]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - include/net/dst.h
net: Do delayed neigh confirmation.
[mirror_ubuntu-zesty-kernel.git] / include / net / dst.h
CommitLineData
1da177e4
LT
1/*
2 * net/dst.h Protocol independent destination cache definitions.
3 *
4 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
5 *
6 */
7
8#ifndef _NET_DST_H
9#define _NET_DST_H
10
86393e52 11#include <net/dst_ops.h>
14c85021 12#include <linux/netdevice.h>
1da177e4
LT
13#include <linux/rtnetlink.h>
14#include <linux/rcupdate.h>
187f1882 15#include <linux/bug.h>
1da177e4
LT
16#include <linux/jiffies.h>
17#include <net/neighbour.h>
18#include <asm/processor.h>
19
1da177e4
LT
20#define DST_GC_MIN (HZ/10)
21#define DST_GC_INC (HZ/2)
22#define DST_GC_MAX (120*HZ)
23
24/* Each dst_entry has reference count and sits in some parent list(s).
25 * When it is removed from parent list, it is "freed" (dst_free).
26 * After this it enters dead state (dst->obsolete > 0) and if its refcnt
27 * is zero, it can be destroyed immediately, otherwise it is added
28 * to gc list and garbage collector periodically checks the refcnt.
29 */
30
31struct sk_buff;
32
fd2c3ef7 33struct dst_entry {
1e19e02c 34 struct rcu_head rcu_head;
1da177e4
LT
35 struct dst_entry *child;
36 struct net_device *dev;
62fa8a84
DM
37 struct dst_ops *ops;
38 unsigned long _metrics;
1716a961
G
39 union {
40 unsigned long expires;
41 /* point to where the dst_entry copied from */
42 struct dst_entry *from;
43 };
f1dd9c37 44 struct dst_entry *path;
f2c31e32 45 struct neighbour __rcu *_neighbour;
def8b4fa 46#ifdef CONFIG_XFRM
1da177e4 47 struct xfrm_state *xfrm;
5635c10d
ED
48#else
49 void *__pad1;
def8b4fa 50#endif
7f95e188
EZ
51 int (*input)(struct sk_buff *);
52 int (*output)(struct sk_buff *);
1da177e4 53
5110effe 54 unsigned short flags;
f6b72b62
DM
55#define DST_HOST 0x0001
56#define DST_NOXFRM 0x0002
57#define DST_NOPOLICY 0x0004
58#define DST_NOHASH 0x0008
59#define DST_NOCACHE 0x0010
60#define DST_NOCOUNT 0x0020
e688a604 61#define DST_NOPEER 0x0040
a881e963 62#define DST_FAKE_RTABLE 0x0080
0c183379 63#define DST_XFRM_TUNNEL 0x0100
f6b72b62 64
5110effe
DM
65 unsigned short pending_confirm;
66
62fa8a84
DM
67 short error;
68 short obsolete;
69 unsigned short header_len; /* more space at head required */
70 unsigned short trailer_len; /* space to reserve at tail */
c7066f70 71#ifdef CONFIG_IP_ROUTE_CLASSID
f1dd9c37 72 __u32 tclassid;
5635c10d
ED
73#else
74 __u32 __pad2;
f1dd9c37
ZY
75#endif
76
5635c10d
ED
77 /*
78 * Align __refcnt to a 64 bytes alignment
79 * (L1_CACHE_SIZE would be too much)
80 */
81#ifdef CONFIG_64BIT
f6b72b62 82 long __pad_to_align_refcnt[2];
5635c10d 83#endif
f1dd9c37
ZY
84 /*
85 * __refcnt wants to be on a different cache line from
86 * input/output/ops or performance tanks badly
87 */
1e19e02c
ED
88 atomic_t __refcnt; /* client references */
89 int __use;
f1dd9c37 90 unsigned long lastuse;
1e19e02c 91 union {
fc766e4c
ED
92 struct dst_entry *next;
93 struct rtable __rcu *rt_next;
94 struct rt6_info *rt6_next;
95 struct dn_route __rcu *dn_next;
1e19e02c 96 };
1da177e4
LT
97};
98
27217455 99static inline struct neighbour *dst_get_neighbour_noref(struct dst_entry *dst)
69cce1d1 100{
f2c31e32
ED
101 return rcu_dereference(dst->_neighbour);
102}
103
27217455 104static inline struct neighbour *dst_get_neighbour_noref_raw(struct dst_entry *dst)
f2c31e32
ED
105{
106 return rcu_dereference_raw(dst->_neighbour);
69cce1d1
DM
107}
108
109static inline void dst_set_neighbour(struct dst_entry *dst, struct neighbour *neigh)
