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
dst_discard_out(struct net * net,struct sock * sk,struct sk_buff * skb)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
dst_init(struct dst_entry * dst,struct dst_ops * ops,struct net_device * dev,int initial_ref,int initial_obsolete,unsigned short flags)47 void dst_init(struct dst_entry *dst, struct dst_ops *ops,
48 struct net_device *dev, int initial_ref, int initial_obsolete,
49 unsigned short flags)
50 {
51 dst->dev = dev;
52 if (dev)
53 dev_hold(dev);
54 dst->ops = ops;
55 dst_init_metrics(dst, dst_default_metrics.metrics, true);
56 dst->expires = 0UL;
57 #ifdef CONFIG_XFRM
58 dst->xfrm = NULL;
59 #endif
60 dst->input = dst_discard;
61 dst->output = dst_discard_out;
62 dst->error = 0;
63 dst->obsolete = initial_obsolete;
64 dst->header_len = 0;
65 dst->trailer_len = 0;
66 #ifdef CONFIG_IP_ROUTE_CLASSID
67 dst->tclassid = 0;
68 #endif
69 dst->lwtstate = NULL;
70 atomic_set(&dst->__refcnt, initial_ref);
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
dst_alloc(struct dst_ops * ops,struct net_device * dev,int initial_ref,int initial_obsolete,unsigned short flags)79 void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
80 int initial_ref, 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_ref, initial_obsolete, flags);
94
95 return dst;
96 }
97 EXPORT_SYMBOL(dst_alloc);
98
dst_destroy(struct dst_entry * dst)99 struct dst_entry *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->flags & DST_NOCOUNT))
113 dst_entries_add(dst->ops, -1);
114
115 if (dst->ops->destroy)
116 dst->ops->destroy(dst);
117 if (dst->dev)
118 dev_put(dst->dev);
119
120 lwtstate_put(dst->lwtstate);
121
122 if (dst->flags & DST_METADATA)
123 metadata_dst_free((struct metadata_dst *)dst);
124 else
125 kmem_cache_free(dst->ops->kmem_cachep, dst);
126
127 dst = child;
128 if (dst)
129 dst_release_immediate(dst);
130 return NULL;
131 }
132 EXPORT_SYMBOL(dst_destroy);
133
dst_destroy_rcu(struct rcu_head * head)134 static void dst_destroy_rcu(struct rcu_head *head)
135 {
136 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
137
138 dst = dst_destroy(dst);
139 }
140
141 /* Operations to mark dst as DEAD and clean up the net device referenced
142 * by dst:
143 * 1. put the dst under blackhole interface and discard all tx/rx packets
144 * on this route.
145 * 2. release the net_device
146 * This function should be called when removing routes from the fib tree
147 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
148 * make the next dst_ops->check() fail.
149 */
dst_dev_put(struct dst_entry * dst)150 void dst_dev_put(struct dst_entry *dst)
151 {
152 struct net_device *dev = dst->dev;
153
154 dst->obsolete = DST_OBSOLETE_DEAD;
155 if (dst->ops->ifdown)
156 dst->ops->ifdown(dst, dev, true);
157 dst->input = dst_discard;
158 dst->output = dst_discard_out;
159 dst->dev = blackhole_netdev;
160 dev_hold(dst->dev);
161 dev_put(dev);
162 }
163 EXPORT_SYMBOL(dst_dev_put);
164
dst_release(struct dst_entry * dst)165 void dst_release(struct dst_entry *dst)
166 {
167 if (dst) {
168 int newrefcnt;
169
170 newrefcnt = atomic_dec_return(&dst->__refcnt);
171 if (WARN_ONCE(newrefcnt < 0, "dst_release underflow"))
172 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
173 __func__, dst, newrefcnt);
174 if (!newrefcnt)
175 call_rcu(&dst->rcu_head, dst_destroy_rcu);
176 }
177 }
178 EXPORT_SYMBOL(dst_release);
179
dst_release_immediate(struct dst_entry * dst)180 void dst_release_immediate(struct dst_entry *dst)
181 {
182 if (dst) {
183 int newrefcnt;
184
185 newrefcnt = atomic_dec_return(&dst->__refcnt);
186 if (WARN_ONCE(newrefcnt < 0, "dst_release_immediate underflow"))
187 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
188 __func__, dst, newrefcnt);
189 if (!newrefcnt)
190 dst_destroy(dst);
191 }
192 }
193 EXPORT_SYMBOL(dst_release_immediate);
194
dst_cow_metrics_generic(struct dst_entry * dst,unsigned long old)195 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
196 {
197 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
198
199 if (p) {
200 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
201 unsigned long prev, new;
202
203 refcount_set(&p->refcnt, 1);
204 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
205
206 new = (unsigned long) p;
207 prev = cmpxchg(&dst->_metrics, old, new);
208
209 if (prev != old) {
210 kfree(p);
211 p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
