ovs_flow_cmd_fill_info(flow,
1332 ovs_header->dp_ifindex,
1333 reply, info->snd_portid,
1334 info->snd_seq, 0,
1335 OVS_FLOW_CMD_SET,
1336 ufid_flags);
1337 BUG_ON(error < 0);
1338 }
1339 } else {
1340 /* Could not alloc without acts before locking. */
1341 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1342 info, OVS_FLOW_CMD_SET, false,
1343 ufid_flags);
1344
1345 if (IS_ERR(reply)) {
1346 error = PTR_ERR(reply);
1347 reply = NULL;
1348 goto err_unlock_ovs;
1349 }
1350 }
1351
1352 /* Clear stats. */
1353 if (a[OVS_FLOW_ATTR_CLEAR])
1354 ovs_flow_stats_clear(flow);
1355 ovs_unlock();
1356
1357 if (reply)
1358 ovs_notify(&dp_flow_genl_family, reply, info);
1359 if (old_acts)
1360 ovs_nla_free_flow_actions_rcu(old_acts);
1361
1362 return 0;
1363
1364err_unlock_ovs:
1365 ovs_unlock();
1366 kfree_skb(reply);
1367err_kfree_acts:
1368 ovs_nla_free_flow_actions(acts);
1369error:
1370 return error;
1371}
1372
1373static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1374{
1375 struct nlattr **a = info->attrs;
1376 struct ovs_header *ovs_header = genl_info_userhdr(info);
1377 struct net *net = sock_net(skb->sk);
1378 struct sw_flow_key key;
1379 struct sk_buff *reply;
1380 struct sw_flow *flow;
1381 struct datapath *dp;
1382 struct sw_flow_match match;
1383 struct sw_flow_id ufid;
1384 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1385 int err = 0;
1386 bool log = !a[OVS_FLOW_ATTR_PROBE];
1387 bool ufid_present;
1388
1389 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1390 if (a[OVS_FLOW_ATTR_KEY]) {
1391 ovs_match_init(&match, &key, true, NULL);
1392 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1393 log);
1394 } else if (!ufid_present) {
1395 OVS_NLERR(log,
1396 "Flow get message rejected, Key attribute missing.");
1397 err = -EINVAL;
1398 }
1399 if (err)
1400 return err;
1401
1402 ovs_lock();
1403 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1404 if (!dp) {
1405 err = -ENODEV;
1406 goto unlock;
1407 }
1408
1409 if (ufid_present)
1410 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1411 else
1412 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1413 if (!flow) {
1414 err = -ENOENT;
1415 goto unlock;
1416 }
1417
1418 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1419 OVS_FLOW_CMD_GET, true, ufid_flags);
1420 if (IS_ERR(reply)) {
1421 err = PTR_ERR(reply);
1422 goto unlock;
1423 }
1424
1425 ovs_unlock();
1426 return genlmsg_reply(reply, info);
1427unlock:
1428 ovs_unlock();
1429 return err;
1430}
1431
1432static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1433{
1434 struct nlattr **a = info->attrs;
1435 struct ovs_header *ovs_header = genl_info_userhdr(info);
1436 struct net *net = sock_net(skb->sk);
1437 struct sw_flow_key key;
1438 struct sk_buff *reply;
1439 struct sw_flow *flow = NULL;
1440 struct datapath *dp;
1441 struct sw_flow_match match;
1442 struct sw_flow_id ufid;
1443 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1444 int err;
1445 bool log = !a[OVS_FLOW_ATTR_PROBE];
1446 bool ufid_present;
1447
1448 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1449 if (a[OVS_FLOW_ATTR_KEY]) {
1450 ovs_match_init(&match, &key, true, NULL);
1451 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1452 NULL, log);
1453 if (unlikely(err))
1454 return err;
1455 }
1456
1457 ovs_lock();
1458 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1459 if (unlikely(!dp)) {
1460 err = -ENODEV;
1461 goto unlock;
1462 }
1463
1464 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1465 err = ovs_flow_tbl_flush(&dp->table);
1466 goto unlock;
1467 }
1468
1469 if (ufid_present)
1470 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1471 else
1472 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1473 if (unlikely(!flow)) {
1474 err = -ENOENT;
1475 goto unlock;
1476 }
1477
1478 reply = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
1479 &flow->id, info, false, ufid_flags);
1480 if (IS_ERR(reply)) {
1481 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1482 PTR_ERR(reply));
1483 reply = NULL;
1484 }
1485
1486 if (likely(reply)) {
1487 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1488 reply, info->snd_portid,
1489 info->snd_seq, 0,
1490 OVS_FLOW_CMD_DEL, ufid_flags);
1491 if (WARN_ON_ONCE(err < 0)) {
1492 kfree_skb(reply);
1493 reply = NULL;
1494 }
1495 }
1496 /* Removal has to happen after ovs_flow_cmd_fill_info(), as it uses
1497 * the flow->mask that can be scheduled to be freed by the
1498 * ovs_flow_tbl_remove() and we're not holding the RCU read lock.
1499 */
1500 ovs_flow_tbl_remove(&dp->table, flow);
1501 ovs_unlock();
1502
1503 if (likely(reply))
1504 ovs_notify(&dp_flow_genl_family, reply, info);
1505
1506 ovs_flow_free(flow, true);
1507 return 0;
1508unlock:
1509 ovs_unlock();
1510 return err;
1511}
1512
1513static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1514{
1515 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1516 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1517 struct table_instance *ti;
1518 struct datapath *dp;
1519 u32 ufid_flags;
1520 int err;
1521
1522 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1523 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1524 if (err)
1525 return err;
1526 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1527
1528 rcu_read_lock();
1529 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1530 if (!dp) {
1531 rcu_read_unlock();
1532 return -ENODEV;
1533 }
1534
1535 /*
1536 * Not needed for safety. Stops every spin_unlock_bh() in
1537 * ovs_flow_stats_get() from running the softirq backlog.
