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 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 atomic_set(&dst->__refcnt, initial_ref);
70 dst->__use = 0;
71 dst->lastuse = jiffies;
72 dst->flags = flags;
73 if (!(flags & DST_NOCOUNT))
74 dst_entries_add(ops, 1);
75 }
76 EXPORT_SYMBOL(dst_init);
77
dst_alloc(struct dst_ops * ops,struct net_device * dev,int initial_ref,int initial_obsolete,unsigned short flags)78 void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
79 int initial_ref, int initial_obsolete, unsigned short flags)
80 {
81 struct dst_entry *dst;
82
83 if (ops->gc &&
84 !(flags & DST_NOCOUNT) &&
85 dst_entries_get_fast(ops) > ops->gc_thresh) {
86 if (ops->gc(ops)) {
87 pr_notice_ratelimited("Route cache is full: consider increasing sysctl net.ipv6.route.max_size.\n");
88 return NULL;
89 }
90 }
91
92 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
93 if (!dst)
94 return NULL;
95
96 dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
97
98 return dst;
99 }
100 EXPORT_SYMBOL(dst_alloc);
101
dst_destroy(struct dst_entry * dst)102 struct dst_entry *dst_destroy(struct dst_entry * dst)
103 {
104 struct dst_entry *child = NULL;
105
106 smp_rmb();
107
108 #ifdef CONFIG_XFRM
109 if (dst->xfrm) {
110 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
111
112 child = xdst->child;
113 }
114 #endif
115 if (!(dst->flags & DST_NOCOUNT))
116 dst_entries_add(dst->ops, -1);
117
118 if (dst->ops->destroy)
119 dst->ops->destroy(dst);
120 netdev_put(dst->dev, &dst->dev_tracker);
121
122 lwtstate_put(dst->lwtstate);
123
124 if (dst->flags & DST_METADATA)
125 metadata_dst_free((struct metadata_dst *)dst);
126 else
127 kmem_cache_free(dst->ops->kmem_cachep, dst);
128
129 dst = child;
130 if (dst)
131 dst_release_immediate(dst);
132 return NULL;
133 }
134 EXPORT_SYMBOL(dst_destroy);
135
dst_destroy_rcu(struct rcu_head * head)136 static void dst_destroy_rcu(struct rcu_head *head)
137 {
138 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
139
140 dst = dst_destroy(dst);
141 }
142
143 /* Operations to mark dst as DEAD and clean up the net device referenced
144 * by dst:
145 * 1. put the dst under blackhole interface and discard all tx/rx packets
146 * on this route.
147 * 2. release the net_device
148 * This function should be called when removing routes from the fib tree
149 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
150 * make the next dst_ops->check() fail.
151 */
dst_dev_put(struct dst_entry * dst)152 void dst_dev_put(struct dst_entry *dst)
153 {
154 struct net_device *dev = dst->dev;
155
156 dst->obsolete = DST_OBSOLETE_DEAD;
157 if (dst->ops->ifdown)
158 dst->ops->ifdown(dst, dev, true);
159 dst->input = dst_discard;
160 dst->output = dst_discard_out;
161 dst->dev = blackhole_netdev;
162 netdev_ref_replace(dev, blackhole_netdev, &dst->dev_tracker,
163 GFP_ATOMIC);
164 }
165 EXPORT_SYMBOL(dst_dev_put);
166
dst_release(struct dst_entry * dst)167 void dst_release(struct dst_entry *dst)
168 {
169 if (dst) {
170 int newrefcnt;
171
172 newrefcnt = atomic_dec_return(&dst->__refcnt);
173 if (WARN_ONCE(newrefcnt < 0, "dst_release underflow"))
174 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
175 __func__, dst, newrefcnt);
176 if (!newrefcnt)
177 call_rcu(&dst->rcu_head, dst_destroy_rcu);
178 }
179 }
180 EXPORT_SYMBOL(dst_release);
181
dst_release_immediate(struct dst_entry * dst)182 void dst_release_immediate(struct dst_entry *dst)
183 {
184 if (dst) {
185 int newrefcnt;
186
187 newrefcnt = atomic_dec_return(&dst->__refcnt);
188 if (WARN_ONCE(newrefcnt < 0, "dst_release_immediate underflow"))
189 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
190 __func__, dst, newrefcnt);
191 if (!newrefcnt)
192 dst_destroy(dst);
193 }
194 }
195 EXPORT_SYMBOL(dst_release_immediate);
196
dst_cow_metrics_generic(struct dst_entry * dst,unsigned long old)197 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
198 {
199 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
200
201 if (p) {
202 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
203 unsigned long prev, new;
204
205 refcount_set(&p->refcnt, 1);
206 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
207
208 new = (unsigned long) p;
209 prev = cmpxchg(&dst->_metrics, old, new);
210
211 if (prev != old) {
212 kfree(p);
213 p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
214 if (prev & DST_METRICS_READ_ONLY)
215 p = NULL;
216 } else if (prev & DST_METRICS_REFCOUNTED) {
217 if (refcount_dec_and_test(&old_p->refcnt))
218 kfree(old_p);
