blob: 4fb6c43817918992f8334c49022d5184ea45c8e4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Alexey Dobriyan86393e52009-08-29 01:34:49 +000011#include <net/dst_ops.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020012#include <linux/netdevice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/rtnetlink.h>
14#include <linux/rcupdate.h>
15#include <linux/jiffies.h>
16#include <net/neighbour.h>
17#include <asm/processor.h>
18
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#define DST_GC_MIN (HZ/10)
20#define DST_GC_INC (HZ/2)
21#define DST_GC_MAX (120*HZ)
22
23/* Each dst_entry has reference count and sits in some parent list(s).
24 * When it is removed from parent list, it is "freed" (dst_free).
25 * After this it enters dead state (dst->obsolete > 0) and if its refcnt
26 * is zero, it can be destroyed immediately, otherwise it is added
27 * to gc list and garbage collector periodically checks the refcnt.
28 */
29
30struct sk_buff;
31
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +000032struct dst_entry {
Eric Dumazet1e19e022007-02-09 16:26:55 -080033 struct rcu_head rcu_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 struct dst_entry *child;
35 struct net_device *dev;
David S. Miller62fa8a82011-01-26 20:51:05 -080036 struct dst_ops *ops;
37 unsigned long _metrics;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 unsigned long expires;
Zhang Yanminf1dd9c32008-03-12 22:52:37 -070039 struct dst_entry *path;
Eric Dumazetf2c31e32011-07-29 19:00:53 +000040 struct neighbour __rcu *_neighbour;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -070041#ifdef CONFIG_XFRM
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 struct xfrm_state *xfrm;
Eric Dumazet5635c102008-11-16 19:46:36 -080043#else
44 void *__pad1;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -070045#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 int (*input)(struct sk_buff*);
47 int (*output)(struct sk_buff*);
48
David S. Millerf6b72b62011-07-14 07:53:20 -070049 int flags;
50#define DST_HOST 0x0001
51#define DST_NOXFRM 0x0002
52#define DST_NOPOLICY 0x0004
53#define DST_NOHASH 0x0008
54#define DST_NOCACHE 0x0010
55#define DST_NOCOUNT 0x0020
56
David S. Miller62fa8a82011-01-26 20:51:05 -080057 short error;
58 short obsolete;
59 unsigned short header_len; /* more space at head required */
60 unsigned short trailer_len; /* space to reserve at tail */
Patrick McHardyc7066f72011-01-14 13:36:42 +010061#ifdef CONFIG_IP_ROUTE_CLASSID
Zhang Yanminf1dd9c32008-03-12 22:52:37 -070062 __u32 tclassid;
Eric Dumazet5635c102008-11-16 19:46:36 -080063#else
64 __u32 __pad2;
Zhang Yanminf1dd9c32008-03-12 22:52:37 -070065#endif
66
Eric Dumazet5635c102008-11-16 19:46:36 -080067 /*
68 * Align __refcnt to a 64 bytes alignment
69 * (L1_CACHE_SIZE would be too much)
70 */
71#ifdef CONFIG_64BIT
David S. Millerf6b72b62011-07-14 07:53:20 -070072 long __pad_to_align_refcnt[2];
Eric Dumazet5635c102008-11-16 19:46:36 -080073#endif
Zhang Yanminf1dd9c32008-03-12 22:52:37 -070074 /*
75 * __refcnt wants to be on a different cache line from
76 * input/output/ops or performance tanks badly
77 */
Eric Dumazet1e19e022007-02-09 16:26:55 -080078 atomic_t __refcnt; /* client references */
79 int __use;
Zhang Yanminf1dd9c32008-03-12 22:52:37 -070080 unsigned long lastuse;
Eric Dumazet1e19e022007-02-09 16:26:55 -080081 union {
Eric Dumazetfc766e42010-10-29 03:09:24 +000082 struct dst_entry *next;
83 struct rtable __rcu *rt_next;
84 struct rt6_info *rt6_next;
85 struct dn_route __rcu *dn_next;
Eric Dumazet1e19e022007-02-09 16:26:55 -080086 };
Linus Torvalds1da177e2005-04-16 15:20:36 -070087};
88
David S. Miller69cce1d2011-07-17 23:09:49 -070089static inline struct neighbour *dst_get_neighbour(struct dst_entry *dst)
90{
Eric Dumazetf2c31e32011-07-29 19:00:53 +000091 return rcu_dereference(dst->_neighbour);
92}
93
94static inline struct neighbour *dst_get_neighbour_raw(struct dst_entry *dst)
95{
96 return rcu_dereference_raw(dst->_neighbour);
David S. Miller69cce1d2011-07-17 23:09:49 -070097}
98
99static inline void dst_set_neighbour(struct dst_entry *dst, struct neighbour *neigh)
100{
Eric Dumazetf2c31e32011-07-29 19:00:53 +0000101 rcu_assign_pointer(dst->_neighbour, neigh);
