projects/linux·blob master · refs

592 lines · 12.8 KiB

1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Generic nexthop implementation
4 *
5 * Copyright (c) 2017-19 Cumulus Networks
6 * Copyright (c) 2017-19 David Ahern <[email protected]>
7 */
8
9#ifndef __LINUX_NEXTHOP_H
10#define __LINUX_NEXTHOP_H
11
12#include <linux/netdevice.h>
13#include <linux/notifier.h>
14#include <linux/route.h>
15#include <linux/types.h>
16#include <net/ip_fib.h>
17#include <net/ip6_fib.h>
18#include <net/netlink.h>
19
20#define NEXTHOP_VALID_USER_FLAGS RTNH_F_ONLINK
21
22struct nexthop;
23
24struct nh_config {
25 u32 nh_id;
26
27 u8 nh_family;
28 u8 nh_protocol;
29 u8 nh_blackhole;
30 u8 nh_fdb;
31 __be16 nh_dst_port;
32 u32 nh_flags;
33
34 int nh_ifindex;
35 struct net_device *dev;
36
37 union {
38 __be32 ipv4;
39 struct in6_addr ipv6;
40 } gw;
41
42 struct nlattr *nh_grp;
43 u16 nh_grp_type;
44 u16 nh_grp_res_num_buckets;
45 unsigned long nh_grp_res_idle_timer;
46 unsigned long nh_grp_res_unbalanced_timer;
47 bool nh_grp_res_has_num_buckets;
48 bool nh_grp_res_has_idle_timer;
49 bool nh_grp_res_has_unbalanced_timer;
50
51 bool nh_hw_stats;
52
53 struct nlattr *nh_encap;
54 u16 nh_encap_type;
55
56 u32 nlflags;
57 struct nl_info nlinfo;
58};
59
60struct nh_info {
61 struct hlist_node dev_hash; /* entry on netns devhash */
62 struct nexthop *nh_parent;
63
64 u8 family;
65 bool reject_nh;
66 bool fdb_nh;
67 __be16 dst_port;
68
69 union {
70 struct fib_nh_common fib_nhc;
71 struct fib_nh fib_nh;
72 struct fib6_nh fib6_nh;
73 };
74};
75
76struct nh_res_bucket {
77 struct nh_grp_entry __rcu *nh_entry;
78 atomic_long_t used_time;
79 unsigned long migrated_time;
80 bool occupied;
81 u8 nh_flags;
82};
83
84struct nh_res_table {
85 struct net *net;
86 u32 nhg_id;
87 struct delayed_work upkeep_dw;
88
89 /* List of NHGEs that have too few buckets ("uw" for underweight).
90 * Reclaimed buckets will be given to entries in this list.
91 */
92 struct list_head uw_nh_entries;
93 unsigned long unbalanced_since;
94
95 u32 idle_timer;
96 u32 unbalanced_timer;
97
98 u16 num_nh_buckets;
99 struct nh_res_bucket nh_buckets[] __counted_by(num_nh_buckets);
100};
101
102struct nh_grp_entry_stats {
103 u64_stats_t packets;
104 struct u64_stats_sync syncp;
105};
106
107struct nh_grp_entry {
108 struct nexthop *nh;
109 struct nh_grp_entry_stats __percpu *stats;
110 u16 weight;
111
112 union {
113 struct {
114 atomic_t upper_bound;
115 } hthr;
116 struct {
117 /* Member on uw_nh_entries. */
118 struct list_head uw_nh_entry;
119
120 u16 count_buckets;
121 u16 wants_buckets;
122 } res;
123 };
124
125 struct list_head nh_list;
126 struct nexthop *nh_parent; /* nexthop of group with this entry */
127 u64 packets_hw;
128};
129
130struct nh_group {
131 struct nh_group *spare; /* spare group for removals */
132 u16 num_nh;
133 bool is_multipath;
134 bool hash_threshold;
135 bool resilient;
