1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/core/dst.c Protocol independent destination cache.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 */
8
9#include <linux/bitops.h>
10#include <linux/errno.h>
11#include <linux/init.h>
12#include <linux/kernel.h>
13#include <linux/workqueue.h>
14#include <linux/mm.h>
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <linux/netdevice.h>
18#include <linux/skbuff.h>
19#include <linux/string.h>
20#include <linux/types.h>
21#include <net/net_namespace.h>
22#include <linux/sched.h>
23#include <linux/prefetch.h>
24#include <net/lwtunnel.h>
25#include <net/xfrm.h>
26
27#include <net/dst.h>
28#include <net/dst_metadata.h>
29
30int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
31{
32 kfree_skb(skb);
33 return 0;
34}
35EXPORT_SYMBOL(dst_discard_out);
36
37const struct dst_metrics dst_default_metrics = {
38 /* This initializer is needed to force linker to place this variable
39 * into const section. Otherwise it might end into bss section.
40 * We really want to avoid false sharing on this variable, and catch
41 * any writes on it.
42 */
43 .refcnt = REFCOUNT_INIT(1),
44};
45EXPORT_SYMBOL(dst_default_metrics);
46
47void dst_init(struct dst_entry *dst, struct dst_ops *ops,
48 struct net_device *dev, int initial_obsolete,
49 unsigned short flags)
50{
51 dst->dev = dev;
52 netdev_hold(dev, tracker: &dst->dev_tracker, GFP_ATOMIC);
53 dst->ops = ops;
54 dst_init_metrics(dst, src_metrics: dst_default_metrics.metrics, read_only: true);
55 dst->expires = 0UL;
56#ifdef CONFIG_XFRM
57 dst->xfrm = NULL;
58#endif
59 dst->input = dst_discard;
60 dst->output = dst_discard_out;
61 dst->error = 0;
62 dst->obsolete = initial_obsolete;
63 dst->header_len = 0;
64 dst->trailer_len = 0;
65#ifdef CONFIG_IP_ROUTE_CLASSID
66 dst->tclassid = 0;
67#endif
68 dst->lwtstate = NULL;
69 rcuref_init(ref: &dst->__rcuref, cnt: 1);
70 INIT_LIST_HEAD(list: &dst->rt_uncached);
71 dst->__use = 0;
72 dst->lastuse = jiffies;
73 dst->flags = flags;
74 if (!(flags & DST_NOCOUNT))
75 dst_entries_add(dst: ops, val: 1);
76}
77EXPORT_SYMBOL(dst_init);
78
79void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
80 int initial_obsolete, unsigned short flags)
81{
82 struct dst_entry *dst;
83
84 if (ops->gc &&
85 !(flags & DST_NOCOUNT) &&
86 dst_entries_get_fast(dst: ops) > ops->gc_thresh)
87 ops->gc(ops);
88
89 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
90 if (!dst)
91 return NULL;
92
93 dst_init(dst, ops, dev, initial_obsolete, flags);
94
95 return dst;
96}
97EXPORT_SYMBOL(dst_alloc);
98
99static void dst_destroy(struct dst_entry *dst)
100{
101 struct dst_entry *child = NULL;
102
103 smp_rmb();
104
105#ifdef CONFIG_XFRM
106 if (dst->xfrm) {
107 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
108
109 child = xdst->child;
110 }
111#endif
112 if (dst->ops->destroy)
113 dst->ops->destroy(dst);
114 netdev_put(dev: dst->dev, tracker: &dst->dev_tracker);
115
116 lwtstate_put(lws: dst->lwtstate);
117
118 if (dst->flags & DST_METADATA)
119 metadata_dst_free((struct metadata_dst *)dst);
120 else
121 kmem_cache_free(s: dst->ops->kmem_cachep, objp: dst);
122
123 dst = child;
124 if (dst)
125 dst_release_immediate(dst);
126}
127
128static void dst_destroy_rcu(struct rcu_head *head)
129{
130 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
131
132 dst_destroy(dst);
133}
134
135/* Operations to mark dst as DEAD and clean up the net device referenced
136 * by dst:
137 * 1. put the dst under blackhole interface and discard all tx/rx packets
138 * on this route.
139 * 2. release the net_device
140 * This function should be called when removing routes from the fib tree
141 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
142 * make the next dst_ops->check() fail.
