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
30 int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
31 {
32 kfree_skb(skb);
33 return 0;
34 }
35 EXPORT_SYMBOL(dst_discard_out);
36
37 const 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 };
45 EXPORT_SYMBOL(dst_default_metrics);
46
47 void 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, &dst->dev_tracker, GFP_ATOMIC);
53 dst->ops = ops;
54 dst_init_metrics(dst, dst_default_metrics.metrics, 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(&dst->__rcuref, 1);
70 INIT_LIST_HEAD(&dst->rt_uncached);
71 dst->__use = 0;
72 dst->lastuse = jiffies;
73 dst->flags = flags;
74 if (!(flags & DST_NOCOUNT))
75 dst_entries_add(ops, 1);
76 }
77 EXPORT_SYMBOL(dst_init);
78
79 void *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(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 }
97 EXPORT_SYMBOL(dst_alloc);
98
99 static 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(dst->dev, &dst->dev_tracker);
115
116 lwtstate_put(dst->lwtstate);
117
118 if (dst->flags & DST_METADATA)
119 metadata_dst_free((struct metadata_dst *)dst);
120 else
121 kmem_cache_free(dst->ops->kmem_cachep, dst);
122
123 dst = child;
124 if (dst)
125 dst_release_immediate(dst);
126 }
127
128 static 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 */
144 void dst_dev_put(struct dst_entry *dst)
145 {
146 struct net_device *dev = dst->dev;
147
148 dst->obsolete = DST_OBSOLETE_DEAD;
149 if (dst->ops->ifdown)
150 dst->ops->ifdown(dst, dev);
151 dst->input = dst_discard;
152 dst->output = dst_discard_out;
153 dst->dev = blackhole_netdev;
154 netdev_ref_replace(dev, blackhole_netdev, &dst->dev_tracker,
155 GFP_ATOMIC);
156 }
157 EXPORT_SYMBOL(dst_dev_put);
158
159 static void dst_count_dec(struct dst_entry *dst)
160 {
161 if (!(dst->flags & DST_NOCOUNT))
162 dst_entries_add(dst->ops, -1);
163 }
164
165 void dst_release(struct dst_entry *dst)
166 {
167 if (dst && rcuref_put(&dst->__rcuref)) {
168 dst_count_dec(dst);
169 call_rcu_hurry(&dst->rcu_head, dst_destroy_rcu);
170 }
171 }
172 EXPORT_SYMBOL(dst_release);
173
174 void dst_release_immediate(struct dst_entry *dst)
175 {
176 if (dst && rcuref_put(&dst->__rcuref)) {
177 dst_count_dec(dst);
178 dst_destroy(dst);
179 }
180 }
181 EXPORT_SYMBOL(dst_release_immediate);
182
183 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
184 {
185 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
186
187 if (p) {
188 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
189 unsigned long prev, new;
190
191 refcount_set(&p->refcnt, 1);
192 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
193
194 new = (unsigned long) p;
195 prev = cmpxchg(&dst->_metrics, old, new);
196
197 if (prev != old) {
198 kfree(p);
199 p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
200 if (prev & DST_METRICS_READ_ONLY)
201 p = NULL;
202 } else if (prev & DST_METRICS_REFCOUNTED) {
203 if (refcount_dec_and_test(&old_p->refcnt))
204 kfree(old_p);
205 }
206 }
207 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
208 return (u32 *)p;
209 }
210 EXPORT_SYMBOL(dst_cow_metrics_generic);
211
212 /* Caller asserts that dst_metrics_read_only(dst) is false. */
213 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
214 {
215 unsigned long prev, new;
216
217 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
218 prev = cmpxchg(&dst->_metrics, old, new);
219 if (prev == old)
220 kfree(__DST_METRICS_PTR(old));
221 }
222 EXPORT_SYMBOL(__dst_destroy_metrics_generic);
223
224 struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie)
225 {
226 return NULL;
227 }
228
229 u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old)
230 {
231 return NULL;
232 }
233
234 struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
235 struct sk_buff *skb,
236 const void *daddr)
237 {
238 return NULL;
239 }
240
241 void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
242 struct sk_buff *skb, u32 mtu,
243 bool confirm_neigh)
244 {
245 }
246 EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu);
247
248 void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
249 struct sk_buff *skb)
250 {
251 }
252 EXPORT_SYMBOL_GPL(dst_blackhole_redirect);
253
254 unsigned int dst_blackhole_mtu(const struct dst_entry *dst)
255 {
256 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
257
258 return mtu ? : dst->dev->mtu;
259 }
260 EXPORT_SYMBOL_GPL(dst_blackhole_mtu);
261
262 static struct dst_ops dst_blackhole_ops = {
263 .family = AF_UNSPEC,
264 .neigh_lookup = dst_blackhole_neigh_lookup,
265 .check = dst_blackhole_check,
266 .cow_metrics = dst_blackhole_cow_metrics,
267 .update_pmtu = dst_blackhole_update_pmtu,
268 .redirect = dst_blackhole_redirect,
269 .mtu = dst_blackhole_mtu,
270 };
271
272 static void __metadata_dst_init(struct metadata_dst *md_dst,
273 enum metadata_type type, u8 optslen)
274 {
275 struct dst_entry *dst;
276
277 dst = &md_dst->dst;
278 dst_init(dst, &dst_blackhole_ops, NULL, DST_OBSOLETE_NONE,
279 DST_METADATA | DST_NOCOUNT);
280 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
281 md_dst->type = type;
282 }
283
284 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
285 gfp_t flags)
286 {
287 struct metadata_dst *md_dst;
288
289 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
290 if (!md_dst)
291 return NULL;
292
293 __metadata_dst_init(md_dst, type, optslen);
294
295 return md_dst;
296 }
297 EXPORT_SYMBOL_GPL(metadata_dst_alloc);
298
299 void metadata_dst_free(struct metadata_dst *md_dst)
300 {
301 #ifdef CONFIG_DST_CACHE
302 if (md_dst->type == METADATA_IP_TUNNEL)
303 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
304 #endif
305 if (md_dst->type == METADATA_XFRM)
306 dst_release(md_dst->u.xfrm_info.dst_orig);
307 kfree(md_dst);
308 }
309 EXPORT_SYMBOL_GPL(metadata_dst_free);
310
311 struct metadata_dst __percpu *
312 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
313 {
314 int cpu;
315 struct metadata_dst __percpu *md_dst;
316
317 md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
318 __alignof__(struct metadata_dst), flags);
319 if (!md_dst)
320 return NULL;
321
322 for_each_possible_cpu(cpu)
323 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
324
325 return md_dst;
326 }
327 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
328
329 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
330 {
331 int cpu;
332
333 for_each_possible_cpu(cpu) {
334 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
335
336 #ifdef CONFIG_DST_CACHE
337 if (one_md_dst->type == METADATA_IP_TUNNEL)
338 dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
339 #endif
340 if (one_md_dst->type == METADATA_XFRM)
341 dst_release(one_md_dst->u.xfrm_info.dst_orig);
342 }
343 free_percpu(md_dst);
344 }
345 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
346
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.