136 bool fdb_nh;
137 bool has_v4;
138 bool hw_stats;
139
140 struct nh_res_table __rcu *res_table;
141 struct nh_grp_entry nh_entries[] __counted_by(num_nh);
142};
143
144struct nexthop {
145 struct rb_node rb_node; /* entry on netns rbtree */
146 struct list_head fi_list; /* v4 entries using nh */
147 struct list_head f6i_list; /* v6 entries using nh */
148 struct list_head fdb_list; /* fdb entries using this nh */
149 struct list_head grp_list; /* nh group entries using this nh */
150 struct net *net;
151
152 u32 id;
153
154 u8 protocol; /* app managing this nh */
155 u8 nh_flags;
156 bool is_group;
157 bool dead;
158 spinlock_t lock; /* protect dead and f6i_list */
159
160 refcount_t refcnt;
161 struct rcu_head rcu;
162
163 union {
164 struct nh_info __rcu *nh_info;
165 struct nh_group __rcu *nh_grp;
166 };
167};
168
169enum nexthop_event_type {
170 NEXTHOP_EVENT_DEL,
171 NEXTHOP_EVENT_REPLACE,
172 NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE,
173 NEXTHOP_EVENT_BUCKET_REPLACE,
174 NEXTHOP_EVENT_HW_STATS_REPORT_DELTA,
175};
176
177enum nh_notifier_info_type {
178 NH_NOTIFIER_INFO_TYPE_SINGLE,
179 NH_NOTIFIER_INFO_TYPE_GRP,
180 NH_NOTIFIER_INFO_TYPE_RES_TABLE,
181 NH_NOTIFIER_INFO_TYPE_RES_BUCKET,
182 NH_NOTIFIER_INFO_TYPE_GRP_HW_STATS,
183};
184
185struct nh_notifier_single_info {
186 struct net_device *dev;
187 u8 gw_family;
188 union {
189 __be32 ipv4;
190 struct in6_addr ipv6;
191 };
192 u32 id;
193 u8 is_reject:1,
194 is_fdb:1,
195 has_encap:1;
196};
197
198struct nh_notifier_grp_entry_info {
199 u16 weight;
200 struct nh_notifier_single_info nh;
201};
202
203struct nh_notifier_grp_info {
204 u16 num_nh;
205 bool is_fdb;
206 bool hw_stats;
207 struct nh_notifier_grp_entry_info nh_entries[] __counted_by(num_nh);
208};
209
210struct nh_notifier_res_bucket_info {
211 u16 bucket_index;
212 unsigned int idle_timer_ms;
213 bool force;
214 struct nh_notifier_single_info old_nh;
215 struct nh_notifier_single_info new_nh;
216};
217
218struct nh_notifier_res_table_info {
219 u16 num_nh_buckets;
220 bool hw_stats;
221 struct nh_notifier_single_info nhs[] __counted_by(num_nh_buckets);
222};
223
224struct nh_notifier_grp_hw_stats_entry_info {
225 u32 id;
226 u64 packets;
227};
228
229struct nh_notifier_grp_hw_stats_info {
230 u16 num_nh;
231 bool hw_stats_used;
232 struct nh_notifier_grp_hw_stats_entry_info stats[] __counted_by(num_nh);
233};
234
235struct nh_notifier_info {
236 struct net *net;
237 struct netlink_ext_ack *extack;
238 u32 id;
239 enum nh_notifier_info_type type;
240 union {
241 struct nh_notifier_single_info *nh;
242 struct nh_notifier_grp_info *nh_grp;
243 struct nh_notifier_res_table_info *nh_res_table;
244 struct nh_notifier_res_bucket_info *nh_res_bucket;
245 struct nh_notifier_grp_hw_stats_info *nh_grp_hw_stats;
246 };
247};
248
249int register_nexthop_notifier(struct net *net, struct notifier_block *nb,
250 struct netlink_ext_ack *extack);