David S. Miller69cce1d2011-07-17 23:09:49 -0700102}
103
David S. Miller62fa8a82011-01-26 20:51:05 -0800104extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
David S. Miller725d1e12011-01-28 14:05:05 -0800105extern const u32 dst_default_metrics[RTAX_MAX];
David S. Miller62fa8a82011-01-26 20:51:05 -0800106
107#define DST_METRICS_READ_ONLY 0x1UL
108#define __DST_METRICS_PTR(Y) \
109 ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
110#define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics)
111
112static inline bool dst_metrics_read_only(const struct dst_entry *dst)
113{
114 return dst->_metrics & DST_METRICS_READ_ONLY;
115}
116
117extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
118
119static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
120{
121 unsigned long val = dst->_metrics;
122 if (!(val & DST_METRICS_READ_ONLY))
123 __dst_destroy_metrics_generic(dst, val);
124}
125
126static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
127{
128 unsigned long p = dst->_metrics;
129
Stephen Hemminger1f370702011-05-24 13:50:52 -0400130 BUG_ON(!p);
131
David S. Miller62fa8a82011-01-26 20:51:05 -0800132 if (p & DST_METRICS_READ_ONLY)
133 return dst->ops->cow_metrics(dst, p);
134 return __DST_METRICS_PTR(p);
135}
136
137/* This may only be invoked before the entry has reached global
138 * visibility.
139 */
140static inline void dst_init_metrics(struct dst_entry *dst,
141 const u32 *src_metrics,
142 bool read_only)
143{
144 dst->_metrics = ((unsigned long) src_metrics) |
145 (read_only ? DST_METRICS_READ_ONLY : 0);
146}
147
148static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
149{
150 u32 *dst_metrics = dst_metrics_write_ptr(dest);
151
152 if (dst_metrics) {
153 u32 *src_metrics = DST_METRICS_PTR(src);
154
155 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
156 }
157}
158
159static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
160{
161 return DST_METRICS_PTR(dst);
162}
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164static inline u32
David S. Miller5170ae82010-12-12 21:35:57 -0800165dst_metric_raw(const struct dst_entry *dst, const int metric)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
David S. Miller62fa8a82011-01-26 20:51:05 -0800167 u32 *p = DST_METRICS_PTR(dst);
168
169 return p[metric-1];
David S. Millerdefb3512010-12-08 21:16:57 -0800170}
171
David S. Miller5170ae82010-12-12 21:35:57 -0800172static inline u32
173dst_metric(const struct dst_entry *dst, const int metric)
174{
David S. Miller0dbaee32010-12-13 12:52:14 -0800175 WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
David S. Millerd33e4552010-12-14 13:01:14 -0800176 metric == RTAX_ADVMSS ||
177 metric == RTAX_MTU);
David S. Miller5170ae82010-12-12 21:35:57 -0800178 return dst_metric_raw(dst, metric);
179}
180
David S. Miller0dbaee32010-12-13 12:52:14 -0800181static inline u32
182dst_metric_advmss(const struct dst_entry *dst)
183{
184 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
185
186 if (!advmss)
187 advmss = dst->ops->default_advmss(dst);
188
189 return advmss;
190}
191
David S. Millerdefb3512010-12-08 21:16:57 -0800192static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
193{
David S. Miller62fa8a82011-01-26 20:51:05 -0800194 u32 *p = dst_metrics_write_ptr(dst);
David S. Millerdefb3512010-12-08 21:16:57 -0800195
David S. Miller62fa8a82011-01-26 20:51:05 -0800196 if (p)
197 p[metric-1] = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198}
199
Gilad Ben-Yossef0c3adfb2009-10-28 04:15:23 +0000200static inline u32
201dst_feature(const struct dst_entry *dst, u32 feature)
202{
David S. Millerbb5b7c12009-12-15 20:56:42 -0800203 return dst_metric(dst, RTAX_FEATURES) & feature;
Gilad Ben-Yossef0c3adfb2009-10-28 04:15:23 +0000204}
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206static inline u32 dst_mtu(const struct dst_entry *dst)
207{
David S. Millerd33e4552010-12-14 13:01:14 -0800208 u32 mtu = dst_metric_raw(dst, RTAX_MTU);
209
210 if (!mtu)
211 mtu = dst->ops->default_mtu(dst);
212
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 return mtu;
214}
215
Stephen Hemmingerc1e20f72008-07-18 23:02:15 -0700216/* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
217static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