212 if (prev & DST_METRICS_READ_ONLY)
213 p = NULL;
214 } else if (prev & DST_METRICS_REFCOUNTED) {
215 if (refcount_dec_and_test(&old_p->refcnt))
216 kfree(old_p);
217 }
218 }
219 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
220 return (u32 *)p;
221 }
222 EXPORT_SYMBOL(dst_cow_metrics_generic);
223
224 /* Caller asserts that dst_metrics_read_only(dst) is false. */
__dst_destroy_metrics_generic(struct dst_entry * dst,unsigned long old)225 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
226 {
227 unsigned long prev, new;
228
229 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
230 prev = cmpxchg(&dst->_metrics, old, new);
231 if (prev == old)
232 kfree(__DST_METRICS_PTR(old));
233 }
234 EXPORT_SYMBOL(__dst_destroy_metrics_generic);
235
dst_blackhole_check(struct dst_entry * dst,u32 cookie)236 struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie)
237 {
238 return NULL;
239 }
240
dst_blackhole_cow_metrics(struct dst_entry * dst,unsigned long old)241 u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old)
242 {
243 return NULL;
244 }
245
dst_blackhole_neigh_lookup(const struct dst_entry * dst,struct sk_buff * skb,const void * daddr)246 struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
247 struct sk_buff *skb,
248 const void *daddr)
249 {
250 return NULL;
251 }
252
dst_blackhole_update_pmtu(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb,u32 mtu,bool confirm_neigh)253 void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
254 struct sk_buff *skb, u32 mtu,
255 bool confirm_neigh)
256 {
257 }
258 EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu);
259
dst_blackhole_redirect(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb)260 void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
261 struct sk_buff *skb)
262 {
263 }
264 EXPORT_SYMBOL_GPL(dst_blackhole_redirect);
265
dst_blackhole_mtu(const struct dst_entry * dst)266 unsigned int dst_blackhole_mtu(const struct dst_entry *dst)
267 {
268 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
269
270 return mtu ? : dst->dev->mtu;
271 }
272 EXPORT_SYMBOL_GPL(dst_blackhole_mtu);
273
274 static struct dst_ops dst_blackhole_ops = {
275 .family = AF_UNSPEC,
276 .neigh_lookup = dst_blackhole_neigh_lookup,
277 .check = dst_blackhole_check,
278 .cow_metrics = dst_blackhole_cow_metrics,
279 .update_pmtu = dst_blackhole_update_pmtu,
280 .redirect = dst_blackhole_redirect,
281 .mtu = dst_blackhole_mtu,
282 };
283
__metadata_dst_init(struct metadata_dst * md_dst,enum metadata_type type,u8 optslen)284 static void __metadata_dst_init(struct metadata_dst *md_dst,
285 enum metadata_type type, u8 optslen)
286 {
287 struct dst_entry *dst;
288
289 dst = &md_dst->dst;
290 dst_init(dst, &dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE,
291 DST_METADATA | DST_NOCOUNT);
292 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
293 md_dst->type = type;
294 }
295
metadata_dst_alloc(u8 optslen,enum metadata_type type,gfp_t flags)296 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
297 gfp_t flags)
298 {
299 struct metadata_dst *md_dst;
300
301 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
302 if (!md_dst)
303 return NULL;
304
305 __metadata_dst_init(md_dst, type, optslen);
306
307 return md_dst;
308 }
309 EXPORT_SYMBOL_GPL(metadata_dst_alloc);
310
metadata_dst_free(struct metadata_dst * md_dst)311 void metadata_dst_free(struct metadata_dst *md_dst)
312 {
313 #ifdef CONFIG_DST_CACHE
314 if (md_dst->type == METADATA_IP_TUNNEL)
315 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
316 #endif
317 kfree(md_dst);
318 }
319 EXPORT_SYMBOL_GPL(metadata_dst_free);
320
321 struct metadata_dst __percpu *
metadata_dst_alloc_percpu(u8 optslen,enum metadata_type type,gfp_t flags)322 metadata_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(sizeof(struct metadata_dst) + 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 }
337 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
338
metadata_dst_free_percpu(struct metadata_dst __percpu * md_dst)339 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
340 {
341 #ifdef CONFIG_DST_CACHE
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 if (one_md_dst->type == METADATA_IP_TUNNEL)
348 dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
349 }
350 #endif
351 free_percpu(md_dst);
352 }
353 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
354