110{
f2c31e32 111 rcu_assign_pointer(dst->_neighbour, neigh);
69cce1d1
DM
112}
113
62fa8a84 114extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
725d1e1b 115extern const u32 dst_default_metrics[RTAX_MAX];
62fa8a84
DM
116
117#define DST_METRICS_READ_ONLY 0x1UL
118#define __DST_METRICS_PTR(Y) \
119 ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
120#define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics)
121
122static inline bool dst_metrics_read_only(const struct dst_entry *dst)
123{
124 return dst->_metrics & DST_METRICS_READ_ONLY;
125}
126
127extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
128
129static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
130{
131 unsigned long val = dst->_metrics;
132 if (!(val & DST_METRICS_READ_ONLY))
133 __dst_destroy_metrics_generic(dst, val);
134}
135
136static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
137{
138 unsigned long p = dst->_metrics;
139
1f37070d
SH
140 BUG_ON(!p);
141
62fa8a84
DM
142 if (p & DST_METRICS_READ_ONLY)
143 return dst->ops->cow_metrics(dst, p);
144 return __DST_METRICS_PTR(p);
145}
146
147/* This may only be invoked before the entry has reached global
148 * visibility.
149 */
150static inline void dst_init_metrics(struct dst_entry *dst,
151 const u32 *src_metrics,
152 bool read_only)
153{
154 dst->_metrics = ((unsigned long) src_metrics) |
155 (read_only ? DST_METRICS_READ_ONLY : 0);
156}
157
158static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
159{
160 u32 *dst_metrics = dst_metrics_write_ptr(dest);
161
162 if (dst_metrics) {
163 u32 *src_metrics = DST_METRICS_PTR(src);
164
165 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
166 }
167}
168
169static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
170{
171 return DST_METRICS_PTR(dst);
172}
173
1da177e4 174static inline u32
5170ae82 175dst_metric_raw(const struct dst_entry *dst, const int metric)
1da177e4 176{
62fa8a84
DM
177 u32 *p = DST_METRICS_PTR(dst);
178
179 return p[metric-1];
defb3519
DM
180}
181
5170ae82
DM
182static inline u32
183dst_metric(const struct dst_entry *dst, const int metric)
184{
0dbaee3b 185 WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
d33e4553
DM
186 metric == RTAX_ADVMSS ||
187 metric == RTAX_MTU);
5170ae82
DM
188 return dst_metric_raw(dst, metric);
189}
190
0dbaee3b
DM
191static inline u32
192dst_metric_advmss(const struct dst_entry *dst)
193{
194 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
195
196 if (!advmss)
197 advmss = dst->ops->default_advmss(dst);
198
199 return advmss;
200}
201
defb3519
DM
202static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
203{
62fa8a84 204 u32 *p = dst_metrics_write_ptr(dst);
defb3519 205
62fa8a84
DM
206 if (p)
207 p[metric-1] = val;
1da177e4
LT
208}
209
0c3adfb8
GBY
210static inline u32
211dst_feature(const struct dst_entry *dst, u32 feature)
212{
bb5b7c11 213 return dst_metric(dst, RTAX_FEATURES) & feature;
0c3adfb8
GBY
214}
215
1da177e4
LT
216static inline u32 dst_mtu(const struct dst_entry *dst)
217{
618f9bc7 218 return dst->ops->mtu(dst);
1da177e4
LT
219}
220
c1e20f7c
SH
221/* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
222static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
223{
224 return msecs_to_jiffies(dst_metric(dst, metric));
225}
226
227static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
228 unsigned long rtt)
229{
defb3519 230 dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
c1e20f7c
SH
231}
232
1da177e4
LT
233static inline u32
234dst_allfrag(const struct dst_entry *dst)
235{