251int __unregister_nexthop_notifier(struct net *net, struct notifier_block *nb);
252int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb);
253void nexthop_set_hw_flags(struct net *net, u32 id, bool offload, bool trap);
254void nexthop_bucket_set_hw_flags(struct net *net, u32 id, u16 bucket_index,
255 bool offload, bool trap);
256void nexthop_res_grp_activity_update(struct net *net, u32 id, u16 num_buckets,
257 unsigned long *activity);
258void nh_grp_hw_stats_report_delta(struct nh_notifier_grp_hw_stats_info *info,
259 unsigned int nh_idx,
260 u64 delta_packets);
261
262/* caller is holding rcu or rtnl; no reference taken to nexthop */
263struct nexthop *nexthop_find_by_id(struct net *net, u32 id);
264void nexthop_free_rcu(struct rcu_head *head);
265
266static inline bool nexthop_get(struct nexthop *nh)
267{
268 return refcount_inc_not_zero(&nh->refcnt);
269}
270
271static inline void nexthop_put(struct nexthop *nh)
272{
273 if (refcount_dec_and_test(&nh->refcnt))
274 call_rcu_hurry(&nh->rcu, nexthop_free_rcu);
275}
276
277static inline bool nexthop_cmp(const struct nexthop *nh1,
278 const struct nexthop *nh2)
279{
280 return nh1 == nh2;
281}
282
283static inline bool nexthop_is_fdb(const struct nexthop *nh)
284{
285 if (nh->is_group) {
286 const struct nh_group *nh_grp;
287
288 nh_grp = rcu_dereference_rtnl(nh->nh_grp);
289 return nh_grp->fdb_nh;
290 } else {
291 const struct nh_info *nhi;
292
293 nhi = rcu_dereference_rtnl(nh->nh_info);
294 return nhi->fdb_nh;
295 }
296}
297
298static inline bool nexthop_has_v4(const struct nexthop *nh)
299{
300 if (nh->is_group) {
301 struct nh_group *nh_grp;
302
303 nh_grp = rcu_dereference_rtnl(nh->nh_grp);
304 return nh_grp->has_v4;
305 }
306 return false;
307}
308
309static inline bool nexthop_is_multipath(const struct nexthop *nh)
310{
311 if (nh->is_group) {
312 struct nh_group *nh_grp;
313
314 nh_grp = rcu_dereference_rtnl(nh->nh_grp);
315 return nh_grp->is_multipath;
316 }
317 return false;
318}
319
320struct nexthop *nexthop_select_path(struct nexthop *nh, int hash);
321
322static inline unsigned int nexthop_num_path(const struct nexthop *nh)
323{
324 unsigned int rc = 1;
325
326 if (nh->is_group) {
327 struct nh_group *nh_grp;
328
329 nh_grp = rcu_dereference_rtnl(nh->nh_grp);
330 if (nh_grp->is_multipath)
331 rc = nh_grp->num_nh;
332 }
333
334 return rc;
335}
336
337static inline
338struct nexthop *nexthop_mpath_select(const struct nh_group *nhg, int nhsel)
339{
340 /* for_nexthops macros in fib_semantics.c grabs a pointer to
341 * the nexthop before checking nhsel
342 */
343 if (nhsel >= nhg->num_nh)
344 return NULL;
345
346 return nhg->nh_entries[nhsel].nh;
347}
348
349static inline
350int nexthop_mpath_fill_node(struct sk_buff *skb, struct nexthop *nh,
351 u8 rt_family)
352{
353 struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
354 int i;
355
356 for (i = 0; i < nhg->num_nh; i++) {
