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;
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
65 unsigned short pending_confirm;
69 unsigned short header_len; /* more space at head required */
70 unsigned short trailer_len; /* space to reserve at tail */
71 #ifdef CONFIG_IP_ROUTE_CLASSID
78 * Align __refcnt to a 64 bytes alignment
79 * (L1_CACHE_SIZE would be too much)
82 long __pad_to_align_refcnt[2];
85 * __refcnt wants to be on a different cache line from
86 * input/output/ops or performance tanks badly
88 atomic_t __refcnt; /* client references */
90 unsigned long lastuse;
92 struct dst_entry *next;
93 struct rtable __rcu *rt_next;
94 struct rt6_info *rt6_next;
95 struct dn_route __rcu *dn_next;
99 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
100 extern const u32 dst_default_metrics[RTAX_MAX];
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)
107 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
109 return dst->_metrics & DST_METRICS_READ_ONLY;
112 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
114 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
116 unsigned long val = dst->_metrics;
117 if (!(val & DST_METRICS_READ_ONLY))
118 __dst_destroy_metrics_generic(dst, val);
121 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
123 unsigned long p = dst->_metrics;
127 if (p & DST_METRICS_READ_ONLY)
128 return dst->ops->cow_metrics(dst, p);
129 return __DST_METRICS_PTR(p);
132 /* This may only be invoked before the entry has reached global
135 static inline void dst_init_metrics(struct dst_entry *dst,
136 const u32 *src_metrics,
139 dst->_metrics = ((unsigned long) src_metrics) |
140 (read_only ? DST_METRICS_READ_ONLY : 0);
143 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
145 u32 *dst_metrics = dst_metrics_write_ptr(dest);
148 u32 *src_metrics = DST_METRICS_PTR(src);
150 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
154 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
156 return DST_METRICS_PTR(dst);
160 dst_metric_raw(const struct dst_entry *dst, const int metric)
162 u32 *p = DST_METRICS_PTR(dst);
168 dst_metric(const struct dst_entry *dst, const int metric)
170 WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
171 metric == RTAX_ADVMSS ||
173 return dst_metric_raw(dst, metric);
177 dst_metric_advmss(const struct dst_entry *dst)
179 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
182 advmss = dst->ops->default_advmss(dst);
187 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
189 u32 *p = dst_metrics_write_ptr(dst);
196 dst_feature(const struct dst_entry *dst, u32 feature)
198 return dst_metric(dst, RTAX_FEATURES) & feature;
201 static inline u32 dst_mtu(const struct dst_entry *dst)
203 return dst->ops->mtu(dst);
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)
209 return msecs_to_jiffies(dst_metric(dst, metric));
212 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
215 dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
219 dst_allfrag(const struct dst_entry *dst)
221 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG);
226 dst_metric_locked(const struct dst_entry *dst, int metric)
228 return dst_metric(dst, RTAX_LOCK) & (1<<metric);
231 static inline void dst_hold(struct dst_entry *dst)
234 * If your kernel compilation stops here, please check
235 * __pad_to_align_refcnt declaration in struct dst_entry
237 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
238 atomic_inc(&dst->__refcnt);
241 static inline void dst_use(struct dst_entry *dst, unsigned long time)
248 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
254 static inline struct dst_entry *dst_clone(struct dst_entry *dst)
257 atomic_inc(&dst->__refcnt);
261 extern void dst_release(struct dst_entry *dst);
263 static inline void refdst_drop(unsigned long refdst)
265 if (!(refdst & SKB_DST_NOREF))
266 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
270 * skb_dst_drop - drops skb dst
273 * Drops dst reference count if a reference was taken.