0c3adfb8 236 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG);
1da177e4
LT
237 return ret;
238}
239
240static inline int
d33e4553 241dst_metric_locked(const struct dst_entry *dst, int metric)
1da177e4
LT
242{
243 return dst_metric(dst, RTAX_LOCK) & (1<<metric);
244}
245
7f95e188 246static inline void dst_hold(struct dst_entry *dst)
1da177e4 247{
5635c10d
ED
248 /*
249 * If your kernel compilation stops here, please check
250 * __pad_to_align_refcnt declaration in struct dst_entry
251 */
252 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
1da177e4
LT
253 atomic_inc(&dst->__refcnt);
254}
255
03f49f34
PE
256static inline void dst_use(struct dst_entry *dst, unsigned long time)
257{
258 dst_hold(dst);
259 dst->__use++;
260 dst->lastuse = time;
261}
262
7fee226a
ED
263static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
264{
265 dst->__use++;
266 dst->lastuse = time;
267}
268
7f95e188 269static inline struct dst_entry *dst_clone(struct dst_entry *dst)
1da177e4
LT
270{
271 if (dst)
272 atomic_inc(&dst->__refcnt);
273 return dst;
274}
275
8d330868 276extern void dst_release(struct dst_entry *dst);
7fee226a
ED
277
278static inline void refdst_drop(unsigned long refdst)
279{
280 if (!(refdst & SKB_DST_NOREF))
281 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
282}
283
284/**
285 * skb_dst_drop - drops skb dst
286 * @skb: buffer
287 *
288 * Drops dst reference count if a reference was taken.
289 */
adf30907
ED
290static inline void skb_dst_drop(struct sk_buff *skb)
291{
7fee226a
ED
292 if (skb->_skb_refdst) {
293 refdst_drop(skb->_skb_refdst);
294 skb->_skb_refdst = 0UL;
295 }
296}
297
298static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
299{
300 nskb->_skb_refdst = oskb->_skb_refdst;
301 if (!(nskb->_skb_refdst & SKB_DST_NOREF))
302 dst_clone(skb_dst(nskb));
303}
304
305/**
306 * skb_dst_force - makes sure skb dst is refcounted
307 * @skb: buffer
308 *
309 * If dst is not yet refcounted, let's do it
310 */
311static inline void skb_dst_force(struct sk_buff *skb)
312{
313 if (skb_dst_is_noref(skb)) {
314 WARN_ON(!rcu_read_lock_held());
315 skb->_skb_refdst &= ~SKB_DST_NOREF;
316 dst_clone(skb_dst(skb));
317 }
adf30907 318}
1da177e4 319
d19d56dd 320
290b895e
ED
321/**
322 * __skb_tunnel_rx - prepare skb for rx reinsert
323 * @skb: buffer
324 * @dev: tunnel device
325 *
326 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
327 * so make some cleanups. (no accounting done)
328 */
329static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
330{
331 skb->dev = dev;
bdeab991
TH
332
333 /*
334 * Clear rxhash so that we can recalulate the hash for the
335 * encapsulated packet, unless we have already determine the hash
336 * over the L4 4-tuple.
337 */
338 if (!skb->l4_rxhash)
339 skb->rxhash = 0;
290b895e
ED
340 skb_set_queue_mapping(skb, 0);
341 skb_dst_drop(skb);
342 nf_reset(skb);
343}
344
d19d56dd
ED
345/**
346 * skb_tunnel_rx - prepare skb for rx reinsert
347 * @skb: buffer
348 * @dev: tunnel device
349 *
350 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
351 * so make some cleanups, and perform accounting.
290b895e 352 * Note: this accounting is not SMP safe.
d19d56dd
ED
353 */
354static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
355{
d19d56dd
ED
356 /* TODO : stats should be SMP safe */
357 dev->stats.rx_packets++;
358 dev->stats.rx_bytes += skb->len;
290b895e 359 __skb_tunnel_rx(skb, dev);
d19d56dd
ED
360}
361
1da177e4
LT
362/* Children define the path of the packet through the
363 * Linux networking. Thus, destinations are stackable.
