net_sched: sfb: optimize enqueue on full queue
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / include / net / dst.h
blob4fb6c43817918992f8334c49022d5184ea45c8e4
1 /*
2 * net/dst.h Protocol independent destination cache definitions.
4 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 */
8 #ifndef _NET_DST_H
9 #define _NET_DST_H
11 #include <net/dst_ops.h>
12 #include <linux/netdevice.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/rcupdate.h>
15 #include <linux/jiffies.h>
16 #include <net/neighbour.h>
17 #include <asm/processor.h>
19 #define DST_GC_MIN (HZ/10)
20 #define DST_GC_INC (HZ/2)
21 #define DST_GC_MAX (120*HZ)
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.
30 struct sk_buff;
32 struct dst_entry {
33 struct rcu_head rcu_head;
34 struct dst_entry *child;
35 struct net_device *dev;
36 struct dst_ops *ops;
37 unsigned long _metrics;
38 unsigned long expires;
39 struct dst_entry *path;
40 struct neighbour __rcu *_neighbour;
41 #ifdef CONFIG_XFRM
42 struct xfrm_state *xfrm;
43 #else
44 void *__pad1;
45 #endif
46 int (*input)(struct sk_buff*);
47 int (*output)(struct sk_buff*);
49 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
57 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 */
61 #ifdef CONFIG_IP_ROUTE_CLASSID
62 __u32 tclassid;
63 #else
64 __u32 __pad2;
65 #endif
68 * Align __refcnt to a 64 bytes alignment
69 * (L1_CACHE_SIZE would be too much)
71 #ifdef CONFIG_64BIT
72 long __pad_to_align_refcnt[2];
73 #endif
75 * __refcnt wants to be on a different cache line from
76 * input/output/ops or performance tanks badly
78 atomic_t __refcnt; /* client references */
79 int __use;
80 unsigned long lastuse;
81 union {
82 struct dst_entry *next;
83 struct rtable __rcu *rt_next;
84 struct rt6_info *rt6_next;
85 struct dn_route __rcu *dn_next;
89 static inline struct neighbour *dst_get_neighbour(struct dst_entry *dst)
91 return rcu_dereference(dst->_neighbour);
94 static inline struct neighbour *dst_get_neighbour_raw(struct dst_entry *dst)
96 return rcu_dereference_raw(dst->_neighbour);
99 static inline void dst_set_neighbour(struct dst_entry *dst, struct neighbour *neigh)
101 rcu_assign_pointer(dst->_neighbour, neigh);
104 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
105 extern const u32 dst_default_metrics[RTAX_MAX];
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)
112 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
114 return dst->_metrics & DST_METRICS_READ_ONLY;
117 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
119 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
121 unsigned long val = dst->_metrics;
122 if (!(val & DST_METRICS_READ_ONLY))
123 __dst_destroy_metrics_generic(dst, val);
126 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
128 unsigned long p = dst->_metrics;
130 BUG_ON(!p);
132 if (p & DST_METRICS_READ_ONLY)
133 return dst->ops->cow_metrics(dst, p);
134 return __DST_METRICS_PTR(p);
137 /* This may only be invoked before the entry has reached global
138 * visibility.
140 static inline void dst_init_metrics(struct dst_entry *dst,
141 const u32 *src_metrics,
142 bool read_only)
144 dst->_metrics = ((unsigned long) src_metrics) |
145 (read_only ? DST_METRICS_READ_ONLY : 0);
148 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
150 u32 *dst_metrics = dst_metrics_write_ptr(dest);
152 if (dst_metrics) {
153 u32 *src_metrics = DST_METRICS_PTR(src);
155 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
159 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
161 return DST_METRICS_PTR(dst);
164 static inline u32
165 dst_metric_raw(const struct dst_entry *dst, const int metric)
167 u32 *p = DST_METRICS_PTR(dst);
169 return p[metric-1];
172 static inline u32
173 dst_metric(const struct dst_entry *dst, const int metric)
175 WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
176 metric == RTAX_ADVMSS ||
177 metric == RTAX_MTU);
178 return dst_metric_raw(dst, metric);
181 static inline u32
182 dst_metric_advmss(const struct dst_entry *dst)
184 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
186 if (!advmss)
187 advmss = dst->ops->default_advmss(dst);
189 return advmss;
192 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
194 u32 *p = dst_metrics_write_ptr(dst);
196 if (p)
197 p[metric-1] = val;
200 static inline u32
201 dst_feature(const struct dst_entry *dst, u32 feature)
203 return dst_metric(dst, RTAX_FEATURES) & feature;
206 static inline u32 dst_mtu(const struct dst_entry *dst)
208 u32 mtu = dst_metric_raw(dst, RTAX_MTU);
210 if (!mtu)
211 mtu = dst->ops->default_mtu(dst);
213 return mtu;
216 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
217 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
219 return msecs_to_jiffies(dst_metric(dst, metric));
222 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
223 unsigned long rtt)
225 dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
228 static inline u32
229 dst_allfrag(const struct dst_entry *dst)
231 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG);
232 return ret;
235 static inline int
236 dst_metric_locked(const struct dst_entry *dst, int metric)
238 return dst_metric(dst, RTAX_LOCK) & (1<<metric);
241 static inline void dst_hold(struct dst_entry * dst)
244 * If your kernel compilation stops here, please check
245 * __pad_to_align_refcnt declaration in struct dst_entry
247 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
248 atomic_inc(&dst->__refcnt);
251 static inline void dst_use(struct dst_entry *dst, unsigned long time)
253 dst_hold(dst);
254 dst->__use++;
255 dst->lastuse = time;
258 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
260 dst->__use++;
261 dst->lastuse = time;
264 static inline
265 struct dst_entry * dst_clone(struct dst_entry * dst)
267 if (dst)
268 atomic_inc(&dst->__refcnt);
269 return dst;
272 extern void dst_release(struct dst_entry *dst);
274 static inline void refdst_drop(unsigned long refdst)
276 if (!(refdst & SKB_DST_NOREF))
277 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
281 * skb_dst_drop - drops skb dst
282 * @skb: buffer
284 * Drops dst reference count if a reference was taken.
