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
;
39 unsigned long expires
;
40 struct dst_entry
*path
;
41 struct dst_entry
*from
;
43 struct xfrm_state
*xfrm
;
47 int (*input
)(struct sk_buff
*);
48 int (*output
)(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
);
51 #define DST_HOST 0x0001
52 #define DST_NOXFRM 0x0002
53 #define DST_NOPOLICY 0x0004
54 #define DST_NOHASH 0x0008
55 #define DST_NOCACHE 0x0010
56 #define DST_NOCOUNT 0x0020
57 #define DST_FAKE_RTABLE 0x0040
58 #define DST_XFRM_TUNNEL 0x0080
59 #define DST_XFRM_QUEUE 0x0100
60 #define DST_METADATA 0x0200
64 /* A non-zero value of dst->obsolete forces by-hand validation
65 * of the route entry. Positive values are set by the generic
66 * dst layer to indicate that the entry has been forcefully
69 * Negative values are used by the implementation layer code to
70 * force invocation of the dst_ops->check() method.
73 #define DST_OBSOLETE_NONE 0
74 #define DST_OBSOLETE_DEAD 2
75 #define DST_OBSOLETE_FORCE_CHK -1
76 #define DST_OBSOLETE_KILL -2
77 unsigned short header_len
; /* more space at head required */
78 unsigned short trailer_len
; /* space to reserve at tail */
79 unsigned short __pad3
;
81 #ifdef CONFIG_IP_ROUTE_CLASSID
89 * Align __refcnt to a 64 bytes alignment
90 * (L1_CACHE_SIZE would be too much)
92 long __pad_to_align_refcnt
[2];
95 * __refcnt wants to be on a different cache line from
96 * input/output/ops or performance tanks badly
98 atomic_t __refcnt
; /* client references */
100 unsigned long lastuse
;
101 struct lwtunnel_state
*lwtstate
;
103 struct dst_entry
*next
;
104 struct rtable __rcu
*rt_next
;
105 struct rt6_info
*rt6_next
;
106 struct dn_route __rcu
*dn_next
;
111 u32 metrics
[RTAX_MAX
];
114 extern const struct dst_metrics dst_default_metrics
;
116 u32
*dst_cow_metrics_generic(struct dst_entry
*dst
, unsigned long old
);
118 #define DST_METRICS_READ_ONLY 0x1UL
119 #define DST_METRICS_REFCOUNTED 0x2UL
120 #define DST_METRICS_FLAGS 0x3UL
121 #define __DST_METRICS_PTR(Y) \
122 ((u32 *)((Y) & ~DST_METRICS_FLAGS))
123 #define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics)
125 static inline bool dst_metrics_read_only(const struct dst_entry
*dst
)
127 return dst
->_metrics
& DST_METRICS_READ_ONLY
;
130 void __dst_destroy_metrics_generic(struct dst_entry
*dst
, unsigned long old
);
132 static inline void dst_destroy_metrics_generic(struct dst_entry
*dst
)
134 unsigned long val
= dst
->_metrics
;
135 if (!(val
& DST_METRICS_READ_ONLY
))
136 __dst_destroy_metrics_generic(dst
, val
);
139 static inline u32
*dst_metrics_write_ptr(struct dst_entry
*dst
)
141 unsigned long p
= dst
->_metrics
;
145 if (p
& DST_METRICS_READ_ONLY
)
146 return dst
->ops
->cow_metrics(dst
, p
);
147 return __DST_METRICS_PTR(p
);
150 /* This may only be invoked before the entry has reached global
153 static inline void dst_init_metrics(struct dst_entry
*dst
,
154 const u32
*src_metrics
,
157 dst
->_metrics
= ((unsigned long) src_metrics
) |
158 (read_only
? DST_METRICS_READ_ONLY
: 0);
161 static inline void dst_copy_metrics(struct dst_entry
*dest
, const struct dst_entry
*src
)
163 u32
*dst_metrics
= dst_metrics_write_ptr(dest
);
166 u32
*src_metrics
= DST_METRICS_PTR(src
);
168 memcpy(dst_metrics
, src_metrics
, RTAX_MAX
* sizeof(u32
));
172 static inline u32
*dst_metrics_ptr(struct dst_entry
*dst
)
174 return DST_METRICS_PTR(dst
);
178 dst_metric_raw(const struct dst_entry
*dst
, const int metric
)
180 u32
*p
= DST_METRICS_PTR(dst
);
186 dst_metric(const struct dst_entry
*dst
, const int metric
)
188 WARN_ON_ONCE(metric
== RTAX_HOPLIMIT
||
189 metric
== RTAX_ADVMSS
||
191 return dst_metric_raw(dst
, metric
);
195 dst_metric_advmss(const struct dst_entry
*dst
)
197 u32 advmss
= dst_metric_raw(dst
, RTAX_ADVMSS
);
200 advmss
= dst
->ops
->default_advmss(dst
);
205 static inline void dst_metric_set(struct dst_entry
*dst
, int metric
, u32 val
)
207 u32
*p
= dst_metrics_write_ptr(dst
);