357 struct nexthop *nhe = nhg->nh_entries[i].nh;
358 struct nh_info *nhi = rcu_dereference_rtnl(nhe->nh_info);
359 struct fib_nh_common *nhc = &nhi->fib_nhc;
360 int weight = nhg->nh_entries[i].weight;
361
362 if (fib_add_nexthop(skb, nhc, weight, rt_family, 0) < 0)
363 return -EMSGSIZE;
364 }
365
366 return 0;
367}
368
369/* called with rcu lock */
370static inline bool nexthop_is_blackhole(const struct nexthop *nh)
371{
372 const struct nh_info *nhi;
373
374 if (nh->is_group) {
375 struct nh_group *nh_grp;
376
377 nh_grp = rcu_dereference_rtnl(nh->nh_grp);
378 if (nh_grp->num_nh > 1)
379 return false;
380
381 nh = nh_grp->nh_entries[0].nh;
382 }
383
384 nhi = rcu_dereference_rtnl(nh->nh_info);
385 return nhi->reject_nh;
386}
387
388static inline void nexthop_path_fib_result(struct fib_result *res, int hash)
389{
390 struct nh_info *nhi;
391 struct nexthop *nh;
392
393 nh = nexthop_select_path(res->fi->nh, hash);
394 nhi = rcu_dereference(nh->nh_info);
395 res->nhc = &nhi->fib_nhc;
396}
397
398/* called with rcu read lock or rtnl held */
399static inline
400struct fib_nh_common *nexthop_fib_nhc(struct nexthop *nh, int nhsel)
401{
402 struct nh_info *nhi;
403
404 BUILD_BUG_ON(offsetof(struct fib_nh, nh_common) != 0);
405 BUILD_BUG_ON(offsetof(struct fib6_nh, nh_common) != 0);
406
407 if (nh->is_group) {
408 struct nh_group *nh_grp;
409
410 nh_grp = rcu_dereference_rtnl(nh->nh_grp);
411 if (nh_grp->is_multipath) {
412 nh = nexthop_mpath_select(nh_grp, nhsel);
413 if (!nh)
414 return NULL;
415 }
416 }
417
418 nhi = rcu_dereference_rtnl(nh->nh_info);
419 return &nhi->fib_nhc;
420}
421
422/* called from fib_table_lookup with rcu_lock */
423static inline
424struct fib_nh_common *nexthop_get_nhc_lookup(const struct nexthop *nh,
425 int fib_flags,
426 const struct flowi4 *flp,
427 int *nhsel)
428{
429 struct nh_info *nhi;
430
431 if (nh->is_group) {
432 struct nh_group *nhg = rcu_dereference(nh->nh_grp);
433 int i;
434
435 for (i = 0; i < nhg->num_nh; i++) {
436 struct nexthop *nhe = nhg->nh_entries[i].nh;
437
438 nhi = rcu_dereference(nhe->nh_info);
439 if (fib_lookup_good_nhc(&nhi->fib_nhc, fib_flags, flp)) {
440 *nhsel = i;
441 return &nhi->fib_nhc;
442 }
443 }
444 } else {
445 nhi = rcu_dereference(nh->nh_info);
446 if (fib_lookup_good_nhc(&nhi->fib_nhc, fib_flags, flp)) {
447 *nhsel = 0;
448 return &nhi->fib_nhc;
449 }
450 }
451
452 return NULL;
453}
454
455static inline bool nexthop_uses_dev(const struct nexthop *nh,
456 const struct net_device *dev)
457{
458 struct nh_info *nhi;
459
460 if (nh->is_group) {
461 struct nh_group *nhg = rcu_dereference(nh->nh_grp);
462 int i;
463
464 for (i = 0; i < nhg->num_nh; i++) {
465 struct nexthop *nhe = nhg->nh_entries[i].nh;
466
467 nhi = rcu_dereference(nhe->nh_info);
468 if (nhc_l3mdev_matches_dev(&nhi->fib_nhc, dev))
469 return true;
470 }
471 } else {
472 nhi = rcu_dereference(nh->nh_info);