219 }
220 }
221 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
222 return (u32 *)p;
223 }
224 EXPORT_SYMBOL(dst_cow_metrics_generic);
225
226 /* Caller asserts that dst_metrics_read_only(dst) is false. */
__dst_destroy_metrics_generic(struct dst_entry * dst,unsigned long old)227 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
228 {
229 unsigned long prev, new;
230
231 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
232 prev = cmpxchg(&dst->_metrics, old, new);
233 if (prev == old)
234 kfree(__DST_METRICS_PTR(old));
235 }
236 EXPORT_SYMBOL(__dst_destroy_metrics_generic);
237
dst_blackhole_check(struct dst_entry * dst,u32 cookie)238 struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie)
239 {
240 return NULL;
241 }
242
dst_blackhole_cow_metrics(struct dst_entry * dst,unsigned long old)243 u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old)
244 {
245 return NULL;
246 }
247
dst_blackhole_neigh_lookup(const struct dst_entry * dst,struct sk_buff * skb,const void * daddr)248 struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
249 struct sk_buff *skb,
250 const void *daddr)
251 {
252 return NULL;
253 }
254
dst_blackhole_update_pmtu(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb,u32 mtu,bool confirm_neigh)255 void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
256 struct sk_buff *skb, u32 mtu,
257 bool confirm_neigh)
258 {
259 }
260 EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu);
261
dst_blackhole_redirect(struct dst_entry * dst,struct sock * sk,struct sk_buff * skb)262 void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
263 struct sk_buff *skb)
264 {
265 }
266 EXPORT_SYMBOL_GPL(dst_blackhole_redirect);
267
dst_blackhole_mtu(const struct dst_entry * dst)268 unsigned int dst_blackhole_mtu(const struct dst_entry *dst)
269 {
270 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
271
272 return mtu ? : dst->dev->mtu;
273 }
274 EXPORT_SYMBOL_GPL(dst_blackhole_mtu);
275
276 static struct dst_ops dst_blackhole_ops = {
277 .family = AF_UNSPEC,
278 .neigh_lookup = dst_blackhole_neigh_lookup,
279 .check = dst_blackhole_check,
280 .cow_metrics = dst_blackhole_cow_metrics,
281 .update_pmtu = dst_blackhole_update_pmtu,
282 .redirect = dst_blackhole_redirect,
283 .mtu = dst_blackhole_mtu,
284 };
285
__metadata_dst_init(struct metadata_dst * md_dst,enum metadata_type type,u8 optslen)286 static void __metadata_dst_init(struct metadata_dst *md_dst,
287 enum metadata_type type, u8 optslen)
288 {
289 struct dst_entry *dst;
290
291 dst = &md_dst->dst;
292 dst_init(dst, &dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE,
293 DST_METADATA | DST_NOCOUNT);
294 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
295 md_dst->type = type;
296 }
297
metadata_dst_alloc(u8 optslen,enum metadata_type type,gfp_t flags)298 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
299 gfp_t flags)
300 {
301 struct metadata_dst *md_dst;
302
303 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
304 if (!md_dst)
305 return NULL;
306
307 __metadata_dst_init(md_dst, type, optslen);
308
309 return md_dst;
310 }
311 EXPORT_SYMBOL_GPL(metadata_dst_alloc);
312
metadata_dst_free(struct metadata_dst * md_dst)313 void metadata_dst_free(struct metadata_dst *md_dst)
314 {
315 #ifdef CONFIG_DST_CACHE
316 if (md_dst->type == METADATA_IP_TUNNEL)
317 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
318 #endif
319 kfree(md_dst);
320 }
321 EXPORT_SYMBOL_GPL(metadata_dst_free);
322
323 struct metadata_dst __percpu *
metadata_dst_alloc_percpu(u8 optslen,enum metadata_type type,gfp_t flags)324 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
325 {
326 int cpu;
327 struct metadata_dst __percpu *md_dst;
328
329 md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
330 __alignof__(struct metadata_dst), flags);
331 if (!md_dst)
332 return NULL;
333
334 for_each_possible_cpu(cpu)
335 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
336
337 return md_dst;
338 }
339 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
340
metadata_dst_free_percpu(struct metadata_dst __percpu * md_dst)341 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
342 {
343 #ifdef CONFIG_DST_CACHE
344 int cpu;
345
346 for_each_possible_cpu(cpu) {
347 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
348
349 if (one_md_dst->type == METADATA_IP_TUNNEL)
350 dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
351 }
352 #endif
353 free_percpu(md_dst);
354 }
355 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
356