1538 */
1539 local_bh_disable();
1540
1541 ti = rcu_dereference(dp->table.ti);
1542 for (;;) {
1543 struct sw_flow *flow;
1544 u32 bucket, obj;
1545
1546 bucket = cb->args[0];
1547 obj = cb->args[1];
1548 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1549 if (!flow)
1550 break;
1551
1552 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1553 NETLINK_CB(cb->skb).portid,
1554 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1555 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1556 break;
1557
1558 cb->args[0] = bucket;
1559 cb->args[1] = obj;
1560 }
1561
1562 local_bh_enable();
1563 rcu_read_unlock();
1564 return skb->len;
1565}
1566
1567static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1568 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1569 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1570 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1571 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1572 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1573 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1574 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1575};
1576
1577static const struct genl_small_ops dp_flow_genl_ops[] = {
1578 { .cmd = OVS_FLOW_CMD_NEW,
1579 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1580 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1581 .doit = ovs_flow_cmd_new
1582 },
1583 { .cmd = OVS_FLOW_CMD_DEL,
1584 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1585 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1586 .doit = ovs_flow_cmd_del
1587 },
1588 { .cmd = OVS_FLOW_CMD_GET,
1589 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1590 .flags = 0, /* OK for unprivileged users. */
1591 .doit = ovs_flow_cmd_get,
1592 .dumpit = ovs_flow_cmd_dump
1593 },
1594 { .cmd = OVS_FLOW_CMD_SET,
1595 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1596 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1597 .doit = ovs_flow_cmd_set,
1598 },
1599};
1600
1601static struct genl_family dp_flow_genl_family __ro_after_init = {
1602 .hdrsize = sizeof(struct ovs_header),
1603 .name = OVS_FLOW_FAMILY,
1604 .version = OVS_FLOW_VERSION,
1605 .maxattr = OVS_FLOW_ATTR_MAX,
1606 .policy = flow_policy,
1607 .netnsok = true,
1608 .parallel_ops = true,
1609 .small_ops = dp_flow_genl_ops,
1610 .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1611 .resv_start_op = OVS_FLOW_CMD_SET + 1,
1612 .mcgrps = &ovs_dp_flow_multicast_group,
1613 .n_mcgrps = 1,
1614 .module = THIS_MODULE,
1615};
1616
1617static size_t ovs_dp_cmd_msg_size(void)
1618{
1619 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1620
1621 msgsize += nla_total_size(IFNAMSIZ);
1622 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1623 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1624 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1625 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1626 msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1627
1628 return msgsize;
1629}
1630
1631/* Called with ovs_mutex. */
1632static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1633 u32 portid, u32 seq, u32 flags, u8 cmd)
1634{
1635 struct ovs_header *ovs_header;
1636 struct ovs_dp_stats dp_stats;
1637 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1638 struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1639 int err, pids_len;
1640
1641 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1642 flags, cmd);
1643 if (!ovs_header)
1644 goto error;
1645
1646 ovs_header->dp_ifindex = get_dpifindex(dp);
1647
1648 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1649 if (err)
1650 goto nla_put_failure;
1651
1652 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1653 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1654 &dp_stats, OVS_DP_ATTR_PAD))
1655 goto nla_put_failure;
1656
1657 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1658 sizeof(struct ovs_dp_megaflow_stats),
1659 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1660 goto nla_put_failure;
1661
1662 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1663 goto nla_put_failure;
1664
1665 if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1666 ovs_flow_tbl_masks_cache_size(&dp->table)))
1667 goto nla_put_failure;
1668
1669 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1670 pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1671 if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1672 goto nla_put_failure;
1673 }
1674
1675 genlmsg_end(skb, ovs_header);
1676 return 0;
1677
1678nla_put_failure:
1679 genlmsg_cancel(skb, ovs_header);
1680error:
1681 return -EMSGSIZE;
1682}
1683
1684static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1685{
1686 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1687}
1688
1689/* Called with rcu_read_lock or ovs_mutex. */
1690static struct datapath *lookup_datapath(struct net *net,
1691 const struct ovs_header *ovs_header,
1692 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1693{
1694 struct datapath *dp;
1695
1696 if (!a[OVS_DP_ATTR_NAME])
1697 dp = get_dp(net, ovs_header->dp_ifindex);
1698 else {
1699 struct vport *vport;
1700
1701 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1702 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1703 }
1704 return dp ? dp : ERR_PTR(-ENODEV);
1705}
1706
1707static void ovs_dp_reset_user_features(struct sk_buff *skb,
1708 struct genl_info *info)
1709{
1710 struct datapath *dp;
1711
1712 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1713 info->attrs);
1714 if (IS_ERR(dp))
1715 return;
1716
1717 pr_warn("%s: Dropping previously announced user features\n",
1718 ovs_dp_name(dp));
1719 dp->user_features = 0;
1720}
1721
1722static int ovs_dp_set_upcall_portids(struct datapath *dp,
1723 const struct nlattr *ids)
1724{
1725 struct dp_nlsk_pids *old, *dp_nlsk_pids;
1726
1727 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1728 return -EINVAL;
1729
1730 old = ovsl_dereference(dp->upcall_portids);
1731
1732 dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1733 GFP_KERNEL);
1734 if (!dp_nlsk_pids)
1735 return -ENOMEM;
1736
1737 dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1738 nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1739
1740 rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1741
1742 kfree_rcu(old, rcu);
1743
1744 return 0;
1745}
1746