473 if (nhc_l3mdev_matches_dev(&nhi->fib_nhc, dev))
474 return true;
475 }
476
477 return false;
478}
479
480static inline unsigned int fib_info_num_path(const struct fib_info *fi)
481{
482 if (unlikely(fi->nh))
483 return nexthop_num_path(fi->nh);
484
485 return fi->fib_nhs;
486}
487
488int fib_check_nexthop(struct nexthop *nh, u8 scope,
489 struct netlink_ext_ack *extack);
490
491static inline struct fib_nh_common *fib_info_nhc(struct fib_info *fi, int nhsel)
492{
493 if (unlikely(fi->nh))
494 return nexthop_fib_nhc(fi->nh, nhsel);
495
496 return &fi->fib_nh[nhsel].nh_common;
497}
498
499/* only used when fib_nh is built into fib_info */
500static inline struct fib_nh *fib_info_nh(struct fib_info *fi, int nhsel)
501{
502 WARN_ON(fi->nh);
503
504 return &fi->fib_nh[nhsel];
505}
506
507/*
508 * IPv6 variants
509 */
510int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
511 struct netlink_ext_ack *extack);
512
513/* Caller should either hold rcu_read_lock(), or RTNL. */
514static inline struct fib6_nh *nexthop_fib6_nh(struct nexthop *nh)
515{
516 struct nh_info *nhi;
517
518 if (nh->is_group) {
519 struct nh_group *nh_grp;
520
521 nh_grp = rcu_dereference_rtnl(nh->nh_grp);
522 nh = nexthop_mpath_select(nh_grp, 0);
523 if (!nh)
524 return NULL;
525 }
526
527 nhi = rcu_dereference_rtnl(nh->nh_info);
528 if (nhi->family == AF_INET6)
529 return &nhi->fib6_nh;
530
531 return NULL;
532}
533
534static inline struct net_device *fib6_info_nh_dev(struct fib6_info *f6i)
535{
536 struct fib6_nh *fib6_nh;
537
538 fib6_nh = f6i->nh ? nexthop_fib6_nh(f6i->nh) : f6i->fib6_nh;
539 return fib6_nh->fib_nh_dev;
540}
541
542static inline void nexthop_path_fib6_result(struct fib6_result *res, int hash)
543{
544 struct nexthop *nh = res->f6i->nh;
545 struct nh_info *nhi;
546
547 nh = nexthop_select_path(nh, hash);
548
549 nhi = rcu_dereference_rtnl(nh->nh_info);
550 if (nhi->reject_nh) {
551 res->fib6_type = RTN_BLACKHOLE;
552 res->fib6_flags |= RTF_REJECT;
553 res->nh = nexthop_fib6_nh(nh);
554 } else {
555 res->nh = &nhi->fib6_nh;
556 }
557}
558
559int nexthop_for_each_fib6_nh(struct nexthop *nh,
560 int (*cb)(struct fib6_nh *nh, void *arg),
561 void *arg);
562
563static inline int nexthop_get_family(struct nexthop *nh)
564{
565 struct nh_info *nhi = rcu_dereference_rtnl(nh->nh_info);
566
567 return nhi->family;
568}
569
570static inline
571struct fib_nh_common *nexthop_fdb_nhc(struct nexthop *nh)
572{
573 struct nh_info *nhi = rcu_dereference_rtnl(nh->nh_info);
574
575 return &nhi->fib_nhc;
576}
577
578static inline struct fib_nh_common *nexthop_path_fdb_result(struct nexthop *nh,
579 int hash,
580 __be16 *dst_port)
581{
582 struct nh_info *nhi;
583 struct nexthop *nhp;
584
585 nhp = nexthop_select_path(nh, hash);
586 if (unlikely(!nhp))
587 return NULL;
588 nhi = rcu_dereference(nhp->nh_info);
589 *dst_port = nhi->dst_port;
590 return &nhi->fib_nhc;
591}
592#endif