]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - net/core/dst.c
Merge tag 'perf-core-for-mingo-4.21-20190104' of git://git.kernel.org/pub/scm/linux...
[mirror_ubuntu-eoan-kernel.git] / net / core / dst.c
CommitLineData
1da177e4
LT
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>
86bba269 12#include <linux/workqueue.h>
1da177e4
LT
13#include <linux/mm.h>
14#include <linux/module.h>
5a0e3ad6 15#include <linux/slab.h>
1da177e4 16#include <linux/netdevice.h>
1da177e4
LT
17#include <linux/skbuff.h>
18#include <linux/string.h>
19#include <linux/types.h>
e9dc8653 20#include <net/net_namespace.h>
2fc1b5dd 21#include <linux/sched.h>
268bb0ce 22#include <linux/prefetch.h>
61adedf3 23#include <net/lwtunnel.h>
b6ca8bd5 24#include <net/xfrm.h>
1da177e4
LT
25
26#include <net/dst.h>
f38a9eb1 27#include <net/dst_metadata.h>
1da177e4 28
86bba269
ED
29/*
30 * Theory of operations:
31 * 1) We use a list, protected by a spinlock, to add
32 * new entries from both BH and non-BH context.
33 * 2) In order to keep spinlock held for a small delay,
34 * we use a second list where are stored long lived
35 * entries, that are handled by the garbage collect thread
36 * fired by a workqueue.
37 * 3) This list is guarded by a mutex,
38 * so that the gc_task and dst_dev_event() can be synchronized.
1da177e4 39 */
1da177e4 40
86bba269
ED
41/*
42 * We want to keep lock & list close together
43 * to dirty as few cache lines as possible in __dst_free().
44 * As this is not a very strong hint, we dont force an alignment on SMP.
45 */
ede2059d 46int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
1da177e4
LT
47{
48 kfree_skb(skb);
49 return 0;
50}
ede2059d 51EXPORT_SYMBOL(dst_discard_out);
1da177e4 52
3fb07daf 53const struct dst_metrics dst_default_metrics = {
a37e6e34
ED
54 /* This initializer is needed to force linker to place this variable
55 * into const section. Otherwise it might end into bss section.
56 * We really want to avoid false sharing on this variable, and catch
57 * any writes on it.
58 */
9620fef2 59 .refcnt = REFCOUNT_INIT(1),
a37e6e34 60};
d4ead6b3 61EXPORT_SYMBOL(dst_default_metrics);
a37e6e34 62
f38a9eb1
TG
63void dst_init(struct dst_entry *dst, struct dst_ops *ops,
64 struct net_device *dev, int initial_ref, int initial_obsolete,
65 unsigned short flags)
1da177e4 66{
5c1e6aa3
DM
67 dst->dev = dev;
68 if (dev)
69 dev_hold(dev);
1da177e4 70 dst->ops = ops;
3fb07daf 71 dst_init_metrics(dst, dst_default_metrics.metrics, true);
cf911662 72 dst->expires = 0UL;
cf911662
DM
73#ifdef CONFIG_XFRM
74 dst->xfrm = NULL;
75#endif
5c1e6aa3 76 dst->input = dst_discard;
ede2059d 77 dst->output = dst_discard_out;
cf911662 78 dst->error = 0;
5c1e6aa3 79 dst->obsolete = initial_obsolete;
cf911662
DM
80 dst->header_len = 0;
81 dst->trailer_len = 0;
82#ifdef CONFIG_IP_ROUTE_CLASSID
83 dst->tclassid = 0;
1da177e4 84#endif
61adedf3 85 dst->lwtstate = NULL;
5c1e6aa3 86 atomic_set(&dst->__refcnt, initial_ref);
cf911662 87 dst->__use = 0;
5c1e6aa3
DM
88 dst->lastuse = jiffies;
89 dst->flags = flags;
957c665f
DM
90 if (!(flags & DST_NOCOUNT))
91 dst_entries_add(ops, 1);
f38a9eb1
TG
92}
93EXPORT_SYMBOL(dst_init);
94
95void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
96 int initial_ref, int initial_obsolete, unsigned short flags)
97{
98 struct dst_entry *dst;
99
100 if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
101 if (ops->gc(ops))
102 return NULL;
103 }
104
105 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
106 if (!dst)
107 return NULL;
