2 * net/dst.h Protocol independent destination cache definitions.
4 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
11 #include <net/dst_ops.h>
12 #include <linux/netdevice.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/rcupdate.h>
15 #include <linux/bug.h>
16 #include <linux/jiffies.h>
17 #include <net/neighbour.h>
18 #include <asm/processor.h>
20 #define DST_GC_MIN (HZ/10)
21 #define DST_GC_INC (HZ/2)
22 #define DST_GC_MAX (120*HZ)
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.
34 struct rcu_head rcu_head;
35 struct dst_entry *child;
36 struct net_device *dev;
38 unsigned long _metrics;
40 unsigned long expires;
41 /* point to where the dst_entry copied from */
42 struct dst_entry *from;
44 struct dst_entry *path;
45 struct neighbour __rcu *_neighbour;
47 struct xfrm_state *xfrm;
51 int (*input)(struct sk_buff*);
52 int (*output)(struct sk_buff*);
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
61 #define DST_NOPEER 0x0040
62 #define DST_FAKE_RTABLE 0x0080
63 #define DST_XFRM_TUNNEL 0x0100
67 unsigned short header_len; /* more space at head required */
68 unsigned short trailer_len; /* space to reserve at tail */
69 #ifdef CONFIG_IP_ROUTE_CLASSID
76 * Align __refcnt to a 64 bytes alignment
77 * (L1_CACHE_SIZE would be too much)
80 long __pad_to_align_refcnt[2];
83 * __refcnt wants to be on a different cache line from
84 * input/output/ops or performance tanks badly
86 atomic_t __refcnt; /* client references */
88 unsigned long lastuse;
90 struct dst_entry *next;
91 struct rtable __rcu *rt_next;
92 struct rt6_info *rt6_next;
93 struct dn_route __rcu *dn_next;
97 static inline struct neighbour *dst_get_neighbour_noref(struct dst_entry *dst)
99 return rcu_dereference(dst->_neighbour);
102 static inline struct neighbour *dst_get_neighbour_noref_raw(struct dst_entry *dst)
104 return rcu_dereference_raw(dst->_neighbour);
107 static inline void dst_set_neighbour(struct dst_entry *dst, struct neighbour *neigh)
109 rcu_assign_pointer(dst->_neighbour, neigh);
112 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
113 extern const u32 dst_default_metrics[RTAX_MAX];
115 #define DST_METRICS_READ_ONLY 0x1UL
116 #define __DST_METRICS_PTR(Y) \
117 ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
118 #define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics)
120 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
122 return dst->_metrics & DST_METRICS_READ_ONLY;
125 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
127 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
129 unsigned long val = dst->_metrics;
130 if (!(val & DST_METRICS_READ_ONLY))
131 __dst_destroy_metrics_generic(dst, val);
134 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
136 unsigned long p = dst->_metrics;
140 if (p & DST_METRICS_READ_ONLY)
141 return dst->ops->cow_metrics(dst, p);
142 return __DST_METRICS_PTR(p);
145 /* This may only be invoked before the entry has reached global
148 static inline void dst_init_metrics(struct dst_entry *dst,
149 const u32 *src_metrics,
152 dst->_metrics = ((unsigned long) src_metrics) |
153 (read_only ? DST_METRICS_READ_ONLY : 0);
156 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
158 u32 *dst_metrics = dst_metrics_write_ptr(dest);
161 u32 *src_metrics = DST_METRICS_PTR(src);
163 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
167 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
169 return DST_METRICS_PTR(dst);
173 dst_metric_raw(const struct dst_entry *dst, const int metric)
175 u32 *p = DST_METRICS_PTR(dst);
181 dst_metric(const struct dst_entry *dst, const int metric)
183 WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
184 metric == RTAX_ADVMSS ||
186 return dst_metric_raw(dst, metric);
190 dst_metric_advmss(const struct dst_entry *dst)
192 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
195 advmss = dst->ops->default_advmss(dst);
200 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
202 u32 *p = dst_metrics_write_ptr(dst);
209 dst_feature(const struct dst_entry *dst, u32 feature)
211 return dst_metric(dst, RTAX_FEATURES) & feature;
214 static inline u32 dst_mtu(const struct dst_entry *dst)
216 return dst->ops->mtu(dst);
219 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
220 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
222 return msecs_to_jiffies(dst_metric(dst, metric));
225 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
228 dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
232 dst_allfrag(const struct dst_entry *dst)
234 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG);
239 dst_metric_locked(const struct dst_entry *dst, int metric)
241 return dst_metric(dst, RTAX_LOCK) & (1<<metric);
244 static inline void dst_hold(struct dst_entry * dst)
247 * If your kernel compilation stops here, please check
248 * __pad_to_align_refcnt declaration in struct dst_entry
250 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
251 atomic_inc(&dst->__refcnt);
254 static inline void dst_use(struct dst_entry *dst, unsigned long time)
261 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
268 struct dst_entry * dst_clone(struct dst_entry * dst)
271 atomic_inc(&dst->__refcnt);
275 extern void dst_release(struct dst_entry *dst);
277 static inline void refdst_drop(unsigned long refdst)
279 if (!(refdst & SKB_DST_NOREF))
280 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
284 * skb_dst_drop - drops skb dst
287 * Drops dst reference count if a reference was taken.
