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
66 unsigned short header_len
; /* more space at head required */
67 unsigned short trailer_len
; /* space to reserve at tail */
68 #ifdef CONFIG_IP_ROUTE_CLASSID
75 * Align __refcnt to a 64 bytes alignment
76 * (L1_CACHE_SIZE would be too much)
79 long __pad_to_align_refcnt
[2];
82 * __refcnt wants to be on a different cache line from
83 * input/output/ops or performance tanks badly
85 atomic_t __refcnt
; /* client references */
87 unsigned long lastuse
;
89 struct dst_entry
*next
;
90 struct rtable __rcu
*rt_next
;
91 struct rt6_info
*rt6_next
;
92 struct dn_route __rcu
*dn_next
;
96 static inline struct neighbour
*dst_get_neighbour_noref(struct dst_entry
*dst
)
98 return rcu_dereference(dst
->_neighbour
);
101 static inline struct neighbour
*dst_get_neighbour_noref_raw(struct dst_entry
*dst
)
103 return rcu_dereference_raw(dst
->_neighbour
);
106 static inline void dst_set_neighbour(struct dst_entry
*dst
, struct neighbour
*neigh
)
108 rcu_assign_pointer(dst
->_neighbour
, neigh
);
111 extern u32
*dst_cow_metrics_generic(struct dst_entry
*dst
, unsigned long old
);
112 extern const u32 dst_default_metrics
[RTAX_MAX
];
114 #define DST_METRICS_READ_ONLY 0x1UL
115 #define __DST_METRICS_PTR(Y) \
116 ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
117 #define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics)
119 static inline bool dst_metrics_read_only(const struct dst_entry
*dst
)
121 return dst
->_metrics
& DST_METRICS_READ_ONLY
;
124 extern void __dst_destroy_metrics_generic(struct dst_entry
*dst
, unsigned long old
);
126 static inline void dst_destroy_metrics_generic(struct dst_entry
*dst
)
128 unsigned long val
= dst
->_metrics
;
129 if (!(val
& DST_METRICS_READ_ONLY
))
130 __dst_destroy_metrics_generic(dst
, val
);
133 static inline u32
*dst_metrics_write_ptr(struct dst_entry
*dst
)
135 unsigned long p
= dst
->_metrics
;
139 if (p
& DST_METRICS_READ_ONLY
)
140 return dst
->ops
->cow_metrics(dst
, p
);
141 return __DST_METRICS_PTR(p
);
144 /* This may only be invoked before the entry has reached global
147 static inline void dst_init_metrics(struct dst_entry
*dst
,
148 const u32
*src_metrics
,
151 dst
->_metrics
= ((unsigned long) src_metrics
) |
152 (read_only
? DST_METRICS_READ_ONLY
: 0);
155 static inline void dst_copy_metrics(struct dst_entry
*dest
, const struct dst_entry
*src
)
157 u32
*dst_metrics
= dst_metrics_write_ptr(dest
);
160 u32
*src_metrics
= DST_METRICS_PTR(src
);
162 memcpy(dst_metrics
, src_metrics
, RTAX_MAX
* sizeof(u32
));
166 static inline u32
*dst_metrics_ptr(struct dst_entry
*dst
)
168 return DST_METRICS_PTR(dst
);
172 dst_metric_raw(const struct dst_entry
*dst
, const int metric
)
174 u32
*p
= DST_METRICS_PTR(dst
);
180 dst_metric(const struct dst_entry
*dst
, const int metric
)
182 WARN_ON_ONCE(metric
== RTAX_HOPLIMIT
||
183 metric
== RTAX_ADVMSS
||
185 return dst_metric_raw(dst
, metric
);
189 dst_metric_advmss(const struct dst_entry
*dst
)
191 u32 advmss
= dst_metric_raw(dst
, RTAX_ADVMSS
);
194 advmss
= dst
->ops
->default_advmss(dst
);
199 static inline void dst_metric_set(struct dst_entry
*dst
, int metric
, u32 val
)
201 u32
*p
= dst_metrics_write_ptr(dst
);
208 dst_feature(const struct dst_entry
*dst
, u32 feature
)
210 return dst_metric(dst
, RTAX_FEATURES
) & feature
;
213 static inline u32
dst_mtu(const struct dst_entry
*dst
)
215 return dst
->ops
->mtu(dst
);
218 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
219 static inline unsigned long dst_metric_rtt(const struct dst_entry
*dst
, int metric
)
221 return msecs_to_jiffies(dst_metric(dst
, metric
));
224 static inline void set_dst_metric_rtt(struct dst_entry
*dst
, int metric
,
227 dst_metric_set(dst
, metric
, jiffies_to_msecs(rtt
));
231 dst_allfrag(const struct dst_entry
*dst
)
233 int ret
= dst_feature(dst
, RTAX_FEATURE_ALLFRAG
);
238 dst_metric_locked(const struct dst_entry
*dst
, int metric
)
240 return dst_metric(dst
, RTAX_LOCK
) & (1<<metric
);
243 static inline void dst_hold(struct dst_entry
* dst
)
246 * If your kernel compilation stops here, please check
247 * __pad_to_align_refcnt declaration in struct dst_entry
249 BUILD_BUG_ON(offsetof(struct dst_entry
, __refcnt
) & 63);
250 atomic_inc(&dst
->__refcnt
);
253 static inline void dst_use(struct dst_entry
*dst
, unsigned long time
)
260 static inline void dst_use_noref(struct dst_entry
*dst
, unsigned long time
)
267 struct dst_entry
* dst_clone(struct dst_entry
* dst
)
270 atomic_inc(&dst
->__refcnt
);
274 extern void dst_release(struct dst_entry
*dst
);
276 static inline void refdst_drop(unsigned long refdst
)
278 if (!(refdst
& SKB_DST_NOREF
))
279 dst_release((struct dst_entry
*)(refdst
& SKB_DST_PTRMASK
));
283 * skb_dst_drop - drops skb dst
286 * Drops dst reference count if a reference was taken.
