Loading...
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_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}
77EXPORT_SYMBOL(dst_init);
78
79void *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 if (ops->gc(ops)) {
88 pr_notice_ratelimited("Route cache is full: consider increasing sysctl net.ipv6.route.max_size.\n");
89 return NULL;
90 }
91 }
92
93 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
94 if (!dst)
95 return NULL;
96
97 dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
98
99 return dst;
100}
101EXPORT_SYMBOL(dst_alloc);
102
103struct dst_entry *dst_destroy(struct dst_entry * dst)
104{
105 struct dst_entry *child = NULL;
106
107 smp_rmb();
108
109#ifdef CONFIG_XFRM
110 if (dst->xfrm) {
111 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
112
113 child = xdst->child;
114 }
115#endif
116 if (!(dst->flags & DST_NOCOUNT))
117 dst_entries_add(dst->ops, -1);
118
119 if (dst->ops->destroy)
120 dst->ops->destroy(dst);
121 if (dst->dev)
122 dev_put(dst->dev);
123
124 lwtstate_put(dst->lwtstate);
125
126 if (dst->flags & DST_METADATA)
127 metadata_dst_free((struct metadata_dst *)dst);
128 else
129 kmem_cache_free(dst->ops->kmem_cachep, dst);
130
131 dst = child;
132 if (dst)
133 dst_release_immediate(dst);
134 return NULL;
135}
136EXPORT_SYMBOL(dst_destroy);
137
138static void dst_destroy_rcu(struct rcu_head *head)
139{
140 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
141
142 dst = dst_destroy(dst);
143}
144
145/* Operations to mark dst as DEAD and clean up the net device referenced
146 * by dst:
147 * 1. put the dst under blackhole interface and discard all tx/rx packets
148 * on this route.
149 * 2. release the net_device
150 * This function should be called when removing routes from the fib tree
151 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
152 * make the next dst_ops->check() fail.
153 */
154void dst_dev_put(struct dst_entry *dst)
155{
156 struct net_device *dev = dst->dev;
157
158 dst->obsolete = DST_OBSOLETE_DEAD;
159 if (dst->ops->ifdown)
160 dst->ops->ifdown(dst, dev, true);
161 dst->input = dst_discard;
162 dst->output = dst_discard_out;
163 dst->dev = blackhole_netdev;
164 dev_hold(dst->dev);
165 dev_put(dev);
166}
167EXPORT_SYMBOL(dst_dev_put);
168
169void dst_release(struct dst_entry *dst)
170{
171 if (dst) {
172 int newrefcnt;
173
174 newrefcnt = atomic_dec_return(&dst->__refcnt);
175 if (WARN_ONCE(newrefcnt < 0, "dst_release underflow"))
176 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
177 __func__, dst, newrefcnt);
178 if (!newrefcnt)
179 call_rcu(&dst->rcu_head, dst_destroy_rcu);
180 }
181}
182EXPORT_SYMBOL(dst_release);
183
184void dst_release_immediate(struct dst_entry *dst)
185{
186 if (dst) {
187 int newrefcnt;
188
189 newrefcnt = atomic_dec_return(&dst->__refcnt);
190 if (WARN_ONCE(newrefcnt < 0, "dst_release_immediate underflow"))
191 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
192 __func__, dst, newrefcnt);
193 if (!newrefcnt)
194 dst_destroy(dst);
195 }
196}
197EXPORT_SYMBOL(dst_release_immediate);
198
199u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
200{
201 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
202
203 if (p) {
204 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
205 unsigned long prev, new;
206
207 refcount_set(&p->refcnt, 1);
208 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
209
210 new = (unsigned long) p;
211 prev = cmpxchg(&dst->_metrics, old, new);
212
213 if (prev != old) {
214 kfree(p);
215 p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
216 if (prev & DST_METRICS_READ_ONLY)
217 p = NULL;
218 } else if (prev & DST_METRICS_REFCOUNTED) {
219 if (refcount_dec_and_test(&old_p->refcnt))
220 kfree(old_p);
221 }
222 }
223 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
224 return (u32 *)p;
225}
226EXPORT_SYMBOL(dst_cow_metrics_generic);
227
228/* Caller asserts that dst_metrics_read_only(dst) is false. */
229void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
230{
231 unsigned long prev, new;
232
233 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
234 prev = cmpxchg(&dst->_metrics, old, new);
235 if (prev == old)
236 kfree(__DST_METRICS_PTR(old));
237}
238EXPORT_SYMBOL(__dst_destroy_metrics_generic);
239
240struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie)