1747u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1748{
1749 struct dp_nlsk_pids *dp_nlsk_pids;
1750
1751 dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1752
1753 if (dp_nlsk_pids) {
1754 if (cpu_id < dp_nlsk_pids->n_pids) {
1755 return dp_nlsk_pids->pids[cpu_id];
1756 } else if (dp_nlsk_pids->n_pids > 0 &&
1757 cpu_id >= dp_nlsk_pids->n_pids) {
1758 /* If the number of netlink PIDs is mismatched with
1759 * the number of CPUs as seen by the kernel, log this
1760 * and send the upcall to an arbitrary socket (0) in
1761 * order to not drop packets
1762 */
1763 pr_info_ratelimited("cpu_id mismatch with handler threads");
1764 return dp_nlsk_pids->pids[cpu_id %
1765 dp_nlsk_pids->n_pids];
1766 } else {
1767 return 0;
1768 }
1769 } else {
1770 return 0;
1771 }
1772}
1773
1774static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1775{
1776 u32 user_features = 0, old_features = dp->user_features;
1777 int err;
1778
1779 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1780 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1781
1782 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1783 OVS_DP_F_UNALIGNED |
1784 OVS_DP_F_TC_RECIRC_SHARING |
1785 OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1786 return -EOPNOTSUPP;
1787
1788#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1789 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1790 return -EOPNOTSUPP;
1791#endif
1792 }
1793
1794 if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1795 int err;
1796 u32 cache_size;
1797
1798 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1799 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1800 if (err)
1801 return err;
1802 }
1803
1804 dp->user_features = user_features;
1805
1806 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1807 a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1808 /* Upcall Netlink Port IDs have been updated */
1809 err = ovs_dp_set_upcall_portids(dp,
1810 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1811 if (err)
1812 return err;
1813 }
1814
1815 if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1816 !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1817 tc_skb_ext_tc_enable();
1818 else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1819 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1820 tc_skb_ext_tc_disable();
1821
1822 return 0;
1823}
1824
1825static int ovs_dp_stats_init(struct datapath *dp)
1826{
1827 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1828 if (!dp->stats_percpu)
1829 return -ENOMEM;
1830
1831 return 0;
1832}
1833
1834static int ovs_dp_vport_init(struct datapath *dp)
1835{
1836 int i;
1837
1838 dp->ports = kmalloc_objs(struct hlist_head, DP_VPORT_HASH_BUCKETS);
1839 if (!dp->ports)
1840 return -ENOMEM;
1841
1842 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1843 INIT_HLIST_HEAD(&dp->ports[i]);
1844
1845 return 0;
1846}
1847
1848static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1849{
1850 struct nlattr **a = info->attrs;
1851 struct vport_parms parms;
1852 struct sk_buff *reply;
1853 struct datapath *dp;
1854 struct vport *vport;
1855 struct ovs_net *ovs_net;
1856 int err;
1857
1858 err = -EINVAL;
1859 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1860 goto err;
1861
1862 reply = ovs_dp_cmd_alloc_info();
1863 if (!reply)
1864 return -ENOMEM;
1865
1866 err = -ENOMEM;
1867 dp = kzalloc_obj(*dp);
1868 if (dp == NULL)
1869 goto err_destroy_reply;
1870
1871 ovs_dp_set_net(dp, sock_net(skb->sk));
1872
1873 /* Allocate table. */
1874 err = ovs_flow_tbl_init(&dp->table);
1875 if (err)
1876 goto err_destroy_dp;
1877
1878 err = ovs_dp_stats_init(dp);
1879 if (err)
1880 goto err_destroy_table;
1881
1882 err = ovs_dp_vport_init(dp);
1883 if (err)
1884 goto err_destroy_stats;
1885
1886 err = ovs_meters_init(dp);
1887 if (err)
1888 goto err_destroy_ports;
1889
1890 /* Set up our datapath device. */
1891 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1892 parms.type = OVS_VPORT_TYPE_INTERNAL;
1893 parms.dp = dp;
1894 parms.port_no = OVSP_LOCAL;
1895 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1896 parms.desired_ifindex = nla_get_s32_default(a[OVS_DP_ATTR_IFINDEX], 0);
1897
1898 /* So far only local changes have been made, now need the lock. */
1899 ovs_lock();
1900
1901 err = ovs_dp_change(dp, a);
1902 if (err)
1903 goto err_unlock_and_destroy_meters;
1904
1905 vport = new_vport(&parms);
1906 if (IS_ERR(vport)) {
1907 err = PTR_ERR(vport);
1908 if (err == -EBUSY)
1909 err = -EEXIST;
1910
1911 if (err == -EEXIST) {
1912 /* An outdated user space instance that does not understand
1913 * the concept of user_features has attempted to create a new
1914 * datapath and is likely to reuse it. Drop all user features.
1915 */
1916 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1917 ovs_dp_reset_user_features(skb, info);
1918 }
1919
1920 goto err_destroy_portids;
1921 }
1922
1923 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1924 info->snd_seq, 0, OVS_DP_CMD_NEW);
1925 BUG_ON(err < 0);
1926
1927 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1928 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1929
1930 ovs_unlock();
1931
1932 ovs_notify(&dp_datapath_genl_family, reply, info);
1933 return 0;
1934
1935err_destroy_portids:
1936 kfree(rcu_dereference_raw(dp->upcall_portids));
1937err_unlock_and_destroy_meters:
1938 ovs_unlock();
1939 ovs_meters_exit(dp);
1940err_destroy_ports:
1941 kfree(dp->ports);
1942err_destroy_stats:
1943 free_percpu(dp->stats_percpu);
1944err_destroy_table:
1945 ovs_flow_tbl_destroy(&dp->table);
1946err_destroy_dp:
1947 kfree(dp);
1948err_destroy_reply:
1949 kfree_skb(reply);
1950err:
1951 return err;
1952}
1953
1954/* Called with ovs_mutex. */
1955static void __dp_destroy(struct datapath *dp)
1956{
1957 struct flow_table *table = &dp->table;
1958 int i;
1959
1960 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1961 tc_skb_ext_tc_disable();
1962
1963 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1964 struct vport *vport;
1965 struct hlist_node *n;
1966
1967 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1968 if (vport->port_no != OVSP_LOCAL)
1969 ovs_dp_detach_port(vport);
1970 }
1971
1972 list_del_rcu(&dp->list_node);
1973
1974 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1975 * all ports in datapath are destroyed first before freeing datapath.
