xref: /openbmc/linux/net/core/dst.c (revision 9df839a711aee437390b16ee39cf0b5c1620be6a)
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_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 	rcuref_init(&dst->__rcuref, 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 
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 		ops->gc(ops);
87 
88 	dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
89 	if (!dst)
90 		return NULL;
91 
92 	dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
93 
94 	return dst;
95 }
96 EXPORT_SYMBOL(dst_alloc);
97 
98 struct dst_entry *dst_destroy(struct dst_entry * dst)
99 {
100 	struct dst_entry *child = NULL;
101 
102 	smp_rmb();
103 
104 #ifdef CONFIG_XFRM
105 	if (dst->xfrm) {
106 		struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
107 
108 		child = xdst->child;
109 	}
110 #endif
111 	if (!(dst->flags & DST_NOCOUNT))
112 		dst_entries_add(dst->ops, -1);
113 
114 	if (dst->ops->destroy)
115 		dst->ops->destroy(dst);
116 	netdev_put(dst->dev, &dst->dev_tracker);
117 
118 	lwtstate_put(dst->lwtstate);
119 
120 	if (dst->flags & DST_METADATA)
121 		metadata_dst_free((struct metadata_dst *)dst);
122 	else
123 		kmem_cache_free(dst->ops->kmem_cachep, dst);
124 
125 	dst = child;
126 	if (dst)
127 		dst_release_immediate(dst);
128 	return NULL;
129 }
130 EXPORT_SYMBOL(dst_destroy);
131 
132 static void dst_destroy_rcu(struct rcu_head *head)
133 {
134 	struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
135 
136 	dst = dst_destroy(dst);
137 }
138 
139 /* Operations to mark dst as DEAD and clean up the net device referenced
140  * by dst:
141  * 1. put the dst under blackhole interface and discard all tx/rx packets
142  *    on this route.
143  * 2. release the net_device
144  * This function should be called when removing routes from the fib tree
145  * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
146  * make the next dst_ops->check() fail.
147  */
148 void dst_dev_put(struct dst_entry *dst)
149 {
150 	struct net_device *dev = dst->dev;
151 
152 	dst->obsolete = DST_OBSOLETE_DEAD;
153 	if (dst->ops->ifdown)
154 		dst->ops->ifdown(dst, dev, true);
155 	dst->input = dst_discard;
156 	dst->output = dst_discard_out;
157 	dst->dev = blackhole_netdev;
158 	netdev_ref_replace(dev, blackhole_netdev, &dst->dev_tracker,
159 			   GFP_ATOMIC);
160 }
161 EXPORT_SYMBOL(dst_dev_put);
162 
163 void dst_release(struct dst_entry *dst)
164 {
165 	if (dst && rcuref_put(&dst->__rcuref))
166 		call_rcu_hurry(&dst->rcu_head, dst_destroy_rcu);
167 }
168 EXPORT_SYMBOL(dst_release);
169 
170 void dst_release_immediate(struct dst_entry *dst)
171 {
172 	if (dst && rcuref_put(&dst->__rcuref))
173 		dst_destroy(dst);
174 }
175 EXPORT_SYMBOL(dst_release_immediate);
176 
177 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
178 {
179 	struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
180 
181 	if (p) {
182 		struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
183 		unsigned long prev, new;
184 
185 		refcount_set(&p->refcnt, 1);
186 		memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
187 
188 		new = (unsigned long) p;
189 		prev = cmpxchg(&dst->_metrics, old, new);
190 
191 		if (prev != old) {
192 			kfree(p);
193 			p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
194 			if (prev & DST_METRICS_READ_ONLY)
195 				p = NULL;
196 		} else if (prev & DST_METRICS_REFCOUNTED) {
197 			if (refcount_dec_and_test(&old_p->refcnt))
198 				kfree(old_p);
199 		}
200 	}
201 	BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
202 	return (u32 *)p;
203 }
204 EXPORT_SYMBOL(dst_cow_metrics_generic);
205 
206 /* Caller asserts that dst_metrics_read_only(dst) is false.  */
207 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