241{
242 return NULL;
243}
244
245u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old)
246{
247 return NULL;
248}
249
250struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
251 struct sk_buff *skb,
252 const void *daddr)
253{
254 return NULL;
255}
256
257void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
258 struct sk_buff *skb, u32 mtu,
259 bool confirm_neigh)
260{
261}
262EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu);
263
264void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
265 struct sk_buff *skb)
266{
267}
268EXPORT_SYMBOL_GPL(dst_blackhole_redirect);
269
270unsigned int dst_blackhole_mtu(const struct dst_entry *dst)
271{
272 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
273
274 return mtu ? : dst->dev->mtu;
275}
276EXPORT_SYMBOL_GPL(dst_blackhole_mtu);
277
278static struct dst_ops dst_blackhole_ops = {
279 .family = AF_UNSPEC,
280 .neigh_lookup = dst_blackhole_neigh_lookup,
281 .check = dst_blackhole_check,
282 .cow_metrics = dst_blackhole_cow_metrics,
283 .update_pmtu = dst_blackhole_update_pmtu,
284 .redirect = dst_blackhole_redirect,
285 .mtu = dst_blackhole_mtu,
286};
287
288static void __metadata_dst_init(struct metadata_dst *md_dst,
289 enum metadata_type type, u8 optslen)
290{
291 struct dst_entry *dst;
292
293 dst = &md_dst->dst;
294 dst_init(dst, &dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE,
295 DST_METADATA | DST_NOCOUNT);
296 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
297 md_dst->type = type;
298}
299
300struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
301 gfp_t flags)
302{
303 struct metadata_dst *md_dst;
304
305 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
306 if (!md_dst)
307 return NULL;
308
309 __metadata_dst_init(md_dst, type, optslen);
310
311 return md_dst;
312}
313EXPORT_SYMBOL_GPL(metadata_dst_alloc);
314
315void metadata_dst_free(struct metadata_dst *md_dst)
316{
317#ifdef CONFIG_DST_CACHE
318 if (md_dst->type == METADATA_IP_TUNNEL)
319 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
320#endif
321 kfree(md_dst);
322}
323EXPORT_SYMBOL_GPL(metadata_dst_free);
324
325struct metadata_dst __percpu *
326metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
327{
328 int cpu;
329 struct metadata_dst __percpu *md_dst;
330
331 md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
332 __alignof__(struct metadata_dst), flags);
333 if (!md_dst)
334 return NULL;
335
336 for_each_possible_cpu(cpu)
337 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
338
339 return md_dst;
340}
341EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
342
343void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
344{
345#ifdef CONFIG_DST_CACHE
346 int cpu;
347
348 for_each_possible_cpu(cpu) {
349 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
350
351 if (one_md_dst->type == METADATA_IP_TUNNEL)
352 dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
353 }
354#endif
355 free_percpu(md_dst);
356}
357EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
1/*
2 * net/core/dst.c Protocol independent destination cache.
3 *
4 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
5 *
6 */
7
8#include <linux/bitops.h>
9#include <linux/errno.h>
10#include <linux/init.h>
11#include <linux/kernel.h>
12#include <linux/workqueue.h>
13#include <linux/mm.h>
14#include <linux/module.h>
15#include <linux/slab.h>
16#include <linux/netdevice.h>
17#include <linux/skbuff.h>
18#include <linux/string.h>
19#include <linux/types.h>
20#include <net/net_namespace.h>
21#include <linux/sched.h>
22#include <linux/prefetch.h>
23#include <net/lwtunnel.h>
24
25#include <net/dst.h>
26#include <net/dst_metadata.h>
27
28/*
29 * Theory of operations:
30 * 1) We use a list, protected by a spinlock, to add
31 * new entries from both BH and non-BH context.
32 * 2) In order to keep spinlock held for a small delay,
33 * we use a second list where are stored long lived
34 * entries, that are handled by the garbage collect thread
35 * fired by a workqueue.
36 * 3) This list is guarded by a mutex,
37 * so that the gc_task and dst_dev_event() can be synchronized.
38 */
39
40/*
41 * We want to keep lock & list close together
42 * to dirty as few cache lines as possible in __dst_free().
43 * As this is not a very strong hint, we dont force an alignment on SMP.
