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->rt_uncached_list = NULL;
72 dst->__use = 0;
73 dst->lastuse = jiffies;
74 dst->flags = flags;
75 if (!(flags & DST_NOCOUNT))
76 dst_entries_add(dst: ops, val: 1);
77}
78EXPORT_SYMBOL(dst_init);
79
80void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
81 int initial_obsolete, unsigned short flags)
82{
83 struct dst_entry *dst;
84
85 if (ops->gc &&
86 !(flags & DST_NOCOUNT) &&
87 dst_entries_get_fast(dst: ops) > ops->gc_thresh)
88 ops->gc(ops);
89
90 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
91 if (!dst)
92 return NULL;
93
94 dst_init(dst, ops, dev, initial_obsolete, flags);
95
96 return dst;
97}
98EXPORT_SYMBOL(dst_alloc);
99
100static void dst_destroy(struct dst_entry *dst)
101{
102 struct dst_entry *child = NULL;
103
104 smp_rmb();
105
106#ifdef CONFIG_XFRM
107 if (dst->xfrm) {
108 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
109
110 child = xdst->child;
111 }
112#endif
113 if (dst->ops->destroy)
114 dst->ops->destroy(dst);
115 netdev_put(dev: dst->dev, tracker: &dst->dev_tracker);
116
117 lwtstate_put(lws: dst->lwtstate);
118
119 if (dst->flags & DST_METADATA)
120 metadata_dst_free((struct metadata_dst *)dst);
121 else
122 kmem_cache_free(s: dst->ops->kmem_cachep, objp: dst);
123
124 dst = child;
125 if (dst)
126 dst_release_immediate(dst);
127}
128
129static void dst_destroy_rcu(struct rcu_head *head)
130{
131 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
132
133 dst_destroy(dst);
134}
135
136/* Operations to mark dst as DEAD and clean up the net device referenced
137 * by dst:
138 * 1. put the dst under blackhole interface and discard all tx/rx packets
139 * on this route.
140 * 2. release the net_device
141 * This function should be called when removing routes from the fib tree
142 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
143 * make the next dst_ops->check() fail.
144 */
145void dst_dev_put(struct dst_entry *dst)
146{
147 struct net_device *dev = dst->dev;
148
149 WRITE_ONCE(dst->obsolete, DST_OBSOLETE_DEAD);
150 if (dst->ops->ifdown)
151 dst->ops->ifdown(dst, dev);
152 WRITE_ONCE(dst->input, dst_discard);
153 WRITE_ONCE(dst->output, dst_discard_out);
154 rcu_assign_pointer(dst->dev_rcu, blackhole_netdev);
155 netdev_ref_replace(odev: dev, ndev: blackhole_netdev, tracker: &dst->dev_tracker,
156 GFP_ATOMIC);
157}
158EXPORT_SYMBOL(dst_dev_put);
159
160static void dst_count_dec(struct dst_entry *dst)
161{
162 if (!(dst->flags & DST_NOCOUNT))
163 dst_entries_add(dst: dst->ops, val: -1);
164}
165
166void dst_release(struct dst_entry *dst)
167{
168 if (dst && rcuref_put(ref: &dst->__rcuref)) {
169#ifdef CONFIG_DST_CACHE
170 if (dst->flags & DST_METADATA) {
171 struct metadata_dst *md_dst = (struct metadata_dst *)dst;
172
173 if (md_dst->type == METADATA_IP_TUNNEL)
174 dst_cache_reset_now(dst_cache: &md_dst->u.tun_info.dst_cache);
175 }
176#endif
177 dst_count_dec(dst);
178 call_rcu_hurry(head: &dst->rcu_head, func: dst_destroy_rcu);
179 }
180}
181EXPORT_SYMBOL(dst_release);
182
183void dst_release_immediate(struct dst_entry *dst)
184{
185 if (dst && rcuref_put(ref: &dst->__rcuref)) {
186 dst_count_dec(dst);
187 dst_destroy(dst);
188 }
189}
190EXPORT_SYMBOL(dst_release_immediate);
191
192u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
193{
194 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
195
196 if (p) {
197 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
198 unsigned long prev, new;
199
200 refcount_set(r: &p->refcnt, n: 1);
201 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
202
203 new = (unsigned long) p;
204 prev = cmpxchg(&dst->_metrics, old, new);
205
206 if (prev != old) {
207 kfree(objp: p);
208 p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
209 if (prev & DST_METRICS_READ_ONLY)
210 p = NULL;
211 } else if (prev & DST_METRICS_REFCOUNTED) {
212 if (refcount_dec_and_test(r: &old_p->refcnt))