208 {
209 	unsigned long prev, new;
210 
211 	new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
212 	prev = cmpxchg(&dst->_metrics, old, new);
213 	if (prev == old)
214 		kfree(__DST_METRICS_PTR(old));
215 }
216 EXPORT_SYMBOL(__dst_destroy_metrics_generic);
217 
218 struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie)
219 {
220 	return NULL;
221 }
222 
223 u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old)
224 {
225 	return NULL;
226 }
227 
228 struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
229 					     struct sk_buff *skb,
230 					     const void *daddr)
231 {
232 	return NULL;
233 }
234 
235 void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
236 			       struct sk_buff *skb, u32 mtu,
237 			       bool confirm_neigh)
238 {
239 }
240 EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu);
241 
242 void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
243 			    struct sk_buff *skb)
244 {
245 }
246 EXPORT_SYMBOL_GPL(dst_blackhole_redirect);
247 
248 unsigned int dst_blackhole_mtu(const struct dst_entry *dst)
249 {
250 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
251 
252 	return mtu ? : dst->dev->mtu;
253 }
254 EXPORT_SYMBOL_GPL(dst_blackhole_mtu);
255 
256 static struct dst_ops dst_blackhole_ops = {
257 	.family		= AF_UNSPEC,
258 	.neigh_lookup	= dst_blackhole_neigh_lookup,
259 	.check		= dst_blackhole_check,
260 	.cow_metrics	= dst_blackhole_cow_metrics,
261 	.update_pmtu	= dst_blackhole_update_pmtu,
262 	.redirect	= dst_blackhole_redirect,
263 	.mtu		= dst_blackhole_mtu,
264 };
265 
266 static void __metadata_dst_init(struct metadata_dst *md_dst,
267 				enum metadata_type type, u8 optslen)
268 {
269 	struct dst_entry *dst;
270 
271 	dst = &md_dst->dst;
272 	dst_init(dst, &dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE,
273 		 DST_METADATA | DST_NOCOUNT);
274 	memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
275 	md_dst->type = type;
276 }
277 
278 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
279 					gfp_t flags)
280 {
281 	struct metadata_dst *md_dst;
282 
283 	md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
284 	if (!md_dst)
285 		return NULL;
286 
287 	__metadata_dst_init(md_dst, type, optslen);
288 
289 	return md_dst;
290 }
291 EXPORT_SYMBOL_GPL(metadata_dst_alloc);
292 
293 void metadata_dst_free(struct metadata_dst *md_dst)
294 {
295 #ifdef CONFIG_DST_CACHE
296 	if (md_dst->type == METADATA_IP_TUNNEL)
297 		dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
298 #endif
299 	if (md_dst->type == METADATA_XFRM)
300 		dst_release(md_dst->u.xfrm_info.dst_orig);
301 	kfree(md_dst);
302 }
303 EXPORT_SYMBOL_GPL(metadata_dst_free);
304 
305 struct metadata_dst __percpu *
306 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
307 {
308 	int cpu;
309 	struct metadata_dst __percpu *md_dst;
310 
311 	md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
312 				    __alignof__(struct metadata_dst), flags);
313 	if (!md_dst)
314 		return NULL;
315 
316 	for_each_possible_cpu(cpu)
317 		__metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
318 
319 	return md_dst;
320 }
321 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
322 
323 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
324 {
325 	int cpu;
326 
327 	for_each_possible_cpu(cpu) {
328 		struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
329 
330 #ifdef CONFIG_DST_CACHE
331 		if (one_md_dst->type == METADATA_IP_TUNNEL)
332 			dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
333 #endif
334 		if (one_md_dst->type == METADATA_XFRM)
335 			dst_release(one_md_dst->u.xfrm_info.dst_orig);
336 	}
337 	free_percpu(md_dst);
338 }
339 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
340