288 static inline void skb_dst_drop(struct sk_buff
*skb
)
290 if (skb
->_skb_refdst
) {
291 refdst_drop(skb
->_skb_refdst
);
292 skb
->_skb_refdst
= 0UL;
296 static inline void skb_dst_copy(struct sk_buff
*nskb
, const struct sk_buff
*oskb
)
298 nskb
->_skb_refdst
= oskb
->_skb_refdst
;
299 if (!(nskb
->_skb_refdst
& SKB_DST_NOREF
))
300 dst_clone(skb_dst(nskb
));
304 * skb_dst_force - makes sure skb dst is refcounted
307 * If dst is not yet refcounted, let's do it
309 static inline void skb_dst_force(struct sk_buff
*skb
)
311 if (skb_dst_is_noref(skb
)) {
312 WARN_ON(!rcu_read_lock_held());
313 skb
->_skb_refdst
&= ~SKB_DST_NOREF
;
314 dst_clone(skb_dst(skb
));
320 * __skb_tunnel_rx - prepare skb for rx reinsert
322 * @dev: tunnel device
324 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
325 * so make some cleanups. (no accounting done)
327 static inline void __skb_tunnel_rx(struct sk_buff
*skb
, struct net_device
*dev
)
332 * Clear rxhash so that we can recalulate the hash for the
333 * encapsulated packet, unless we have already determine the hash
334 * over the L4 4-tuple.
338 skb_set_queue_mapping(skb
, 0);
344 * skb_tunnel_rx - prepare skb for rx reinsert
346 * @dev: tunnel device
348 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
349 * so make some cleanups, and perform accounting.
350 * Note: this accounting is not SMP safe.
352 static inline void skb_tunnel_rx(struct sk_buff
*skb
, struct net_device
*dev
)
354 /* TODO : stats should be SMP safe */
355 dev
->stats
.rx_packets
++;
356 dev
->stats
.rx_bytes
+= skb
->len
;
357 __skb_tunnel_rx(skb
, dev
);
360 /* Children define the path of the packet through the
361 * Linux networking. Thus, destinations are stackable.
364 static inline struct dst_entry
*skb_dst_pop(struct sk_buff
*skb
)
366 struct dst_entry
*child
= dst_clone(skb_dst(skb
)->child
);
372 extern int dst_discard(struct sk_buff
*skb
);
373 extern void *dst_alloc(struct dst_ops
* ops
, struct net_device
*dev
,
374 int initial_ref
, int initial_obsolete
, int flags
);
375 extern void __dst_free(struct dst_entry
* dst
);
376 extern struct dst_entry
*dst_destroy(struct dst_entry
* dst
);
378 static inline void dst_free(struct dst_entry
* dst
)
380 if (dst
->obsolete
> 1)
382 if (!atomic_read(&dst
->__refcnt
)) {
383 dst
= dst_destroy(dst
);
390 static inline void dst_rcu_free(struct rcu_head
*head
)
392 struct dst_entry
*dst
= container_of(head
, struct dst_entry
, rcu_head
);
396 static inline void dst_confirm(struct dst_entry
*dst
)
402 n
= dst_get_neighbour_noref(dst
);
408 static inline struct neighbour
*dst_neigh_lookup(const struct dst_entry
*dst
, const void *daddr
)
410 return dst
->ops
->neigh_lookup(dst
, daddr
);
413 static inline void dst_link_failure(struct sk_buff
*skb
)
415 struct dst_entry
*dst
= skb_dst(skb
);
416 if (dst
&& dst
->ops
&& dst
->ops
->link_failure
)
417 dst
->ops
->link_failure(skb
);
420 static inline void dst_set_expires(struct dst_entry
*dst
, int timeout
)
422 unsigned long expires
= jiffies
+ timeout
;
427 if (dst
->expires
== 0 || time_before(expires
, dst
->expires
))
428 dst
->expires
= expires
;
431 /* Output packet to network from transport. */
432 static inline int dst_output(struct sk_buff
*skb
)
434 return skb_dst(skb
)->output(skb
);
437 /* Input packet from network to transport. */
438 static inline int dst_input(struct sk_buff
*skb
)
440 return skb_dst(skb
)->input(skb
);
443 static inline struct dst_entry
*dst_check(struct dst_entry
*dst
, u32 cookie
)
446 dst
= dst
->ops
->check(dst
, cookie
);
450 extern void dst_init(void);
452 /* Flags for xfrm_lookup flags argument. */
454 XFRM_LOOKUP_ICMP
= 1 << 0,
459 static inline struct dst_entry
*xfrm_lookup(struct net
*net
,
460 struct dst_entry
*dst_orig
,
461 const struct flowi
*fl
, struct sock
*sk
,
467 extern struct dst_entry
*xfrm_lookup(struct net
*net
, struct dst_entry
*dst_orig
,
468 const struct flowi
*fl
, struct sock
*sk
,
472 #endif /* _NET_DST_H */