213 /* Kernel-internal feature bits that are unallocated in user space. */
214 #define DST_FEATURE_ECN_CA (1 << 31)
216 #define DST_FEATURE_MASK (DST_FEATURE_ECN_CA)
217 #define DST_FEATURE_ECN_MASK (DST_FEATURE_ECN_CA | RTAX_FEATURE_ECN)
220 dst_feature(const struct dst_entry
*dst
, u32 feature
)
222 return dst_metric(dst
, RTAX_FEATURES
) & feature
;
225 static inline u32
dst_mtu(const struct dst_entry
*dst
)
227 return dst
->ops
->mtu(dst
);
230 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
231 static inline unsigned long dst_metric_rtt(const struct dst_entry
*dst
, int metric
)
233 return msecs_to_jiffies(dst_metric(dst
, metric
));
237 dst_allfrag(const struct dst_entry
*dst
)
239 int ret
= dst_feature(dst
, RTAX_FEATURE_ALLFRAG
);
244 dst_metric_locked(const struct dst_entry
*dst
, int metric
)
246 return dst_metric(dst
, RTAX_LOCK
) & (1<<metric
);
249 static inline void dst_hold(struct dst_entry
*dst
)
252 * If your kernel compilation stops here, please check
253 * __pad_to_align_refcnt declaration in struct dst_entry
255 BUILD_BUG_ON(offsetof(struct dst_entry
, __refcnt
) & 63);
256 atomic_inc(&dst
->__refcnt
);
259 static inline void dst_use(struct dst_entry
*dst
, unsigned long time
)
266 static inline void dst_use_noref(struct dst_entry
*dst
, unsigned long time
)
272 static inline struct dst_entry
*dst_clone(struct dst_entry
*dst
)
275 atomic_inc(&dst
->__refcnt
);
279 void dst_release(struct dst_entry
*dst
);
281 static inline void refdst_drop(unsigned long refdst
)
283 if (!(refdst
& SKB_DST_NOREF
))
284 dst_release((struct dst_entry
*)(refdst
& SKB_DST_PTRMASK
));
288 * skb_dst_drop - drops skb dst
291 * Drops dst reference count if a reference was taken.
293 static inline void skb_dst_drop(struct sk_buff
*skb
)
295 if (skb
->_skb_refdst
) {
296 refdst_drop(skb
->_skb_refdst
);
297 skb
->_skb_refdst
= 0UL;
301 static inline void __skb_dst_copy(struct sk_buff
*nskb
, unsigned long refdst
)
303 nskb
->_skb_refdst
= refdst
;
304 if (!(nskb
->_skb_refdst
& SKB_DST_NOREF
))
305 dst_clone(skb_dst(nskb
));
308 static inline void skb_dst_copy(struct sk_buff
*nskb
, const struct sk_buff
*oskb
)
310 __skb_dst_copy(nskb
, oskb
->_skb_refdst
);
314 * skb_dst_force - makes sure skb dst is refcounted
317 * If dst is not yet refcounted, let's do it
319 static inline void skb_dst_force(struct sk_buff
*skb
)
321 if (skb_dst_is_noref(skb
)) {
322 WARN_ON(!rcu_read_lock_held());
323 skb
->_skb_refdst
&= ~SKB_DST_NOREF
;
324 dst_clone(skb_dst(skb
));
329 * dst_hold_safe - Take a reference on a dst if possible
330 * @dst: pointer to dst entry
332 * This helper returns false if it could not safely
333 * take a reference on a dst.
335 static inline bool dst_hold_safe(struct dst_entry
*dst
)
337 if (dst
->flags
& DST_NOCACHE
)
338 return atomic_inc_not_zero(&dst
->__refcnt
);
344 * skb_dst_force_safe - makes sure skb dst is refcounted
347 * If dst is not yet refcounted and not destroyed, grab a ref on it.
349 static inline void skb_dst_force_safe(struct sk_buff
*skb
)
351 if (skb_dst_is_noref(skb
)) {
352 struct dst_entry
*dst
= skb_dst(skb
);
354 if (!dst_hold_safe(dst
))
357 skb
->_skb_refdst
= (unsigned long)dst
;
363 * __skb_tunnel_rx - prepare skb for rx reinsert
365 * @dev: tunnel device
366 * @net: netns for packet i/o
368 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
369 * so make some cleanups. (no accounting done)
371 static inline void __skb_tunnel_rx(struct sk_buff
*skb
, struct net_device
*dev
,
377 * Clear hash so that we can recalulate the hash for the
378 * encapsulated packet, unless we have already determine the hash
379 * over the L4 4-tuple.
381 skb_clear_hash_if_not_l4(skb
);
382 skb_set_queue_mapping(skb
, 0);
383 skb_scrub_packet(skb
, !net_eq(net
, dev_net(dev
)));
387 * skb_tunnel_rx - prepare skb for rx reinsert
389 * @dev: tunnel device
391 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
392 * so make some cleanups, and perform accounting.