364 */
365
8764ab2c 366static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
1da177e4 367{
e433430a 368 struct dst_entry *child = dst_clone(skb_dst(skb)->child);
1da177e4 369
8764ab2c 370 skb_dst_drop(skb);
1da177e4
LT
371 return child;
372}
373
352e512c 374extern int dst_discard(struct sk_buff *skb);
7f95e188 375extern void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
5110effe
DM
376 int initial_ref, int initial_obsolete,
377 unsigned short flags);
7f95e188
EZ
378extern void __dst_free(struct dst_entry *dst);
379extern struct dst_entry *dst_destroy(struct dst_entry *dst);
1da177e4 380
7f95e188 381static inline void dst_free(struct dst_entry *dst)
1da177e4
LT
382{
383 if (dst->obsolete > 1)
384 return;
385 if (!atomic_read(&dst->__refcnt)) {
386 dst = dst_destroy(dst);
387 if (!dst)
388 return;
389 }
390 __dst_free(dst);
391}
392
393static inline void dst_rcu_free(struct rcu_head *head)
394{
395 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
396 dst_free(dst);
397}
398
399static inline void dst_confirm(struct dst_entry *dst)
400{
5110effe
DM
401 dst->pending_confirm = 1;
402}
f2c31e32 403
5110effe
DM
404static inline int dst_neigh_output(struct dst_entry *dst, struct neighbour *n,
405 struct sk_buff *skb)
406{
407 struct hh_cache *hh;
408
409 if (unlikely(dst->pending_confirm)) {
410 n->confirmed = jiffies;
411 dst->pending_confirm = 0;
69cce1d1 412 }
5110effe
DM
413
414 hh = &n->hh;
415 if ((n->nud_state & NUD_CONNECTED) && hh->hh_len)
416 return neigh_hh_output(hh, skb);
417 else
418 return n->output(n, skb);
1da177e4
LT
419}
420
d3aaeb38
DM
421static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
422{
423 return dst->ops->neigh_lookup(dst, daddr);
424}
425
1da177e4
LT
426static inline void dst_link_failure(struct sk_buff *skb)
427{
adf30907 428 struct dst_entry *dst = skb_dst(skb);
1da177e4
LT
429 if (dst && dst->ops && dst->ops->link_failure)
430 dst->ops->link_failure(skb);
431}
432
433static inline void dst_set_expires(struct dst_entry *dst, int timeout)
434{
435 unsigned long expires = jiffies + timeout;
436
437 if (expires == 0)
438 expires = 1;
439
440 if (dst->expires == 0 || time_before(expires, dst->expires))
441 dst->expires = expires;
442}
443
444/* Output packet to network from transport. */
445static inline int dst_output(struct sk_buff *skb)
446{
adf30907 447 return skb_dst(skb)->output(skb);
1da177e4
LT
448}
449
450/* Input packet from network to transport. */
451static inline int dst_input(struct sk_buff *skb)
452{
adf30907 453 return skb_dst(skb)->input(skb);
1da177e4
LT
454}
455
456static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
457{
458 if (dst->obsolete)
459 dst = dst->ops->check(dst, cookie);
460 return dst;
461}
462
463extern void dst_init(void);
464
815f4e57
HX
465/* Flags for xfrm_lookup flags argument. */
466enum {
80c0bc9e 467 XFRM_LOOKUP_ICMP = 1 << 0,
815f4e57
HX
468};
469
1da177e4
LT
470struct flowi;
471#ifndef CONFIG_XFRM
452edd59
DM
472static inline struct dst_entry *xfrm_lookup(struct net *net,
473 struct dst_entry *dst_orig,
474 const struct flowi *fl, struct sock *sk,
475 int flags)
1da177e4 476{
452edd59 477 return dst_orig;
1da177e4
LT
478}
479#else
452edd59
DM
480extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
481 const struct flowi *fl, struct sock *sk,
482 int flags);
1da177e4 483#endif
1da177e4
LT
484
485#endif /* _NET_DST_H */