1976 */
1977 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1978
1979 /* Flush sw_flow in the tables. RCU cb only releases resource
1980 * such as dp, ports and tables. That may avoid some issues
1981 * such as RCU usage warning.
1982 */
1983 table_instance_flow_flush(table, ovsl_dereference(table->ti),
1984 ovsl_dereference(table->ufid_ti));
1985
1986 /* RCU destroy the ports, meters and flow tables. */
1987 call_rcu(&dp->rcu, destroy_dp_rcu);
1988}
1989
1990static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1991{
1992 struct sk_buff *reply;
1993 struct datapath *dp;
1994 int err;
1995
1996 reply = ovs_dp_cmd_alloc_info();
1997 if (!reply)
1998 return -ENOMEM;
1999
2000 ovs_lock();
2001 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2002 info->attrs);
2003 err = PTR_ERR(dp);
2004 if (IS_ERR(dp))
2005 goto err_unlock_free;
2006
2007 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2008 info->snd_seq, 0, OVS_DP_CMD_DEL);
2009 BUG_ON(err < 0);
2010
2011 __dp_destroy(dp);
2012 ovs_unlock();
2013
2014 ovs_notify(&dp_datapath_genl_family, reply, info);
2015
2016 return 0;
2017
2018err_unlock_free:
2019 ovs_unlock();
2020 kfree_skb(reply);
2021 return err;
2022}
2023
2024static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
2025{
2026 struct sk_buff *reply;
2027 struct datapath *dp;
2028 int err;
2029
2030 reply = ovs_dp_cmd_alloc_info();
2031 if (!reply)
2032 return -ENOMEM;
2033
2034 ovs_lock();
2035 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2036 info->attrs);
2037 err = PTR_ERR(dp);
2038 if (IS_ERR(dp))
2039 goto err_unlock_free;
2040
2041 err = ovs_dp_change(dp, info->attrs);
2042 if (err)
2043 goto err_unlock_free;
2044
2045 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2046 info->snd_seq, 0, OVS_DP_CMD_SET);
2047 BUG_ON(err < 0);
2048
2049 ovs_unlock();
2050 ovs_notify(&dp_datapath_genl_family, reply, info);
2051
2052 return 0;
2053
2054err_unlock_free:
2055 ovs_unlock();
2056 kfree_skb(reply);
2057 return err;
2058}
2059
2060static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
2061{
2062 struct sk_buff *reply;
2063 struct datapath *dp;
2064 int err;
2065
2066 reply = ovs_dp_cmd_alloc_info();
2067 if (!reply)
2068 return -ENOMEM;
2069
2070 ovs_lock();
2071 dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2072 info->attrs);
2073 if (IS_ERR(dp)) {
2074 err = PTR_ERR(dp);
2075 goto err_unlock_free;
2076 }
2077 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2078 info->snd_seq, 0, OVS_DP_CMD_GET);
2079 BUG_ON(err < 0);
2080 ovs_unlock();
2081
2082 return genlmsg_reply(reply, info);
2083
2084err_unlock_free:
2085 ovs_unlock();
2086 kfree_skb(reply);
2087 return err;
2088}
2089
2090static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2091{
2092 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2093 struct datapath *dp;
2094 int skip = cb->args[0];
2095 int i = 0;
2096
2097 ovs_lock();
2098 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2099 if (i >= skip &&
2100 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2101 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2102 OVS_DP_CMD_GET) < 0)
2103 break;
2104 i++;
2105 }
2106 ovs_unlock();
2107
2108 cb->args[0] = i;
2109
2110 return skb->len;
2111}
2112
2113static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2114 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2115 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2116 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2117 [OVS_DP_ATTR_MASKS_CACHE_SIZE] = NLA_POLICY_RANGE(NLA_U32, 0,
2118 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2119 [OVS_DP_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2120};
2121
2122static const struct genl_small_ops dp_datapath_genl_ops[] = {
2123 { .cmd = OVS_DP_CMD_NEW,
2124 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2125 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2126 .doit = ovs_dp_cmd_new
2127 },
2128 { .cmd = OVS_DP_CMD_DEL,
2129 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2130 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2131 .doit = ovs_dp_cmd_del
2132 },
2133 { .cmd = OVS_DP_CMD_GET,
2134 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2135 .flags = 0, /* OK for unprivileged users. */
2136 .doit = ovs_dp_cmd_get,
2137 .dumpit = ovs_dp_cmd_dump
2138 },
2139 { .cmd = OVS_DP_CMD_SET,
2140 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2141 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2142 .doit = ovs_dp_cmd_set,
2143 },
2144};
2145
2146static struct genl_family dp_datapath_genl_family __ro_after_init = {
2147 .hdrsize = sizeof(struct ovs_header),
2148 .name = OVS_DATAPATH_FAMILY,
2149 .version = OVS_DATAPATH_VERSION,
2150 .maxattr = OVS_DP_ATTR_MAX,
2151 .policy = datapath_policy,
2152 .netnsok = true,
2153 .parallel_ops = true,
2154 .small_ops = dp_datapath_genl_ops,
2155 .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2156 .resv_start_op = OVS_DP_CMD_SET + 1,
2157 .mcgrps = &ovs_dp_datapath_multicast_group,
2158 .n_mcgrps = 1,
2159 .module = THIS_MODULE,
2160};
2161
2162/* Called with ovs_mutex or RCU read lock. */
2163static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2164 struct net *net, u32 portid, u32 seq,