393 * Note: this accounting is not SMP safe.
395 static inline void skb_tunnel_rx(struct sk_buff
*skb
, struct net_device
*dev
,
398 /* TODO : stats should be SMP safe */
399 dev
->stats
.rx_packets
++;
400 dev
->stats
.rx_bytes
+= skb
->len
;
401 __skb_tunnel_rx(skb
, dev
, net
);
404 static inline u32
dst_tclassid(const struct sk_buff
*skb
)
406 #ifdef CONFIG_IP_ROUTE_CLASSID
407 const struct dst_entry
*dst
;
411 return dst
->tclassid
;
416 int dst_discard_out(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
);
417 static inline int dst_discard(struct sk_buff
*skb
)
419 return dst_discard_out(&init_net
, skb
->sk
, skb
);
421 void *dst_alloc(struct dst_ops
*ops
, struct net_device
*dev
, int initial_ref
,
422 int initial_obsolete
, unsigned short flags
);
423 void dst_init(struct dst_entry
*dst
, struct dst_ops
*ops
,
424 struct net_device
*dev
, int initial_ref
, int initial_obsolete
,
425 unsigned short flags
);
426 void __dst_free(struct dst_entry
*dst
);
427 struct dst_entry
*dst_destroy(struct dst_entry
*dst
);
429 static inline void dst_free(struct dst_entry
*dst
)
431 if (dst
->obsolete
> 0)
433 if (!atomic_read(&dst
->__refcnt
)) {
434 dst
= dst_destroy(dst
);
441 static inline void dst_rcu_free(struct rcu_head
*head
)
443 struct dst_entry
*dst
= container_of(head
, struct dst_entry
, rcu_head
);
447 static inline void dst_confirm(struct dst_entry
*dst
)
451 static inline struct neighbour
*dst_neigh_lookup(const struct dst_entry
*dst
, const void *daddr
)
453 struct neighbour
*n
= dst
->ops
->neigh_lookup(dst
, NULL
, daddr
);
454 return IS_ERR(n
) ? NULL
: n
;
457 static inline struct neighbour
*dst_neigh_lookup_skb(const struct dst_entry
*dst
,
460 struct neighbour
*n
= dst
->ops
->neigh_lookup(dst
, skb
, NULL
);
461 return IS_ERR(n
) ? NULL
: n
;
464 static inline void dst_confirm_neigh(const struct dst_entry
*dst
,
467 if (dst
->ops
->confirm_neigh
)
468 dst
->ops
->confirm_neigh(dst
, daddr
);
471 static inline void dst_link_failure(struct sk_buff
*skb
)
473 struct dst_entry
*dst
= skb_dst(skb
);
474 if (dst
&& dst
->ops
&& dst
->ops
->link_failure
)
475 dst
->ops
->link_failure(skb
);
478 static inline void dst_set_expires(struct dst_entry
*dst
, int timeout
)
480 unsigned long expires
= jiffies
+ timeout
;
485 if (dst
->expires
== 0 || time_before(expires
, dst
->expires
))
486 dst
->expires
= expires
;
489 /* Output packet to network from transport. */
490 static inline int dst_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
492 return skb_dst(skb
)->output(net
, sk
, skb
);
495 /* Input packet from network to transport. */
496 static inline int dst_input(struct sk_buff
*skb
)
498 return skb_dst(skb
)->input(skb
);
501 static inline struct dst_entry
*dst_check(struct dst_entry
*dst
, u32 cookie
)
504 dst
= dst
->ops
->check(dst
, cookie
);
508 void dst_subsys_init(void);
510 /* Flags for xfrm_lookup flags argument. */
512 XFRM_LOOKUP_ICMP
= 1 << 0,
513 XFRM_LOOKUP_QUEUE
= 1 << 1,
514 XFRM_LOOKUP_KEEP_DST_REF
= 1 << 2,
519 static inline struct dst_entry
*xfrm_lookup(struct net
*net
,
520 struct dst_entry
*dst_orig
,
521 const struct flowi
*fl
,
522 const struct sock
*sk
,
528 static inline struct dst_entry
*xfrm_lookup_route(struct net
*net
,
529 struct dst_entry
*dst_orig
,
530 const struct flowi
*fl
,
531 const struct sock
*sk
,
537 static inline struct xfrm_state
*dst_xfrm(const struct dst_entry
*dst
)
543 struct dst_entry
*xfrm_lookup(struct net
*net
, struct dst_entry
*dst_orig
,
544 const struct flowi
*fl
, const struct sock
*sk
,
547 struct dst_entry
*xfrm_lookup_route(struct net
*net
, struct dst_entry
*dst_orig
,
548 const struct flowi
*fl
, const struct sock
*sk
,
551 /* skb attached with this dst needs transformation if dst->xfrm is valid */
552 static inline struct xfrm_state
*dst_xfrm(const struct dst_entry
*dst
)
558 #endif /* _NET_DST_H */