218{
219 return msecs_to_jiffies(dst_metric(dst, metric));
220}
221
222static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
223 unsigned long rtt)
224{
David S. Millerdefb3512010-12-08 21:16:57 -0800225 dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
Stephen Hemmingerc1e20f72008-07-18 23:02:15 -0700226}
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228static inline u32
229dst_allfrag(const struct dst_entry *dst)
230{
Gilad Ben-Yossef0c3adfb2009-10-28 04:15:23 +0000231 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 return ret;
233}
234
235static inline int
David S. Millerd33e4552010-12-14 13:01:14 -0800236dst_metric_locked(const struct dst_entry *dst, int metric)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
238 return dst_metric(dst, RTAX_LOCK) & (1<<metric);
239}
240
241static inline void dst_hold(struct dst_entry * dst)
242{
Eric Dumazet5635c102008-11-16 19:46:36 -0800243 /*
244 * If your kernel compilation stops here, please check
245 * __pad_to_align_refcnt declaration in struct dst_entry
246 */
247 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 atomic_inc(&dst->__refcnt);
249}
250
Pavel Emelyanov03f49f32007-11-10 21:28:34 -0800251static inline void dst_use(struct dst_entry *dst, unsigned long time)
252{
253 dst_hold(dst);
254 dst->__use++;
255 dst->lastuse = time;
256}
257
Eric Dumazet7fee2262010-05-11 23:19:48 +0000258static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
259{
260 dst->__use++;
261 dst->lastuse = time;
262}
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264static inline
265struct dst_entry * dst_clone(struct dst_entry * dst)
266{
267 if (dst)
268 atomic_inc(&dst->__refcnt);
269 return dst;
270}
271
Ilpo Järvinen8d330862008-03-27 17:53:31 -0700272extern void dst_release(struct dst_entry *dst);
Eric Dumazet7fee2262010-05-11 23:19:48 +0000273
274static inline void refdst_drop(unsigned long refdst)
275{
276 if (!(refdst & SKB_DST_NOREF))
277 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
278}
279
280/**
281 * skb_dst_drop - drops skb dst
282 * @skb: buffer
283 *
284 * Drops dst reference count if a reference was taken.
285 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000286static inline void skb_dst_drop(struct sk_buff *skb)
287{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000288 if (skb->_skb_refdst) {
289 refdst_drop(skb->_skb_refdst);
290 skb->_skb_refdst = 0UL;
291 }
292}
293
294static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
295{
296 nskb->_skb_refdst = oskb->_skb_refdst;
297 if (!(nskb->_skb_refdst & SKB_DST_NOREF))
298 dst_clone(skb_dst(nskb));
299}
300
301/**
302 * skb_dst_force - makes sure skb dst is refcounted
303 * @skb: buffer
304 *
305 * If dst is not yet refcounted, let's do it
306 */
307static inline void skb_dst_force(struct sk_buff *skb)
308{
309 if (skb_dst_is_noref(skb)) {
310 WARN_ON(!rcu_read_lock_held());
311 skb->_skb_refdst &= ~SKB_DST_NOREF;
312 dst_clone(skb_dst(skb));
313 }
Eric Dumazetadf30902009-06-02 05:19:30 +0000314}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
Eric Dumazetd19d56d2010-05-17 22:36:55 -0700316
317/**
Eric Dumazet290b8952010-09-27 00:33:35 +0000318 * __skb_tunnel_rx - prepare skb for rx reinsert
319 * @skb: buffer
320 * @dev: tunnel device
321 *
322 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
323 * so make some cleanups. (no accounting done)
324 */
325static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
326{
327 skb->dev = dev;
Tom Herbertbdeab992011-08-14 19:45:55 +0000328
329 /*
330 * Clear rxhash so that we can recalulate the hash for the
331 * encapsulated packet, unless we have already determine the hash
332 * over the L4 4-tuple.
333 */
334 if (!skb->l4_rxhash)
335 skb->rxhash = 0;
Eric Dumazet290b8952010-09-27 00:33:35 +0000336 skb_set_queue_mapping(skb, 0);
337 skb_dst_drop(skb);
338 nf_reset(skb);
339}
340
341/**
Eric Dumazetd19d56d2010-05-17 22:36:55 -0700342 * skb_tunnel_rx - prepare skb for rx reinsert
343 * @skb: buffer
344 * @dev: tunnel device
345 *
346 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
347 * so make some cleanups, and perform accounting.
Eric Dumazet290b8952010-09-27 00:33:35 +0000348 * Note: this accounting is not SMP safe.