2165 u32 flags, u8 cmd, gfp_t gfp)
2166{
2167 struct ovs_header *ovs_header;
2168 struct ovs_vport_stats vport_stats;
2169 struct net *net_vport;
2170 int err;
2171
2172 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2173 flags, cmd);
2174 if (!ovs_header)
2175 return -EMSGSIZE;
2176
2177 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2178
2179 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2180 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2181 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2182 ovs_vport_name(vport)) ||
2183 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2184 goto nla_put_failure;
2185
2186 rcu_read_lock();
2187 net_vport = dev_net_rcu(vport->dev);
2188 if (!net_eq(net, net_vport)) {
2189 int id = peernet2id_alloc(net, net_vport, GFP_ATOMIC);
2190
2191 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2192 goto nla_put_failure_unlock;
2193 }
2194 rcu_read_unlock();
2195
2196 ovs_vport_get_stats(vport, &vport_stats);
2197 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2198 sizeof(struct ovs_vport_stats), &vport_stats,
2199 OVS_VPORT_ATTR_PAD))
2200 goto nla_put_failure;
2201
2202 if (ovs_vport_get_upcall_stats(vport, skb))
2203 goto nla_put_failure;
2204
2205 if (ovs_vport_get_upcall_portids(vport, skb))
2206 goto nla_put_failure;
2207
2208 genlmsg_end(skb, ovs_header);
2209 return 0;
2210
2211nla_put_failure_unlock:
2212 rcu_read_unlock();
2213nla_put_failure:
2214 err = -EMSGSIZE;
2215 genlmsg_cancel(skb, ovs_header);
2216 return err;
2217}
2218
2219static size_t ovs_vport_cmd_msg_size(void)
2220{
2221 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
2222
2223 msgsize += nla_total_size(sizeof(u32)); /* OVS_VPORT_ATTR_PORT_NO */
2224 msgsize += nla_total_size(sizeof(u32)); /* OVS_VPORT_ATTR_TYPE */
2225 msgsize += nla_total_size(IFNAMSIZ); /* OVS_VPORT_ATTR_NAME */
2226 msgsize += nla_total_size(sizeof(u32)); /* OVS_VPORT_ATTR_IFINDEX */
2227 msgsize += nla_total_size(sizeof(s32)); /* OVS_VPORT_ATTR_NETNSID */
2228
2229 /* OVS_VPORT_ATTR_STATS */
2230 msgsize += nla_total_size_64bit(sizeof(struct ovs_vport_stats));
2231
2232 /* OVS_VPORT_ATTR_UPCALL_STATS(OVS_VPORT_UPCALL_ATTR_SUCCESS +
2233 * OVS_VPORT_UPCALL_ATTR_FAIL)
2234 */
2235 msgsize += nla_total_size(nla_total_size_64bit(sizeof(u64)) +
2236 nla_total_size_64bit(sizeof(u64)));
2237
2238 /* OVS_VPORT_ATTR_UPCALL_PID */
2239 msgsize += nla_total_size(nr_cpu_ids * sizeof(u32));
2240
2241 return msgsize;
2242}
2243
2244static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2245{
2246 return genlmsg_new(ovs_vport_cmd_msg_size(), GFP_KERNEL);
2247}
2248
2249/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2250struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2251 u32 portid, u32 seq, u8 cmd)
2252{
2253 struct sk_buff *skb;
2254 int retval;
2255
2256 skb = ovs_vport_cmd_alloc_info();
2257 if (!skb)
2258 return ERR_PTR(-ENOMEM);
2259
2260 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2261 GFP_KERNEL);
2262 BUG_ON(retval < 0);
2263
2264 return skb;
2265}
2266
2267/* Called with ovs_mutex or RCU read lock. */
2268static struct vport *lookup_vport(struct net *net,
2269 const struct ovs_header *ovs_header,
2270 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2271{
2272 struct datapath *dp;
2273 struct vport *vport;
2274
2275 if (a[OVS_VPORT_ATTR_IFINDEX])
2276 return ERR_PTR(-EOPNOTSUPP);
2277 if (a[OVS_VPORT_ATTR_NAME]) {
2278 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2279 if (!vport)
2280 return ERR_PTR(-ENODEV);
2281 if (ovs_header->dp_ifindex &&
2282 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2283 return ERR_PTR(-ENODEV);
2284 return vport;
2285 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2286 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2287
2288 if (port_no >= DP_MAX_PORTS)
2289 return ERR_PTR(-EFBIG);
2290
2291 dp = get_dp(net, ovs_header->dp_ifindex);
2292 if (!dp)
2293 return ERR_PTR(-ENODEV);
2294
2295 vport = ovs_vport_ovsl_rcu(dp, port_no);
2296 if (!vport)
2297 return ERR_PTR(-ENODEV);
2298 return vport;
2299 } else
2300 return ERR_PTR(-EINVAL);
2301
2302}
2303
2304static unsigned int ovs_get_max_headroom(struct datapath *dp)
2305{
2306 unsigned int dev_headroom, max_headroom = 0;
2307 struct net_device *dev;
2308 struct vport *vport;
2309 int i;
2310
2311 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2312 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2313 lockdep_ovsl_is_held()) {
2314 dev = vport->dev;
2315 dev_headroom = netdev_get_fwd_headroom(dev);
2316 if (dev_headroom > max_headroom)
2317 max_headroom = dev_headroom;
2318 }
2319 }
2320
2321 return max_headroom;
2322}
2323
2324/* Called with ovs_mutex */
2325static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2326{
2327 struct vport *vport;
2328 int i;
2329
2330 dp->max_headroom = new_headroom;
2331 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2332 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2333 lockdep_ovsl_is_held())
2334 netdev_set_rx_headroom(vport->dev, new_headroom);
2335 }
2336}
2337
2338static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2339{
2340 struct nlattr **a = info->attrs;