213 kfree(objp: old_p);
214 }
215 }
216 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
217 return (u32 *)p;
218}
219EXPORT_SYMBOL(dst_cow_metrics_generic);
220
221/* Caller asserts that dst_metrics_read_only(dst) is false. */
222void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
223{
224 unsigned long prev, new;
225
226 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
227 prev = cmpxchg(&dst->_metrics, old, new);
228 if (prev == old)
229 kfree(__DST_METRICS_PTR(old));
230}
231EXPORT_SYMBOL(__dst_destroy_metrics_generic);
232
233struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie)
234{
235 return NULL;
236}
237
238u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old)
239{
240 return NULL;
241}
242
243struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
244 struct sk_buff *skb,
245 const void *daddr)
246{
247 return NULL;
248}
249
250void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
251 struct sk_buff *skb, u32 mtu,
252 bool confirm_neigh)
253{
254}
255EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu);
256
257void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
258 struct sk_buff *skb)
259{
260}
261EXPORT_SYMBOL_GPL(dst_blackhole_redirect);
262
263unsigned int dst_blackhole_mtu(const struct dst_entry *dst)
264{
265 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
266
267 return mtu ? : dst_dev(dst)->mtu;
268}
269EXPORT_SYMBOL_GPL(dst_blackhole_mtu);
270
271static struct dst_ops dst_blackhole_ops = {
272 .family = AF_UNSPEC,
273 .neigh_lookup = dst_blackhole_neigh_lookup,
274 .check = dst_blackhole_check,
275 .cow_metrics = dst_blackhole_cow_metrics,
276 .update_pmtu = dst_blackhole_update_pmtu,
277 .redirect = dst_blackhole_redirect,
278 .mtu = dst_blackhole_mtu,
279};
280
281static void __metadata_dst_init(struct metadata_dst *md_dst,
282 enum metadata_type type, u8 optslen)
283{
284 struct dst_entry *dst;
285
286 dst = &md_dst->dst;
287 dst_init(dst, &dst_blackhole_ops, NULL, DST_OBSOLETE_NONE,
288 DST_METADATA | DST_NOCOUNT);
289 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
290 md_dst->type = type;
291}
292
293struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
294 gfp_t flags)
295{
296 struct metadata_dst *md_dst;
297
298 md_dst = kmalloc(struct_size(md_dst, u.tun_info.options, optslen),
299 flags);
300 if (!md_dst)
301 return NULL;
302
303 __metadata_dst_init(md_dst, type, optslen);
304
305 return md_dst;
306}
307EXPORT_SYMBOL_GPL(metadata_dst_alloc);
308
309void metadata_dst_free(struct metadata_dst *md_dst)
310{
311#ifdef CONFIG_DST_CACHE
312 if (md_dst->type == METADATA_IP_TUNNEL)
313 dst_cache_destroy(dst_cache: &md_dst->u.tun_info.dst_cache);
314#endif
315 if (md_dst->type == METADATA_XFRM)
316 dst_release(md_dst->u.xfrm_info.dst_orig);
317 kfree(objp: md_dst);
318}
319EXPORT_SYMBOL_GPL(metadata_dst_free);
320
321struct metadata_dst __percpu *
322metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
323{
324 int cpu;
325 struct metadata_dst __percpu *md_dst;
326
327 md_dst = __alloc_percpu_gfp(struct_size(md_dst, u.tun_info.options,
328 optslen),
329 __alignof__(struct metadata_dst), flags);
330 if (!md_dst)
331 return NULL;
332
333 for_each_possible_cpu(cpu)
334 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
335
336 return md_dst;
337}
338EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
339
340void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
341{
342 int cpu;
343
344 for_each_possible_cpu(cpu) {
345 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
346
347#ifdef CONFIG_DST_CACHE
348 if (one_md_dst->type == METADATA_IP_TUNNEL)
349 dst_cache_destroy(dst_cache: &one_md_dst->u.tun_info.dst_cache);
350#endif
351 if (one_md_dst->type == METADATA_XFRM)
352 dst_release(one_md_dst->u.xfrm_info.dst_orig);
353 }
354 free_percpu(pdata: md_dst);
355}
356EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
357

source code of linux/net/core/dst.c