44 */
45static struct {
46 spinlock_t lock;
47 struct dst_entry *list;
48 unsigned long timer_inc;
49 unsigned long timer_expires;
50} dst_garbage = {
51 .lock = __SPIN_LOCK_UNLOCKED(dst_garbage.lock),
52 .timer_inc = DST_GC_MAX,
53};
54static void dst_gc_task(struct work_struct *work);
55static void ___dst_free(struct dst_entry *dst);
56
57static DECLARE_DELAYED_WORK(dst_gc_work, dst_gc_task);
58
59static DEFINE_MUTEX(dst_gc_mutex);
60/*
61 * long lived entries are maintained in this list, guarded by dst_gc_mutex
62 */
63static struct dst_entry *dst_busy_list;
64
65static void dst_gc_task(struct work_struct *work)
66{
67 int delayed = 0;
68 int work_performed = 0;
69 unsigned long expires = ~0L;
70 struct dst_entry *dst, *next, head;
71 struct dst_entry *last = &head;
72
73 mutex_lock(&dst_gc_mutex);
74 next = dst_busy_list;
75
76loop:
77 while ((dst = next) != NULL) {
78 next = dst->next;
79 prefetch(&next->next);
80 cond_resched();
81 if (likely(atomic_read(&dst->__refcnt))) {
82 last->next = dst;
83 last = dst;
84 delayed++;
85 continue;
86 }
87 work_performed++;
88
89 dst = dst_destroy(dst);
90 if (dst) {
91 /* NOHASH and still referenced. Unless it is already
92 * on gc list, invalidate it and add to gc list.
93 *
94 * Note: this is temporary. Actually, NOHASH dst's
95 * must be obsoleted when parent is obsoleted.
96 * But we do not have state "obsoleted, but
97 * referenced by parent", so it is right.
98 */
99 if (dst->obsolete > 0)
100 continue;
101
102 ___dst_free(dst);
103 dst->next = next;
104 next = dst;
105 }
106 }
107
108 spin_lock_bh(&dst_garbage.lock);
109 next = dst_garbage.list;
110 if (next) {
111 dst_garbage.list = NULL;
112 spin_unlock_bh(&dst_garbage.lock);
113 goto loop;
114 }
115 last->next = NULL;
116 dst_busy_list = head.next;
117 if (!dst_busy_list)
118 dst_garbage.timer_inc = DST_GC_MAX;
119 else {
120 /*
121 * if we freed less than 1/10 of delayed entries,
122 * we can sleep longer.
123 */
124 if (work_performed <= delayed/10) {
125 dst_garbage.timer_expires += dst_garbage.timer_inc;
126 if (dst_garbage.timer_expires > DST_GC_MAX)
127 dst_garbage.timer_expires = DST_GC_MAX;
128 dst_garbage.timer_inc += DST_GC_INC;
129 } else {
130 dst_garbage.timer_inc = DST_GC_INC;
131 dst_garbage.timer_expires = DST_GC_MIN;
132 }
133 expires = dst_garbage.timer_expires;
134 /*
135 * if the next desired timer is more than 4 seconds in the
136 * future then round the timer to whole seconds
137 */
138 if (expires > 4*HZ)
139 expires = round_jiffies_relative(expires);
140 schedule_delayed_work(&dst_gc_work, expires);
141 }
142
143 spin_unlock_bh(&dst_garbage.lock);
144 mutex_unlock(&dst_gc_mutex);
145}
146
147int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
148{
149 kfree_skb(skb);
150 return 0;
151}
152EXPORT_SYMBOL(dst_discard_out);
153
154const u32 dst_default_metrics[RTAX_MAX + 1] = {
155 /* This initializer is needed to force linker to place this variable
156 * into const section. Otherwise it might end into bss section.
157 * We really want to avoid false sharing on this variable, and catch
158 * any writes on it.
159 */
160 [RTAX_MAX] = 0xdeadbeef,
161};
162
163void dst_init(struct dst_entry *dst, struct dst_ops *ops,
164 struct net_device *dev, int initial_ref, int initial_obsolete,
165 unsigned short flags)
166{
167 dst->child = NULL;
168 dst->dev = dev;
169 if (dev)
170 dev_hold(dev);
171 dst->ops = ops;
172 dst_init_metrics(dst, dst_default_metrics, true);
173 dst->expires = 0UL;
174 dst->path = dst;
175 dst->from = NULL;
176#ifdef CONFIG_XFRM
177 dst->xfrm = NULL;
178#endif
179 dst->input = dst_discard;
180 dst->output = dst_discard_out;
181 dst->error = 0;
182 dst->obsolete = initial_obsolete;
183 dst->header_len = 0;
184 dst->trailer_len = 0;
185#ifdef CONFIG_IP_ROUTE_CLASSID
186 dst->tclassid = 0;
187#endif
188 dst->lwtstate = NULL;
189 atomic_set(&dst->__refcnt, initial_ref);
190 dst->__use = 0;
191 dst->lastuse = jiffies;
192 dst->flags = flags;
193 dst->pending_confirm = 0;
194 dst->next = NULL;
195 if (!(flags & DST_NOCOUNT))
196 dst_entries_add(ops, 1);
197}
198EXPORT_SYMBOL(dst_init);
199
200void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
201 int initial_ref, int initial_obsolete, unsigned short flags)
202{
203 struct dst_entry *dst;
204
205 if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
206 if (ops->gc(ops))
207 return NULL;
208 }
209
210 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
211 if (!dst)
212 return NULL;
213
214 dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
215
216 return dst;
217}
218EXPORT_SYMBOL(dst_alloc);
219
220static void ___dst_free(struct dst_entry *dst)
221{
222 /* The first case (dev==NULL) is required, when
223 protocol module is unloaded.