2341 struct ovs_header *ovs_header = genl_info_userhdr(info);
2342 struct vport_parms parms;
2343 struct sk_buff *reply;
2344 struct vport *vport;
2345 struct datapath *dp;
2346 unsigned int new_headroom;
2347 u32 port_no;
2348 int err;
2349
2350 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2351 !a[OVS_VPORT_ATTR_UPCALL_PID])
2352 return -EINVAL;
2353
2354 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2355
2356 if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2357 return -EOPNOTSUPP;
2358
2359 port_no = nla_get_u32_default(a[OVS_VPORT_ATTR_PORT_NO], 0);
2360 if (port_no >= DP_MAX_PORTS)
2361 return -EFBIG;
2362
2363 reply = ovs_vport_cmd_alloc_info();
2364 if (!reply)
2365 return -ENOMEM;
2366
2367 ovs_lock();
2368 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2369 err = -ENODEV;
2370 if (!dp)
2371 goto exit_unlock_free;
2372
2373 if (port_no) {
2374 vport = ovs_vport_ovsl(dp, port_no);
2375 err = -EBUSY;
2376 if (vport)
2377 goto exit_unlock_free;
2378 } else {
2379 for (port_no = 1; ; port_no++) {
2380 if (port_no >= DP_MAX_PORTS) {
2381 err = -EFBIG;
2382 goto exit_unlock_free;
2383 }
2384 vport = ovs_vport_ovsl(dp, port_no);
2385 if (!vport)
2386 break;
2387 }
2388 }
2389
2390 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2391 parms.dp = dp;
2392 parms.port_no = port_no;
2393 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2394 parms.desired_ifindex = nla_get_s32_default(a[OVS_VPORT_ATTR_IFINDEX],
2395 0);
2396
2397 vport = new_vport(&parms);
2398 err = PTR_ERR(vport);
2399 if (IS_ERR(vport))
2400 goto exit_unlock_free;
2401
2402 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2403 info->snd_portid, info->snd_seq, 0,
2404 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2405
2406 new_headroom = netdev_get_fwd_headroom(vport->dev);
2407
2408 if (new_headroom > dp->max_headroom)
2409 ovs_update_headroom(dp, new_headroom);
2410 else
2411 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2412
2413 BUG_ON(err < 0);
2414 ovs_unlock();
2415
2416 ovs_notify(&dp_vport_genl_family, reply, info);
2417 return 0;
2418
2419exit_unlock_free:
2420 ovs_unlock();
2421 kfree_skb(reply);
2422 return err;
2423}
2424
2425static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2426{
2427 struct nlattr **a = info->attrs;
2428 struct sk_buff *reply;
2429 struct vport *vport;
2430 int err;
2431
2432 reply = ovs_vport_cmd_alloc_info();
2433 if (!reply)
2434 return -ENOMEM;
2435
2436 ovs_lock();
2437 vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2438 err = PTR_ERR(vport);
2439 if (IS_ERR(vport))
2440 goto exit_unlock_free;
2441
2442 if (a[OVS_VPORT_ATTR_TYPE] &&
2443 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2444 err = -EINVAL;
2445 goto exit_unlock_free;
2446 }
2447
2448 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2449 /* There are no vport types that support legacy options. */
2450 err = -EOPNOTSUPP;
2451 goto exit_unlock_free;
2452 }
2453
2454 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2455 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2456
2457 err = ovs_vport_set_upcall_portids(vport, ids);
2458 if (err)
2459 goto exit_unlock_free;
2460 }
2461
2462 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2463 info->snd_portid, info->snd_seq, 0,
2464 OVS_VPORT_CMD_SET, GFP_KERNEL);
2465 BUG_ON(err < 0);
2466
2467 ovs_unlock();
2468 ovs_notify(&dp_vport_genl_family, reply, info);
2469 return 0;
2470
2471exit_unlock_free:
2472 ovs_unlock();
2473 kfree_skb(reply);
2474 return err;
2475}
2476
2477static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2478{
2479 bool update_headroom = false;
2480 struct nlattr **a = info->attrs;
2481 struct sk_buff *reply;
2482 struct datapath *dp;
2483 struct vport *vport;
2484 unsigned int new_headroom;
2485 int err;
2486
2487 reply = ovs_vport_cmd_alloc_info();
2488 if (!reply)
2489 return -ENOMEM;
2490
2491 ovs_lock();
2492 vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2493 err = PTR_ERR(vport);
2494 if (IS_ERR(vport))
2495 goto exit_unlock_free;
2496
2497 if (vport->port_no == OVSP_LOCAL) {
2498 err = -EINVAL;
2499 goto exit_unlock_free;
2500 }
2501
2502 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2503 info->snd_portid, info->snd_seq, 0,
2504 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2505 BUG_ON(err < 0);
2506
2507 /* the vport deletion may trigger dp headroom update */
2508 dp = vport->dp;
2509 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2510 update_headroom = true;
2511
2512 netdev_reset_rx_headroom(vport->dev);
2513 ovs_dp_detach_port(vport);
2514
2515 if (update_headroom) {
2516 new_headroom = ovs_get_max_headroom(dp);
2517
2518 if (new_headroom < dp->max_headroom)
2519 ovs_update_headroom(dp, new_headroom);
2520 }
2521 ovs_unlock();
2522
2523 ovs_notify(&dp_vport_genl_family, reply, info);
2524 return 0;
2525
2526exit_unlock_free:
2527 ovs_unlock();
2528 kfree_skb(reply);
2529 return err;
2530}
2531
2532static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2533{
2534 struct nlattr **a = info->attrs;
2535 struct ovs_header *ovs_header = genl_info_userhdr(info);
2536 struct sk_buff *reply;
2537 struct vport *vport;
2538 int err;