108
109 dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
110
1da177e4
LT
111 return dst;
112}
598ed936 113EXPORT_SYMBOL(dst_alloc);
1da177e4 114
1da177e4
LT
115struct dst_entry *dst_destroy(struct dst_entry * dst)
116{
b92cf4aa 117 struct dst_entry *child = NULL;
1da177e4
LT
118
119 smp_rmb();
120
b92cf4aa 121#ifdef CONFIG_XFRM
b6ca8bd5
DM
122 if (dst->xfrm) {
123 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
124
125 child = xdst->child;
126 }
b92cf4aa 127#endif
957c665f
DM
128 if (!(dst->flags & DST_NOCOUNT))
129 dst_entries_add(dst->ops, -1);
1da177e4
LT
130
131 if (dst->ops->destroy)
132 dst->ops->destroy(dst);
133 if (dst->dev)
134 dev_put(dst->dev);
f38a9eb1 135
e252b3d1
WC
136 lwtstate_put(dst->lwtstate);
137
f38a9eb1 138 if (dst->flags & DST_METADATA)
d71785ff 139 metadata_dst_free((struct metadata_dst *)dst);
f38a9eb1
TG
140 else
141 kmem_cache_free(dst->ops->kmem_cachep, dst);
1da177e4
LT
142
143 dst = child;
52df157f
WW
144 if (dst)
145 dst_release_immediate(dst);
1da177e4
LT
146 return NULL;
147}
598ed936 148EXPORT_SYMBOL(dst_destroy);
1da177e4 149
f8864972
ED
150static void dst_destroy_rcu(struct rcu_head *head)
151{
152 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
153
154 dst = dst_destroy(dst);
f8864972
ED
155}
156
4a6ce2b6
WW
157/* Operations to mark dst as DEAD and clean up the net device referenced
158 * by dst:
159 * 1. put the dst under loopback interface and discard all tx/rx packets
160 * on this route.
161 * 2. release the net_device
162 * This function should be called when removing routes from the fib tree
163 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
164 * make the next dst_ops->check() fail.
165 */
166void dst_dev_put(struct dst_entry *dst)
167{
168 struct net_device *dev = dst->dev;
169
170 dst->obsolete = DST_OBSOLETE_DEAD;
171 if (dst->ops->ifdown)
172 dst->ops->ifdown(dst, dev, true);
173 dst->input = dst_discard;
174 dst->output = dst_discard_out;
175 dst->dev = dev_net(dst->dev)->loopback_dev;
176 dev_hold(dst->dev);
177 dev_put(dev);
178}
179EXPORT_SYMBOL(dst_dev_put);
180
8d330868
IJ
181void dst_release(struct dst_entry *dst)
182{
183 if (dst) {
598ed936 184 int newrefcnt;
ef711cf1 185
598ed936 186 newrefcnt = atomic_dec_return(&dst->__refcnt);
8bf4ada2
KK
187 if (unlikely(newrefcnt < 0))
188 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
189 __func__, dst, newrefcnt);
b2a9c0ed 190 if (!newrefcnt)
f8864972 191 call_rcu(&dst->rcu_head, dst_destroy_rcu);
8d330868
IJ
192 }
193}
194EXPORT_SYMBOL(dst_release);
195
5f56f409
WW
196void dst_release_immediate(struct dst_entry *dst)
197{
198 if (dst) {
199 int newrefcnt;
200
201 newrefcnt = atomic_dec_return(&dst->__refcnt);
202 if (unlikely(newrefcnt < 0))
203 net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
204 __func__, dst, newrefcnt);
205 if (!newrefcnt)
206 dst_destroy(dst);
207 }
208}
209EXPORT_SYMBOL(dst_release_immediate);
210
62fa8a84
DM
211u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
212{
3fb07daf 213 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
62fa8a84
DM
214
215 if (p) {
3fb07daf 216 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
62fa8a84
DM
217 unsigned long prev, new;
218
9620fef2 219 refcount_set(&p->refcnt, 1);
3fb07daf 220 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
62fa8a84
DM
221
222 new = (unsigned long) p;
223 prev = cmpxchg(&dst->_metrics, old, new);
224
225 if (prev != old) {
226 kfree(p);
3fb07daf 227 p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