224 */
225 if (dst->dev == NULL || !(dst->dev->flags&IFF_UP)) {
226 dst->input = dst_discard;
227 dst->output = dst_discard_out;
228 }
229 dst->obsolete = DST_OBSOLETE_DEAD;
230}
231
232void __dst_free(struct dst_entry *dst)
233{
234 spin_lock_bh(&dst_garbage.lock);
235 ___dst_free(dst);
236 dst->next = dst_garbage.list;
237 dst_garbage.list = dst;
238 if (dst_garbage.timer_inc > DST_GC_INC) {
239 dst_garbage.timer_inc = DST_GC_INC;
240 dst_garbage.timer_expires = DST_GC_MIN;
241 mod_delayed_work(system_wq, &dst_gc_work,
242 dst_garbage.timer_expires);
243 }
244 spin_unlock_bh(&dst_garbage.lock);
245}
246EXPORT_SYMBOL(__dst_free);
247
248struct dst_entry *dst_destroy(struct dst_entry * dst)
249{
250 struct dst_entry *child;
251
252 smp_rmb();
253
254again:
255 child = dst->child;
256
257 if (!(dst->flags & DST_NOCOUNT))
258 dst_entries_add(dst->ops, -1);
259
260 if (dst->ops->destroy)
261 dst->ops->destroy(dst);
262 if (dst->dev)
263 dev_put(dst->dev);
264
265 lwtstate_put(dst->lwtstate);
266
267 if (dst->flags & DST_METADATA)
268 metadata_dst_free((struct metadata_dst *)dst);
269 else
270 kmem_cache_free(dst->ops->kmem_cachep, dst);
271
272 dst = child;
273 if (dst) {
274 int nohash = dst->flags & DST_NOHASH;
275
276 if (atomic_dec_and_test(&dst->__refcnt)) {
277 /* We were real parent of this dst, so kill child. */
278 if (nohash)
279 goto again;
280 } else {
281 /* Child is still referenced, return it for freeing. */
282 if (nohash)
283 return dst;
284 /* Child is still in his hash table */
285 }
286 }
287 return NULL;
288}
289EXPORT_SYMBOL(dst_destroy);
290
291static void dst_destroy_rcu(struct rcu_head *head)
292{
293 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
294
295 dst = dst_destroy(dst);
296 if (dst)
297 __dst_free(dst);
298}
299
300void dst_release(struct dst_entry *dst)
301{
302 if (dst) {
303 int newrefcnt;
304 unsigned short nocache = dst->flags & DST_NOCACHE;
305
306 newrefcnt = atomic_dec_return(&dst->__refcnt);
307 if (unlikely(newrefcnt < 0))
308 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
309 __func__, dst, newrefcnt);
310 if (!newrefcnt && unlikely(nocache))
311 call_rcu(&dst->rcu_head, dst_destroy_rcu);
312 }
313}
314EXPORT_SYMBOL(dst_release);
315
316u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
317{
318 u32 *p = kmalloc(sizeof(u32) * RTAX_MAX, GFP_ATOMIC);
319
320 if (p) {
321 u32 *old_p = __DST_METRICS_PTR(old);
322 unsigned long prev, new;
323
324 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
325
326 new = (unsigned long) p;
327 prev = cmpxchg(&dst->_metrics, old, new);
328
329 if (prev != old) {
330 kfree(p);
331 p = __DST_METRICS_PTR(prev);
332 if (prev & DST_METRICS_READ_ONLY)
333 p = NULL;
334 }
335 }
336 return p;
337}
338EXPORT_SYMBOL(dst_cow_metrics_generic);
339
340/* Caller asserts that dst_metrics_read_only(dst) is false. */
341void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
342{
343 unsigned long prev, new;
344
345 new = ((unsigned long) dst_default_metrics) | DST_METRICS_READ_ONLY;
346 prev = cmpxchg(&dst->_metrics, old, new);
347 if (prev == old)
348 kfree(__DST_METRICS_PTR(old));
349}
350EXPORT_SYMBOL(__dst_destroy_metrics_generic);
351
352static struct dst_ops md_dst_ops = {
353 .family = AF_UNSPEC,
354};
355
356static int dst_md_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
357{
358 WARN_ONCE(1, "Attempting to call output on metadata dst\n");