2539
2540 reply = ovs_vport_cmd_alloc_info();
2541 if (!reply)
2542 return -ENOMEM;
2543
2544 rcu_read_lock();
2545 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2546 err = PTR_ERR(vport);
2547 if (IS_ERR(vport))
2548 goto exit_unlock_free;
2549 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2550 info->snd_portid, info->snd_seq, 0,
2551 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2552 BUG_ON(err < 0);
2553 rcu_read_unlock();
2554
2555 return genlmsg_reply(reply, info);
2556
2557exit_unlock_free:
2558 rcu_read_unlock();
2559 kfree_skb(reply);
2560 return err;
2561}
2562
2563static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2564{
2565 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2566 struct datapath *dp;
2567 int bucket = cb->args[0], skip = cb->args[1];
2568 int i, j = 0;
2569
2570 rcu_read_lock();
2571 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2572 if (!dp) {
2573 rcu_read_unlock();
2574 return -ENODEV;
2575 }
2576 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2577 struct vport *vport;
2578
2579 j = 0;
2580 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2581 if (j >= skip &&
2582 ovs_vport_cmd_fill_info(vport, skb,
2583 sock_net(skb->sk),
2584 NETLINK_CB(cb->skb).portid,
2585 cb->nlh->nlmsg_seq,
2586 NLM_F_MULTI,
2587 OVS_VPORT_CMD_GET,
2588 GFP_ATOMIC) < 0)
2589 goto out;
2590
2591 j++;
2592 }
2593 skip = 0;
2594 }
2595out:
2596 rcu_read_unlock();
2597
2598 cb->args[0] = i;
2599 cb->args[1] = j;
2600
2601 return skb->len;
2602}
2603
2604static void ovs_dp_masks_rebalance(struct work_struct *work)
2605{
2606 struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2607 masks_rebalance.work);
2608 struct datapath *dp;
2609
2610 ovs_lock();
2611
2612 list_for_each_entry(dp, &ovs_net->dps, list_node)
2613 ovs_flow_masks_rebalance(&dp->table);
2614
2615 ovs_unlock();
2616
2617 schedule_delayed_work(&ovs_net->masks_rebalance,
2618 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2619}
2620
2621static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2622 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2623 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2624 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2625 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2626 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2627 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, /* Unused. */
2628 [OVS_VPORT_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2629 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2630 [OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
2631};
2632
2633static const struct genl_small_ops dp_vport_genl_ops[] = {
2634 { .cmd = OVS_VPORT_CMD_NEW,
2635 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2636 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2637 .doit = ovs_vport_cmd_new
2638 },
2639 { .cmd = OVS_VPORT_CMD_DEL,
2640 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2641 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2642 .doit = ovs_vport_cmd_del
2643 },
2644 { .cmd = OVS_VPORT_CMD_GET,
2645 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2646 .flags = 0, /* OK for unprivileged users. */
2647 .doit = ovs_vport_cmd_get,
2648 .dumpit = ovs_vport_cmd_dump
2649 },
2650 { .cmd = OVS_VPORT_CMD_SET,
2651 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2652 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2653 .doit = ovs_vport_cmd_set,
2654 },
2655};
2656
2657struct genl_family dp_vport_genl_family __ro_after_init = {
2658 .hdrsize = sizeof(struct ovs_header),
2659 .name = OVS_VPORT_FAMILY,
2660 .version = OVS_VPORT_VERSION,
2661 .maxattr = OVS_VPORT_ATTR_MAX,
2662 .policy = vport_policy,
2663 .netnsok = true,
2664 .parallel_ops = true,
2665 .small_ops = dp_vport_genl_ops,
2666 .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2667 .resv_start_op = OVS_VPORT_CMD_SET + 1,
2668 .mcgrps = &ovs_dp_vport_multicast_group,
2669 .n_mcgrps = 1,
2670 .module = THIS_MODULE,
2671};
2672
2673static struct genl_family * const dp_genl_families[] = {
2674 &dp_datapath_genl_family,
2675 &dp_vport_genl_family,
2676 &dp_flow_genl_family,
2677 &dp_packet_genl_family,
2678 &dp_meter_genl_family,
2679#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2680 &dp_ct_limit_genl_family,
2681#endif
2682};
2683
2684static void dp_unregister_genl(int n_families)
2685{
2686 int i;
2687
2688 for (i = 0; i < n_families; i++)
2689 genl_unregister_family(dp_genl_families[i]);
2690}
2691
2692static int __init dp_register_genl(void)
2693{
2694 int err;
2695 int i;
2696
2697 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2698
2699 err = genl_register_family(dp_genl_families[i]);
2700 if (err)
2701 goto error;
2702 }
2703
2704 return 0;
2705
2706error:
2707 dp_unregister_genl(i);
2708 return err;
2709}
2710
2711static int __net_init ovs_init_net(struct net *net)
2712{
2713 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2714 int err;
2715
2716 INIT_LIST_HEAD(&ovs_net->dps);
2717 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2718 INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2719
2720 err = ovs_ct_init(net);
2721 if (err)
2722 return err;
2723