286 static inline void skb_dst_drop(struct sk_buff *skb)
288 if (skb->_skb_refdst) {
289 refdst_drop(skb->_skb_refdst);
290 skb->_skb_refdst = 0UL;
294 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
296 nskb->_skb_refdst = oskb->_skb_refdst;
297 if (!(nskb->_skb_refdst & SKB_DST_NOREF))
298 dst_clone(skb_dst(nskb));
302 * skb_dst_force - makes sure skb dst is refcounted
303 * @skb: buffer
305 * If dst is not yet refcounted, let's do it
307 static inline void skb_dst_force(struct sk_buff *skb)
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));
318 * __skb_tunnel_rx - prepare skb for rx reinsert
319 * @skb: buffer
320 * @dev: tunnel device
322 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
323 * so make some cleanups. (no accounting done)
325 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
327 skb->dev = dev;
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.
334 if (!skb->l4_rxhash)
335 skb->rxhash = 0;
336 skb_set_queue_mapping(skb, 0);
337 skb_dst_drop(skb);
338 nf_reset(skb);
342 * skb_tunnel_rx - prepare skb for rx reinsert
343 * @skb: buffer
344 * @dev: tunnel device
346 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
347 * so make some cleanups, and perform accounting.
348 * Note: this accounting is not SMP safe.
350 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
352 /* TODO : stats should be SMP safe */
353 dev->stats.rx_packets++;
354 dev->stats.rx_bytes += skb->len;
355 __skb_tunnel_rx(skb, dev);
358 /* Children define the path of the packet through the
359 * Linux networking. Thus, destinations are stackable.
362 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
364 struct dst_entry *child = dst_clone(skb_dst(skb)->child);
366 skb_dst_drop(skb);
367 return child;
370 extern int dst_discard(struct sk_buff *skb);
371 extern void *dst_alloc(struct dst_ops * ops, struct net_device *dev,
372 int initial_ref, int initial_obsolete, int flags);
373 extern void __dst_free(struct dst_entry * dst);
374 extern struct dst_entry *dst_destroy(struct dst_entry * dst);
376 static inline void dst_free(struct dst_entry * dst)
378 if (dst->obsolete > 1)
379 return;
380 if (!atomic_read(&dst->__refcnt)) {
381 dst = dst_destroy(dst);
382 if (!dst)
383 return;
385 __dst_free(dst);
388 static inline void dst_rcu_free(struct rcu_head *head)
390 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
391 dst_free(dst);
394 static inline void dst_confirm(struct dst_entry *dst)
396 if (dst) {
397 struct neighbour *n;
399 rcu_read_lock();
400 n = dst_get_neighbour(dst);
401 neigh_confirm(n);
402 rcu_read_unlock();
406 static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
408 return dst->ops->neigh_lookup(dst, daddr);
411 static inline void dst_link_failure(struct sk_buff *skb)
413 struct dst_entry *dst = skb_dst(skb);
414 if (dst && dst->ops && dst->ops->link_failure)
415 dst->ops->link_failure(skb);
418 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
420 unsigned long expires = jiffies + timeout;
422 if (expires == 0)
423 expires = 1;
425 if (dst->expires == 0 || time_before(expires, dst->expires))
426 dst->expires = expires;
429 /* Output packet to network from transport. */
430 static inline int dst_output(struct sk_buff *skb)
432 return skb_dst(skb)->output(skb);
435 /* Input packet from network to transport. */
436 static inline int dst_input(struct sk_buff *skb)
438 return skb_dst(skb)->input(skb);
441 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
443 if (dst->obsolete)
444 dst = dst->ops->check(dst, cookie);
445 return dst;
448 extern void dst_init(void);
450 /* Flags for xfrm_lookup flags argument. */
451 enum {
452 XFRM_LOOKUP_ICMP = 1 << 0,
455 struct flowi;
456 #ifndef CONFIG_XFRM
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)
462 return dst_orig;
464 #else
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);
468 #endif
470 #endif /* _NET_DST_H */