359 kfree_skb(skb);
360 return 0;
361}
362
363static int dst_md_discard(struct sk_buff *skb)
364{
365 WARN_ONCE(1, "Attempting to call input on metadata dst\n");
366 kfree_skb(skb);
367 return 0;
368}
369
370static void __metadata_dst_init(struct metadata_dst *md_dst, u8 optslen)
371{
372 struct dst_entry *dst;
373
374 dst = &md_dst->dst;
375 dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE,
376 DST_METADATA | DST_NOCACHE | DST_NOCOUNT);
377
378 dst->input = dst_md_discard;
379 dst->output = dst_md_discard_out;
380
381 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
382}
383
384struct metadata_dst *metadata_dst_alloc(u8 optslen, gfp_t flags)
385{
386 struct metadata_dst *md_dst;
387
388 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
389 if (!md_dst)
390 return NULL;
391
392 __metadata_dst_init(md_dst, optslen);
393
394 return md_dst;
395}
396EXPORT_SYMBOL_GPL(metadata_dst_alloc);
397
398void metadata_dst_free(struct metadata_dst *md_dst)
399{
400#ifdef CONFIG_DST_CACHE
401 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
402#endif
403 kfree(md_dst);
404}
405
406struct metadata_dst __percpu *metadata_dst_alloc_percpu(u8 optslen, gfp_t flags)
407{
408 int cpu;
409 struct metadata_dst __percpu *md_dst;
410
411 md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
412 __alignof__(struct metadata_dst), flags);
413 if (!md_dst)
414 return NULL;
415
416 for_each_possible_cpu(cpu)
417 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), optslen);
418
419 return md_dst;
420}
421EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
422
423/* Dirty hack. We did it in 2.2 (in __dst_free),
424 * we have _very_ good reasons not to repeat
425 * this mistake in 2.3, but we have no choice
426 * now. _It_ _is_ _explicit_ _deliberate_
427 * _race_ _condition_.
428 *
429 * Commented and originally written by Alexey.
430 */
431static void dst_ifdown(struct dst_entry *dst, struct net_device *dev,
432 int unregister)
433{
434 if (dst->ops->ifdown)
435 dst->ops->ifdown(dst, dev, unregister);
436
437 if (dev != dst->dev)
438 return;
439
440 if (!unregister) {
441 dst->input = dst_discard;
442 dst->output = dst_discard_out;
443 } else {
444 dst->dev = dev_net(dst->dev)->loopback_dev;
445 dev_hold(dst->dev);
446 dev_put(dev);
447 }
448}
449
450static int dst_dev_event(struct notifier_block *this, unsigned long event,
451 void *ptr)
452{
453 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
454 struct dst_entry *dst, *last = NULL;
455
456 switch (event) {
457 case NETDEV_UNREGISTER_FINAL:
458 case NETDEV_DOWN:
459 mutex_lock(&dst_gc_mutex);
460 for (dst = dst_busy_list; dst; dst = dst->next) {
461 last = dst;
462 dst_ifdown(dst, dev, event != NETDEV_DOWN);
463 }
464
465 spin_lock_bh(&dst_garbage.lock);
466 dst = dst_garbage.list;
467 dst_garbage.list = NULL;
468 spin_unlock_bh(&dst_garbage.lock);
469
470 if (last)
471 last->next = dst;
472 else
473 dst_busy_list = dst;
474 for (; dst; dst = dst->next)
475 dst_ifdown(dst, dev, event != NETDEV_DOWN);
476 mutex_unlock(&dst_gc_mutex);
477 break;
478 }
479 return NOTIFY_DONE;
480}
481
482static struct notifier_block dst_dev_notifier = {
483 .notifier_call = dst_dev_event,
484 .priority = -10, /* must be called after other network notifiers */
485};
486
487void __init dst_subsys_init(void)
488{
489 register_netdevice_notifier(&dst_dev_notifier);
490}