Commit | Line | Data |
---|---|---|
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 | ||
31 | struct sk_buff; | |
32 | ||
fd2c3ef7 | 33 | struct 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; |
36bdbcae | 45 | void *__pad0; |
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 | ||
62fa8a84 | 99 | extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old); |
725d1e1b | 100 | extern const u32 dst_default_metrics[RTAX_MAX]; |
62fa8a84 DM |
101 | |
102 | #define DST_METRICS_READ_ONLY 0x1UL | |
103 | #define __DST_METRICS_PTR(Y) \ | |
104 | ((u32 *)((Y) & ~DST_METRICS_READ_ONLY)) | |
105 | #define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics) | |
106 | ||
107 | static inline bool dst_metrics_read_only(const struct dst_entry *dst) | |
108 | { | |
109 | return dst->_metrics & DST_METRICS_READ_ONLY; | |
110 | } | |
111 | ||
112 | extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old); | |
113 | ||
114 | static inline void dst_destroy_metrics_generic(struct dst_entry *dst) | |
115 | { | |
116 | unsigned long val = dst->_metrics; | |
117 | if (!(val & DST_METRICS_READ_ONLY)) | |
118 | __dst_destroy_metrics_generic(dst, val); | |
119 | } | |
120 | ||
121 | static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst) | |
122 | { | |
123 | unsigned long p = dst->_metrics; | |
124 | ||
1f37070d SH |
125 | BUG_ON(!p); |
126 | ||
62fa8a84 DM |
127 | if (p & DST_METRICS_READ_ONLY) |
128 | return dst->ops->cow_metrics(dst, p); | |
129 | return __DST_METRICS_PTR(p); | |
130 | } | |
131 | ||
132 | /* This may only be invoked before the entry has reached global | |
133 | * visibility. | |
134 | */ | |
135 | static inline void dst_init_metrics(struct dst_entry *dst, | |
136 | const u32 *src_metrics, | |
137 | bool read_only) | |
138 | { | |
139 | dst->_metrics = ((unsigned long) src_metrics) | | |
140 | (read_only ? DST_METRICS_READ_ONLY : 0); | |
141 | } | |
142 | ||
143 | static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src) | |
144 | { | |
145 | u32 *dst_metrics = dst_metrics_write_ptr(dest); | |
146 | ||
147 | if (dst_metrics) { | |
148 | u32 *src_metrics = DST_METRICS_PTR(src); | |
149 | ||
150 | memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32)); | |
151 | } | |
152 | } | |
153 | ||
154 | static inline u32 *dst_metrics_ptr(struct dst_entry *dst) | |
155 | { | |
156 | return DST_METRICS_PTR(dst); | |
157 | } | |
158 | ||
1da177e4 | 159 | static inline u32 |
5170ae82 | 160 | dst_metric_raw(const struct dst_entry *dst, const int metric) |
1da177e4 | 161 | { |
62fa8a84 DM |
162 | u32 *p = DST_METRICS_PTR(dst); |
163 | ||
164 | return p[metric-1]; | |
defb3519 DM |
165 | } |
166 | ||
5170ae82 DM |
167 | static inline u32 |
168 | dst_metric(const struct dst_entry *dst, const int metric) | |
169 | { | |
0dbaee3b | 170 | WARN_ON_ONCE(metric == RTAX_HOPLIMIT || |
d33e4553 DM |
171 | metric == RTAX_ADVMSS || |
172 | metric == RTAX_MTU); | |
5170ae82 DM |
173 | return dst_metric_raw(dst, metric); |
174 | } | |
175 | ||
0dbaee3b DM |
176 | static inline u32 |
177 | dst_metric_advmss(const struct dst_entry *dst) | |
178 | { | |
179 | u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS); | |
180 | ||
181 | if (!advmss) | |
182 | advmss = dst->ops->default_advmss(dst); | |
183 | ||
184 | return advmss; | |
185 | } | |
186 | ||
defb3519 DM |
187 | static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val) |
188 | { | |
62fa8a84 | 189 | u32 *p = dst_metrics_write_ptr(dst); |
defb3519 | 190 | |
62fa8a84 DM |
191 | if (p) |
192 | p[metric-1] = val; | |
1da177e4 LT |
193 | } |
194 | ||
0c3adfb8 GBY |
195 | static inline u32 |
196 | dst_feature(const struct dst_entry *dst, u32 feature) | |
197 | { | |
bb5b7c11 | 198 | return dst_metric(dst, RTAX_FEATURES) & feature; |
0c3adfb8 GBY |
199 | } |
200 | ||
1da177e4 LT |
201 | static inline u32 dst_mtu(const struct dst_entry *dst) |
202 | { | |
618f9bc7 | 203 | return dst->ops->mtu(dst); |
1da177e4 LT |
204 | } |
205 | ||
c1e20f7c SH |
206 | /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */ |
207 | static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric) | |
208 | { | |
209 | return msecs_to_jiffies(dst_metric(dst, metric)); | |