62fa8a84
DM
228 if (prev & DST_METRICS_READ_ONLY)
229 p = NULL;
3fb07daf 230 } else if (prev & DST_METRICS_REFCOUNTED) {
9620fef2 231 if (refcount_dec_and_test(&old_p->refcnt))
3fb07daf 232 kfree(old_p);
62fa8a84
DM
233 }
234 }
3fb07daf
ED
235 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
236 return (u32 *)p;
62fa8a84
DM
237}
238EXPORT_SYMBOL(dst_cow_metrics_generic);
239
240/* Caller asserts that dst_metrics_read_only(dst) is false. */
241void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
242{
243 unsigned long prev, new;
244
3fb07daf 245 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
62fa8a84
DM
246 prev = cmpxchg(&dst->_metrics, old, new);
247 if (prev == old)
248 kfree(__DST_METRICS_PTR(old));
249}
250EXPORT_SYMBOL(__dst_destroy_metrics_generic);
251
f38a9eb1
TG
252static struct dst_ops md_dst_ops = {
253 .family = AF_UNSPEC,
254};
255
ede2059d 256static int dst_md_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
f38a9eb1
TG
257{
258 WARN_ONCE(1, "Attempting to call output on metadata dst\n");
259 kfree_skb(skb);
260 return 0;
261}
262
263static int dst_md_discard(struct sk_buff *skb)
264{
265 WARN_ONCE(1, "Attempting to call input on metadata dst\n");
266 kfree_skb(skb);
267 return 0;
268}
269
3fcece12
JK
270static void __metadata_dst_init(struct metadata_dst *md_dst,
271 enum metadata_type type, u8 optslen)
272
f38a9eb1 273{
f38a9eb1
TG
274 struct dst_entry *dst;
275
f38a9eb1
TG
276 dst = &md_dst->dst;
277 dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE,
a4c2fd7f 278 DST_METADATA | DST_NOCOUNT);
f38a9eb1
TG
279
280 dst->input = dst_md_discard;
ede2059d 281 dst->output = dst_md_discard_out;
f38a9eb1
TG
282
283 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
3fcece12 284 md_dst->type = type;
d3aa45ce
AS
285}
286
3fcece12
JK
287struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
288 gfp_t flags)
d3aa45ce
AS
289{
290 struct metadata_dst *md_dst;
291
292 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
293 if (!md_dst)
294 return NULL;
295
3fcece12 296 __metadata_dst_init(md_dst, type, optslen);
f38a9eb1
TG
297
298 return md_dst;
299}
300EXPORT_SYMBOL_GPL(metadata_dst_alloc);
301
d71785ff
PA
302void metadata_dst_free(struct metadata_dst *md_dst)
303{
304#ifdef CONFIG_DST_CACHE
e65a4955
DL
305 if (md_dst->type == METADATA_IP_TUNNEL)
306 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
d71785ff
PA
307#endif
308 kfree(md_dst);
309}
af308b94 310EXPORT_SYMBOL_GPL(metadata_dst_free);
d71785ff 311
3fcece12
JK
312struct metadata_dst __percpu *
313metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
d3aa45ce
AS
314{
315 int cpu;
316 struct metadata_dst __percpu *md_dst;
317
318 md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
319 __alignof__(struct metadata_dst), flags);
320 if (!md_dst)
321 return NULL;
322
323 for_each_possible_cpu(cpu)
3fcece12 324 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
d3aa45ce
AS
325
326 return md_dst;
327}
328EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
d66f2b91
JK
329
330void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
331{
833e0e2f 332#ifdef CONFIG_DST_CACHE
d66f2b91
JK
333 int cpu;
334
d66f2b91
JK
335 for_each_possible_cpu(cpu) {
336 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
337
338 if (one_md_dst->type == METADATA_IP_TUNNEL)
339 dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
340 }
341#endif
342 free_percpu(md_dst);
343}
344EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);