2724 schedule_delayed_work(&ovs_net->masks_rebalance,
2725 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2726 return 0;
2727}
2728
2729static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2730 struct list_head *head)
2731{
2732 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2733 struct datapath *dp;
2734
2735 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2736 int i;
2737
2738 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2739 struct vport *vport;
2740
2741 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2742 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2743 continue;
2744
2745 if (dev_net(vport->dev) == dnet)
2746 list_add(&vport->detach_list, head);
2747 }
2748 }
2749 }
2750}
2751
2752static void __net_exit ovs_pre_exit_net(struct net *dnet)
2753{
2754 ovs_lock();
2755 ovs_ct_exit_start(dnet);
2756 ovs_unlock();
2757}
2758
2759static void __net_exit ovs_exit_net(struct net *dnet)
2760{
2761 struct datapath *dp, *dp_next;
2762 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2763 struct vport *vport, *vport_next;
2764 struct net *net;
2765 LIST_HEAD(head);
2766
2767 ovs_lock();
2768
2769 ovs_ct_exit_finish(dnet);
2770
2771 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2772 __dp_destroy(dp);
2773
2774 down_read(&net_rwsem);
2775 for_each_net(net)
2776 list_vports_from_net(net, dnet, &head);
2777 up_read(&net_rwsem);
2778
2779 /* Detach all vports from given namespace. */
2780 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2781 list_del(&vport->detach_list);
2782 ovs_dp_detach_port(vport);
2783 }
2784
2785 ovs_unlock();
2786
2787 cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2788 cancel_work_sync(&ovs_net->dp_notify_work);
2789}
2790
2791static struct pernet_operations ovs_net_ops = {
2792 .init = ovs_init_net,
2793 .pre_exit = ovs_pre_exit_net,
2794 .exit = ovs_exit_net,
2795 .id = &ovs_net_id,
2796 .size = sizeof(struct ovs_net),
2797};
2798
2799static const char * const ovs_drop_reasons[] = {
2800#define S(x) [(x) & ~SKB_DROP_REASON_SUBSYS_MASK] = (#x),
2801 OVS_DROP_REASONS(S)
2802#undef S
2803};
2804
2805static struct drop_reason_list drop_reason_list_ovs = {
2806 .reasons = ovs_drop_reasons,
2807 .n_reasons = ARRAY_SIZE(ovs_drop_reasons),
2808};
2809
2810static int __init ovs_alloc_percpu_storage(void)
2811{
2812 unsigned int cpu;
2813
2814 ovs_pcpu_storage = alloc_percpu(*ovs_pcpu_storage);
2815 if (!ovs_pcpu_storage)
2816 return -ENOMEM;
2817
2818 for_each_possible_cpu(cpu) {
2819 struct ovs_pcpu_storage *ovs_pcpu;
2820
2821 ovs_pcpu = per_cpu_ptr(ovs_pcpu_storage, cpu);
2822 local_lock_init(&ovs_pcpu->bh_lock);
2823 }
2824 return 0;
2825}
2826
2827static void ovs_free_percpu_storage(void)
2828{
2829 free_percpu(ovs_pcpu_storage);
2830}
2831
2832static int __init dp_init(void)
2833{
2834 int err;
2835
2836 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2837 sizeof_field(struct sk_buff, cb));
2838
2839 pr_info("Open vSwitch switching datapath\n");
2840
2841 err = ovs_alloc_percpu_storage();
2842 if (err)
2843 goto error;
2844
2845 err = ovs_internal_dev_rtnl_link_register();
2846 if (err)
2847 goto error;
2848
2849 err = ovs_flow_init();
2850 if (err)
2851 goto error_unreg_rtnl_link;
2852
2853 err = ovs_vport_init();
2854 if (err)
2855 goto error_flow_exit;
2856
2857 err = register_pernet_device(&ovs_net_ops);
2858 if (err)
2859 goto error_vport_exit;
2860
2861 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2862 if (err)
2863 goto error_netns_exit;
2864
2865 err = ovs_netdev_init();
2866 if (err)
2867 goto error_unreg_notifier;
2868
2869 err = dp_register_genl();
2870 if (err < 0)
2871 goto error_unreg_netdev;
2872
2873 drop_reasons_register_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH,
2874 &drop_reason_list_ovs);
2875
2876 return 0;
2877
2878error_unreg_netdev:
2879 ovs_netdev_exit();
2880error_unreg_notifier:
2881 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2882error_netns_exit:
2883 unregister_pernet_device(&ovs_net_ops);
2884error_vport_exit:
2885 ovs_vport_exit();
2886error_flow_exit:
2887 ovs_flow_exit();
2888error_unreg_rtnl_link:
2889 ovs_internal_dev_rtnl_link_unregister();
2890error:
2891 ovs_free_percpu_storage();
2892 return err;
2893}
2894
2895static void dp_cleanup(void)
2896{
2897 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2898 ovs_netdev_exit();
2899 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2900 unregister_pernet_device(&ovs_net_ops);
2901 drop_reasons_unregister_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH);
2902 rcu_barrier();
2903 ovs_vport_exit();
2904 ovs_flow_exit();
2905 ovs_internal_dev_rtnl_link_unregister();
2906 ovs_free_percpu_storage();
2907}
2908
2909module_init(dp_init);
2910module_exit(dp_cleanup);
2911
2912MODULE_DESCRIPTION("Open vSwitch switching datapath");
2913MODULE_LICENSE("GPL");
2914MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2915MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2916MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2917MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2918MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2919MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);