210 | } | |
211 | ||
1da177e4 LT |
212 | static inline u32 |
213 | dst_allfrag(const struct dst_entry *dst) | |
214 | { | |
0c3adfb8 | 215 | int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG); |
1da177e4 LT |
216 | return ret; |
217 | } | |
218 | ||
219 | static inline int | |
d33e4553 | 220 | dst_metric_locked(const struct dst_entry *dst, int metric) |
1da177e4 LT |
221 | { |
222 | return dst_metric(dst, RTAX_LOCK) & (1<<metric); | |
223 | } | |
224 | ||
7f95e188 | 225 | static inline void dst_hold(struct dst_entry *dst) |
1da177e4 | 226 | { |
5635c10d ED |
227 | /* |
228 | * If your kernel compilation stops here, please check | |
229 | * __pad_to_align_refcnt declaration in struct dst_entry | |
230 | */ | |
231 | BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63); | |
1da177e4 LT |
232 | atomic_inc(&dst->__refcnt); |
233 | } | |
234 | ||
03f49f34 PE |
235 | static inline void dst_use(struct dst_entry *dst, unsigned long time) |
236 | { | |
237 | dst_hold(dst); | |
238 | dst->__use++; | |
239 | dst->lastuse = time; | |
240 | } | |
241 | ||
7fee226a ED |
242 | static inline void dst_use_noref(struct dst_entry *dst, unsigned long time) |
243 | { | |
244 | dst->__use++; | |
245 | dst->lastuse = time; | |
246 | } | |
247 | ||
7f95e188 | 248 | static inline struct dst_entry *dst_clone(struct dst_entry *dst) |
1da177e4 LT |
249 | { |
250 | if (dst) | |
251 | atomic_inc(&dst->__refcnt); | |
252 | return dst; | |
253 | } | |
254 | ||
8d330868 | 255 | extern void dst_release(struct dst_entry *dst); |
7fee226a ED |
256 | |
257 | static inline void refdst_drop(unsigned long refdst) | |
258 | { | |
259 | if (!(refdst & SKB_DST_NOREF)) | |
260 | dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK)); | |
261 | } | |
262 | ||
263 | /** | |
264 | * skb_dst_drop - drops skb dst | |
265 | * @skb: buffer | |
266 | * | |
267 | * Drops dst reference count if a reference was taken. | |
268 | */ | |
adf30907 ED |
269 | static inline void skb_dst_drop(struct sk_buff *skb) |
270 | { | |
7fee226a ED |
271 | if (skb->_skb_refdst) { |
272 | refdst_drop(skb->_skb_refdst); | |
273 | skb->_skb_refdst = 0UL; | |
274 | } | |
275 | } | |
276 | ||
277 | static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb) | |
278 | { | |
279 | nskb->_skb_refdst = oskb->_skb_refdst; | |
280 | if (!(nskb->_skb_refdst & SKB_DST_NOREF)) | |
281 | dst_clone(skb_dst(nskb)); | |
282 | } | |
283 | ||
284 | /** | |
285 | * skb_dst_force - makes sure skb dst is refcounted | |
286 | * @skb: buffer | |
287 | * | |
288 | * If dst is not yet refcounted, let's do it | |
289 | */ | |
290 | static inline void skb_dst_force(struct sk_buff *skb) | |
291 | { | |
292 | if (skb_dst_is_noref(skb)) { | |
293 | WARN_ON(!rcu_read_lock_held()); | |
294 | skb->_skb_refdst &= ~SKB_DST_NOREF; | |
295 | dst_clone(skb_dst(skb)); | |
296 | } | |
adf30907 | 297 | } |
1da177e4 | 298 | |
d19d56dd | 299 | |
290b895e ED |
300 | /** |
301 | * __skb_tunnel_rx - prepare skb for rx reinsert | |
302 | * @skb: buffer | |
303 | * @dev: tunnel device | |
304 | * | |
305 | * After decapsulation, packet is going to re-enter (netif_rx()) our stack, | |
306 | * so make some cleanups. (no accounting done) | |
307 | */ | |
308 | static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev) | |
309 | { | |
310 | skb->dev = dev; | |
bdeab991 TH |
311 | |
312 | /* | |
313 | * Clear rxhash so that we can recalulate the hash for the | |
314 | * encapsulated packet, unless we have already determine the hash | |
315 | * over the L4 4-tuple. | |
316 | */ | |
317 | if (!skb->l4_rxhash) | |
318 | skb->rxhash = 0; | |
290b895e ED |
319 | skb_set_queue_mapping(skb, 0); |
320 | skb_dst_drop(skb); | |
321 | nf_reset(skb); | |
322 | } | |
323 | ||
d19d56dd ED |
324 | /** |
325 | * skb_tunnel_rx - prepare skb for rx reinsert | |
326 | * @skb: buffer | |
327 | * @dev: tunnel device | |
328 | * | |
329 | * After decapsulation, packet is going to re-enter (netif_rx()) our stack, | |
330 | * so make some cleanups, and perform accounting. | |
290b895e | 331 | * Note: this accounting is not SMP safe. |
d19d56dd ED |
332 | */ |
333 | static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev) | |
334 | { | |