275 static inline void skb_dst_drop(struct sk_buff *skb)
277 if (skb->_skb_refdst) {
278 refdst_drop(skb->_skb_refdst);
279 skb->_skb_refdst = 0UL;
283 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
285 nskb->_skb_refdst = oskb->_skb_refdst;
286 if (!(nskb->_skb_refdst & SKB_DST_NOREF))
287 dst_clone(skb_dst(nskb));
291 * skb_dst_force - makes sure skb dst is refcounted
294 * If dst is not yet refcounted, let's do it
296 static inline void skb_dst_force(struct sk_buff *skb)
298 if (skb_dst_is_noref(skb)) {
299 WARN_ON(!rcu_read_lock_held());
300 skb->_skb_refdst &= ~SKB_DST_NOREF;
301 dst_clone(skb_dst(skb));
307 * __skb_tunnel_rx - prepare skb for rx reinsert
309 * @dev: tunnel device
311 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
312 * so make some cleanups. (no accounting done)
314 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
319 * Clear rxhash so that we can recalulate the hash for the
320 * encapsulated packet, unless we have already determine the hash
321 * over the L4 4-tuple.
325 skb_set_queue_mapping(skb, 0);
331 * skb_tunnel_rx - prepare skb for rx reinsert
333 * @dev: tunnel device
335 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
336 * so make some cleanups, and perform accounting.
337 * Note: this accounting is not SMP safe.
339 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
341 /* TODO : stats should be SMP safe */
342 dev->stats.rx_packets++;
343 dev->stats.rx_bytes += skb->len;
344 __skb_tunnel_rx(skb, dev);
347 /* Children define the path of the packet through the
348 * Linux networking. Thus, destinations are stackable.
351 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
353 struct dst_entry *child = dst_clone(skb_dst(skb)->child);
359 extern int dst_discard(struct sk_buff *skb);
360 extern void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
361 int initial_ref, int initial_obsolete,
362 unsigned short flags);
363 extern void __dst_free(struct dst_entry *dst);
364 extern struct dst_entry *dst_destroy(struct dst_entry *dst);
366 static inline void dst_free(struct dst_entry *dst)
368 if (dst->obsolete > 1)
370 if (!atomic_read(&dst->__refcnt)) {
371 dst = dst_destroy(dst);
378 static inline void dst_rcu_free(struct rcu_head *head)
380 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
384 static inline void dst_confirm(struct dst_entry *dst)
386 dst->pending_confirm = 1;
389 static inline int dst_neigh_output(struct dst_entry *dst, struct neighbour *n,
394 if (unlikely(dst->pending_confirm)) {
395 n->confirmed = jiffies;
396 dst->pending_confirm = 0;
400 if ((n->nud_state & NUD_CONNECTED) && hh->hh_len)
401 return neigh_hh_output(hh, skb);
403 return n->output(n, skb);
406 static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
408 return dst->ops->neigh_lookup(dst, NULL, daddr);
411 static inline struct neighbour *dst_neigh_lookup_skb(const struct dst_entry *dst,
414 return dst->ops->neigh_lookup(dst, skb, NULL);
417 static inline void dst_link_failure(struct sk_buff *skb)
419 struct dst_entry *dst = skb_dst(skb);
420 if (dst && dst->ops && dst->ops->link_failure)
421 dst->ops->link_failure(skb);
424 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
426 unsigned long expires = jiffies + timeout;
431 if (dst->expires == 0 || time_before(expires, dst->expires))
432 dst->expires = expires;
435 /* Output packet to network from transport. */
436 static inline int dst_output(struct sk_buff *skb)
438 return skb_dst(skb)->output(skb);
441 /* Input packet from network to transport. */
442 static inline int dst_input(struct sk_buff *skb)
444 return skb_dst(skb)->input(skb);
447 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
450 dst = dst->ops->check(dst, cookie);
454 extern void dst_init(void);
456 /* Flags for xfrm_lookup flags argument. */
458 XFRM_LOOKUP_ICMP = 1 << 0,
463 static inline struct dst_entry *xfrm_lookup(struct net *net,
464 struct dst_entry *dst_orig,
465 const struct flowi *fl, struct sock *sk,
471 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
472 const struct flowi *fl, struct sock *sk,
476 #endif /* _NET_DST_H */