Eric Dumazetd19d56d2010-05-17 22:36:55 -0700349 */
350static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
351{
Eric Dumazetd19d56d2010-05-17 22:36:55 -0700352 /* TODO : stats should be SMP safe */
353 dev->stats.rx_packets++;
354 dev->stats.rx_bytes += skb->len;
Eric Dumazet290b8952010-09-27 00:33:35 +0000355 __skb_tunnel_rx(skb, dev);
Eric Dumazetd19d56d2010-05-17 22:36:55 -0700356}
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358/* Children define the path of the packet through the
359 * Linux networking. Thus, destinations are stackable.
360 */
361
Steffen Klassert8764ab22010-06-04 01:57:38 +0000362static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Steffen Klasserte4334302011-03-15 21:09:32 +0000364 struct dst_entry *child = dst_clone(skb_dst(skb)->child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Steffen Klassert8764ab22010-06-04 01:57:38 +0000366 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 return child;
368}
369
Herbert Xu352e5122007-11-13 21:34:06 -0800370extern int dst_discard(struct sk_buff *skb);
David S. Miller5c1e6aa2011-04-28 14:13:38 -0700371extern void *dst_alloc(struct dst_ops * ops, struct net_device *dev,
372 int initial_ref, int initial_obsolete, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373extern void __dst_free(struct dst_entry * dst);
374extern struct dst_entry *dst_destroy(struct dst_entry * dst);
375
376static inline void dst_free(struct dst_entry * dst)
377{
378 if (dst->obsolete > 1)
379 return;
380 if (!atomic_read(&dst->__refcnt)) {
381 dst = dst_destroy(dst);
382 if (!dst)
383 return;
384 }
385 __dst_free(dst);
386}
387
388static inline void dst_rcu_free(struct rcu_head *head)
389{
390 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
391 dst_free(dst);
392}
393
394static inline void dst_confirm(struct dst_entry *dst)
395{
David S. Miller69cce1d2011-07-17 23:09:49 -0700396 if (dst) {
Eric Dumazetf2c31e32011-07-29 19:00:53 +0000397 struct neighbour *n;
398
399 rcu_read_lock();
400 n = dst_get_neighbour(dst);
David S. Miller69cce1d2011-07-17 23:09:49 -0700401 neigh_confirm(n);
Eric Dumazetf2c31e32011-07-29 19:00:53 +0000402 rcu_read_unlock();
David S. Miller69cce1d2011-07-17 23:09:49 -0700403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
405
David S. Millerd3aaeb32011-07-18 00:40:17 -0700406static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
407{
408 return dst->ops->neigh_lookup(dst, daddr);
409}
410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411static inline void dst_link_failure(struct sk_buff *skb)
412{
Eric Dumazetadf30902009-06-02 05:19:30 +0000413 struct dst_entry *dst = skb_dst(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (dst && dst->ops && dst->ops->link_failure)
415 dst->ops->link_failure(skb);
416}
417
418static inline void dst_set_expires(struct dst_entry *dst, int timeout)
419{
420 unsigned long expires = jiffies + timeout;
421
422 if (expires == 0)
423 expires = 1;
424
425 if (dst->expires == 0 || time_before(expires, dst->expires))
426 dst->expires = expires;
427}
428
429/* Output packet to network from transport. */
430static inline int dst_output(struct sk_buff *skb)
431{
Eric Dumazetadf30902009-06-02 05:19:30 +0000432 return skb_dst(skb)->output(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
435/* Input packet from network to transport. */
436static inline int dst_input(struct sk_buff *skb)
437{
Eric Dumazetadf30902009-06-02 05:19:30 +0000438 return skb_dst(skb)->input(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
440
441static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
442{
443 if (dst->obsolete)
444 dst = dst->ops->check(dst, cookie);
445 return dst;
446}
447
448extern void dst_init(void);
449
Herbert Xu815f4e52007-12-12 10:36:59 -0800450/* Flags for xfrm_lookup flags argument. */
451enum {
David S. Miller80c0bc92011-03-01 14:36:37 -0800452 XFRM_LOOKUP_ICMP = 1 << 0,
Herbert Xu815f4e52007-12-12 10:36:59 -0800453};
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455struct flowi;
456#ifndef CONFIG_XFRM
David S. Miller452edd52011-03-02 13:27:41 -0800457static inline struct dst_entry *xfrm_lookup(struct net *net,
458 struct dst_entry *dst_orig,
459 const struct flowi *fl, struct sock *sk,
460 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
David S. Miller452edd52011-03-02 13:27:41 -0800462 return dst_orig;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464#else
David S. Miller452edd52011-03-02 13:27:41 -0800465extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
466 const struct flowi *fl, struct sock *sk,
467 int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470#endif /* _NET_DST_H */