d19d56dd ED |
335 | /* TODO : stats should be SMP safe */ |
336 | dev->stats.rx_packets++; | |
337 | dev->stats.rx_bytes += skb->len; | |
290b895e | 338 | __skb_tunnel_rx(skb, dev); |
d19d56dd ED |
339 | } |
340 | ||
1da177e4 LT |
341 | /* Children define the path of the packet through the |
342 | * Linux networking. Thus, destinations are stackable. | |
343 | */ | |
344 | ||
8764ab2c | 345 | static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb) |
1da177e4 | 346 | { |
e433430a | 347 | struct dst_entry *child = dst_clone(skb_dst(skb)->child); |
1da177e4 | 348 | |
8764ab2c | 349 | skb_dst_drop(skb); |
1da177e4 LT |
350 | return child; |
351 | } | |
352 | ||
352e512c | 353 | extern int dst_discard(struct sk_buff *skb); |
7f95e188 | 354 | extern void *dst_alloc(struct dst_ops *ops, struct net_device *dev, |
5110effe DM |
355 | int initial_ref, int initial_obsolete, |
356 | unsigned short flags); | |
7f95e188 EZ |
357 | extern void __dst_free(struct dst_entry *dst); |
358 | extern struct dst_entry *dst_destroy(struct dst_entry *dst); | |
1da177e4 | 359 | |
7f95e188 | 360 | static inline void dst_free(struct dst_entry *dst) |
1da177e4 LT |
361 | { |
362 | if (dst->obsolete > 1) | |
363 | return; | |
364 | if (!atomic_read(&dst->__refcnt)) { | |
365 | dst = dst_destroy(dst); | |
366 | if (!dst) | |
367 | return; | |
368 | } | |
369 | __dst_free(dst); | |
370 | } | |
371 | ||
372 | static inline void dst_rcu_free(struct rcu_head *head) | |
373 | { | |
374 | struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head); | |
375 | dst_free(dst); | |
376 | } | |
377 | ||
378 | static inline void dst_confirm(struct dst_entry *dst) | |
379 | { | |
5110effe DM |
380 | dst->pending_confirm = 1; |
381 | } | |
f2c31e32 | 382 | |
5110effe DM |
383 | static inline int dst_neigh_output(struct dst_entry *dst, struct neighbour *n, |
384 | struct sk_buff *skb) | |
385 | { | |
386 | struct hh_cache *hh; | |
387 | ||
388 | if (unlikely(dst->pending_confirm)) { | |
389 | n->confirmed = jiffies; | |
390 | dst->pending_confirm = 0; | |
69cce1d1 | 391 | } |
5110effe DM |
392 | |
393 | hh = &n->hh; | |
394 | if ((n->nud_state & NUD_CONNECTED) && hh->hh_len) | |
395 | return neigh_hh_output(hh, skb); | |
396 | else | |
397 | return n->output(n, skb); | |
1da177e4 LT |
398 | } |
399 | ||
d3aaeb38 DM |
400 | static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr) |
401 | { | |
f894cbf8 DM |
402 | return dst->ops->neigh_lookup(dst, NULL, daddr); |
403 | } | |
404 | ||
405 | static inline struct neighbour *dst_neigh_lookup_skb(const struct dst_entry *dst, | |
406 | struct sk_buff *skb) | |
407 | { | |
408 | return dst->ops->neigh_lookup(dst, skb, NULL); | |
d3aaeb38 DM |
409 | } |
410 | ||
1da177e4 LT |
411 | static inline void dst_link_failure(struct sk_buff *skb) |
412 | { | |
adf30907 | 413 | struct dst_entry *dst = skb_dst(skb); |
1da177e4 LT |
414 | if (dst && dst->ops && dst->ops->link_failure) |
415 | dst->ops->link_failure(skb); | |
416 | } | |
417 | ||
418 | static 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. */ | |
430 | static inline int dst_output(struct sk_buff *skb) | |
431 | { | |
adf30907 | 432 | return skb_dst(skb)->output(skb); |
1da177e4 LT |
433 | } |
434 | ||
435 | /* Input packet from network to transport. */ | |
436 | static inline int dst_input(struct sk_buff *skb) | |
437 | { | |
adf30907 | 438 | return skb_dst(skb)->input(skb); |
1da177e4 LT |
439 | } |
440 | ||
441 | static 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 | ||
448 | extern void dst_init(void); | |
449 | ||
815f4e57 HX |
450 | /* Flags for xfrm_lookup flags argument. */ |
451 | enum { | |
80c0bc9e | 452 | XFRM_LOOKUP_ICMP = 1 << 0, |
815f4e57 HX |
453 | }; |
454 | ||
1da177e4 LT |
455 | struct flowi; |
456 | #ifndef CONFIG_XFRM | |
452edd59 DM |
457 | static 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) | |
1da177e4 | 461 | { |
452edd59 | 462 | return dst_orig; |
1da177e4 LT |
463 | } |
464 | #else | |
452edd59 DM |
465 | extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, |
466 | const struct flowi *fl, struct sock *sk, | |
467 | int flags); | |
1da177e4 | 468 | #endif |
1da177e4 LT |
469 | |
470 | #endif /* _NET_DST_H */ |