143 */
144void dst_dev_put(struct dst_entry *dst)
145{
146 struct net_device *dev = dst->dev;
147
148 WRITE_ONCE(dst->obsolete, DST_OBSOLETE_DEAD);
149 if (dst->ops->ifdown)
150 dst->ops->ifdown(dst, dev);
151 WRITE_ONCE(dst->input, dst_discard);
152 WRITE_ONCE(dst->output, dst_discard_out);
153 rcu_assign_pointer(dst->dev_rcu, blackhole_netdev);
154 netdev_ref_replace(odev: dev, ndev: blackhole_netdev, tracker: &dst->dev_tracker,
155 GFP_ATOMIC);
156}
157EXPORT_SYMBOL(dst_dev_put);
158
159static void dst_count_dec(struct dst_entry *dst)
160{
161 if (!(dst->flags & DST_NOCOUNT))
162 dst_entries_add(dst: dst->ops, val: -1);
163}
164
165void dst_release(struct dst_entry *dst)
166{
167 if (dst && rcuref_put(ref: &dst->__rcuref)) {
168#ifdef CONFIG_DST_CACHE
169 if (dst->flags & DST_METADATA) {
170 struct metadata_dst *md_dst = (struct metadata_dst *)dst;
171
172 if (md_dst->type == METADATA_IP_TUNNEL)
173 dst_cache_reset_now(dst_cache: &md_dst->u.tun_info.dst_cache);
174 }
175#endif
176 dst_count_dec(dst);
177 call_rcu_hurry(head: &dst->rcu_head, func: dst_destroy_rcu);
178 }
179}
180EXPORT_SYMBOL(dst_release);
181
182void dst_release_immediate(struct dst_entry *dst)
183{
184 if (dst && rcuref_put(ref: &dst->__rcuref)) {
185 dst_count_dec(dst);
186 dst_destroy(dst);
187 }
188}
189EXPORT_SYMBOL(dst_release_immediate);
190
191u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
192{
193 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
194
195 if (p) {
196 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
197 unsigned long prev, new;
198
199 refcount_set(r: &p->refcnt, n: 1);
200 memcpy(to: p->metrics, from: old_p->metrics, len: sizeof(p->metrics));
201
202 new = (unsigned long) p;
203 prev = cmpxchg(&dst->_metrics, old, new);
204
205 if (prev != old) {
206 kfree(objp: p);
207 p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
208 if (prev & DST_METRICS_READ_ONLY)
209 p = NULL;
210 } else if (prev & DST_METRICS_REFCOUNTED) {
211 if (refcount_dec_and_test(r: &old_p->refcnt))
212 kfree(objp: old_p);
213 }
214 }
215 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
216 return (u32 *)p;
217}
218EXPORT_SYMBOL(dst_cow_metrics_generic);
219
220/* Caller asserts that dst_metrics_read_only(dst) is false. */
221void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
222{
223 unsigned long prev, new;
224
225 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
226 prev = cmpxchg(&dst->_metrics, old, new);
227 if (prev == old)
228 kfree(__DST_METRICS_PTR(old));
229}
230EXPORT_SYMBOL(__dst_destroy_metrics_generic);
231
232struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie)
233{
234 return NULL;
235}
236
237u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old)
238{
239 return NULL;
240}
241
242struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
243 struct sk_buff *skb,
244 const void *daddr)
245{
246 return NULL;
247}
248
249void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
250 struct sk_buff *skb, u32 mtu,
251 bool confirm_neigh)
252{
253}
254EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu);
255
256void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
257 struct sk_buff *skb)
258{
259}
260EXPORT_SYMBOL_GPL(dst_blackhole_redirect);
261
262unsigned int dst_blackhole_mtu(const struct dst_entry *dst)
263{
264 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
265
266 return mtu ? : dst_dev(dst)->mtu;
267}
268EXPORT_SYMBOL_GPL(dst_blackhole_mtu);
269
270static struct dst_ops dst_blackhole_ops = {
271 .family = AF_UNSPEC,
272 .neigh_lookup = dst_blackhole_neigh_lookup,
273 .check = dst_blackhole_check,
274 .cow_metrics = dst_blackhole_cow_metrics,
275 .update_pmtu = dst_blackhole_update_pmtu,
276 .redirect = dst_blackhole_redirect,
277 .mtu = dst_blackhole_mtu,
278};
279
280static void __metadata_dst_init(struct metadata_dst *md_dst,
281 enum metadata_type type, u8 optslen)
282{
283 struct dst_entry *dst;
284
285 dst = &md_dst->dst;
286 dst_init(dst, &dst_blackhole_ops, NULL, DST_OBSOLETE_NONE,
287 DST_METADATA | DST_NOCOUNT);
288 memset(s: dst + 1, c: 0, n: sizeof(*md_dst) + optslen - sizeof(*dst));
289 md_dst->type = type;
290}
291
292struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
293 gfp_t flags)
294{
295 struct metadata_dst *md_dst;
296
297 md_dst = kmalloc(struct_size(md_dst, u.tun_info.options, optslen),
298 flags);
299 if (!md_dst)
300 return NULL;
301
302 __metadata_dst_init(md_dst, type, optslen);
303
304 return md_dst;
305}
306EXPORT_SYMBOL_GPL(metadata_dst_alloc);
307
308void metadata_dst_free(struct metadata_dst *md_dst)
309{
310#ifdef CONFIG_DST_CACHE
311 if (md_dst->type == METADATA_IP_TUNNEL)
312 dst_cache_destroy(dst_cache: &md_dst->u.tun_info.dst_cache);
313#endif
314 if (md_dst->type == METADATA_XFRM)
315 dst_release(md_dst->u.xfrm_info.dst_orig);
316 kfree(objp: md_dst);
317}
318EXPORT_SYMBOL_GPL(metadata_dst_free);
319
320struct metadata_dst __percpu *
321metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
322{
323 int cpu;
324 struct metadata_dst __percpu *md_dst;
325
326 md_dst = __alloc_percpu_gfp(struct_size(md_dst, u.tun_info.options,
327 optslen),
328 __alignof__(struct metadata_dst), flags);
329 if (!md_dst)
330 return NULL;
331
332 for_each_possible_cpu(cpu)
333 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
334
335 return md_dst;
336}
337EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
338
339void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
340{
341 int cpu;
342
343 for_each_possible_cpu(cpu) {
344 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
345
346#ifdef CONFIG_DST_CACHE
347 if (one_md_dst->type == METADATA_IP_TUNNEL)
348 dst_cache_destroy(dst_cache: &one_md_dst->u.tun_info.dst_cache);
349#endif
350 if (one_md_dst->type == METADATA_XFRM)
351 dst_release(one_md_dst->u.xfrm_info.dst_orig);
352 }
353 free_percpu(pdata: md_dst);
354}
355EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
356