289 static inline void skb_dst_drop(struct sk_buff *skb)
291 if (skb->_skb_refdst) {
292 refdst_drop(skb->_skb_refdst);
293 skb->_skb_refdst = 0UL;
297 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
299 nskb->_skb_refdst = oskb->_skb_refdst;
300 if (!(nskb->_skb_refdst & SKB_DST_NOREF))
301 dst_clone(skb_dst(nskb));
305 * skb_dst_force - makes sure skb dst is refcounted
308 * If dst is not yet refcounted, let's do it
310 static inline void skb_dst_force(struct sk_buff *skb)
312 if (skb_dst_is_noref(skb)) {
313 WARN_ON(!rcu_read_lock_held());
314 skb->_skb_refdst &= ~SKB_DST_NOREF;
315 dst_clone(skb_dst(skb));
321 * __skb_tunnel_rx - prepare skb for rx reinsert
323 * @dev: tunnel device
325 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
326 * so make some cleanups. (no accounting done)
328 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
333 * Clear rxhash so that we can recalulate the hash for the
334 * encapsulated packet, unless we have already determine the hash
335 * over the L4 4-tuple.
339 skb_set_queue_mapping(skb, 0);
345 * skb_tunnel_rx - prepare skb for rx reinsert
347 * @dev: tunnel device
349 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
350 * so make some cleanups, and perform accounting.
351 * Note: this accounting is not SMP safe.
353 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
355 /* TODO : stats should be SMP safe */
356 dev->stats.rx_packets++;
357 dev->stats.rx_bytes += skb->len;
358 __skb_tunnel_rx(skb, dev);
361 /* Children define the path of the packet through the
362 * Linux networking. Thus, destinations are stackable.
365 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
367 struct dst_entry *child = dst_clone(skb_dst(skb)->child);
373 extern int dst_discard(struct sk_buff *skb);
374 extern void *dst_alloc(struct dst_ops * ops, struct net_device *dev,
375 int initial_ref, int initial_obsolete, int flags);
376 extern void __dst_free(struct dst_entry * dst);
377 extern struct dst_entry *dst_destroy(struct dst_entry * dst);
379 static inline void dst_free(struct dst_entry * dst)
381 if (dst->obsolete > 1)
383 if (!atomic_read(&dst->__refcnt)) {
384 dst = dst_destroy(dst);
391 static inline void dst_rcu_free(struct rcu_head *head)
393 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
397 static inline void dst_confirm(struct dst_entry *dst)
403 n = dst_get_neighbour_noref(dst);
409 static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
411 return dst->ops->neigh_lookup(dst, daddr);
414 static inline void dst_link_failure(struct sk_buff *skb)
416 struct dst_entry *dst = skb_dst(skb);
417 if (dst && dst->ops && dst->ops->link_failure)
418 dst->ops->link_failure(skb);
421 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
423 unsigned long expires = jiffies + timeout;
428 if (dst->expires == 0 || time_before(expires, dst->expires))
429 dst->expires = expires;
432 /* Output packet to network from transport. */
433 static inline int dst_output(struct sk_buff *skb)
435 return skb_dst(skb)->output(skb);
438 /* Input packet from network to transport. */
439 static inline int dst_input(struct sk_buff *skb)
441 return skb_dst(skb)->input(skb);
444 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
447 dst = dst->ops->check(dst, cookie);
451 extern void dst_init(void);
453 /* Flags for xfrm_lookup flags argument. */
455 XFRM_LOOKUP_ICMP = 1 << 0,
460 static inline struct dst_entry *xfrm_lookup(struct net *net,
461 struct dst_entry *dst_orig,
462 const struct flowi *fl, struct sock *sk,
468 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
469 const struct flowi *fl, struct sock *sk,
473 #endif /* _NET_DST_H */