xref: /linux/include/net/dst.h (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
1 /*
2  * net/dst.h	Protocol independent destination cache definitions.
3  *
4  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
5  *
6  */
7 
8 #ifndef _NET_DST_H
9 #define _NET_DST_H
10 
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>
18 
19 /*
20  * 0 - no debugging messages
21  * 1 - rare events and bugs (default)
22  * 2 - trace mode.
23  */
24 #define RT_CACHE_DEBUG		0
25 
26 #define DST_GC_MIN	(HZ/10)
27 #define DST_GC_INC	(HZ/2)
28 #define DST_GC_MAX	(120*HZ)
29 
30 /* Each dst_entry has reference count and sits in some parent list(s).
31  * When it is removed from parent list, it is "freed" (dst_free).
32  * After this it enters dead state (dst->obsolete > 0) and if its refcnt
33  * is zero, it can be destroyed immediately, otherwise it is added
34  * to gc list and garbage collector periodically checks the refcnt.
35  */
36 
37 struct sk_buff;
38 
39 struct dst_entry {
40 	struct rcu_head		rcu_head;
41 	struct dst_entry	*child;
42 	struct net_device       *dev;
43 	short			error;
44 	short			obsolete;
45 	int			flags;
46 #define DST_HOST		1
47 #define DST_NOXFRM		2
48 #define DST_NOPOLICY		4
49 #define DST_NOHASH		8
50 	unsigned long		expires;
51 
52 	unsigned short		header_len;	/* more space at head required */
53 	unsigned short		trailer_len;	/* space to reserve at tail */
54 
55 	unsigned int		rate_tokens;
56 	unsigned long		rate_last;	/* rate limiting for ICMP */
57 
58 	struct dst_entry	*path;
59 
60 	struct neighbour	*neighbour;
61 	struct hh_cache		*hh;
62 #ifdef CONFIG_XFRM
63 	struct xfrm_state	*xfrm;
64 #else
65 	void			*__pad1;
66 #endif
67 	int			(*input)(struct sk_buff*);
68 	int			(*output)(struct sk_buff*);
69 
70 	struct  dst_ops	        *ops;
71 
72 	u32			metrics[RTAX_MAX];
73 
74 #ifdef CONFIG_NET_CLS_ROUTE
75 	__u32			tclassid;
76 #else
77 	__u32			__pad2;
78 #endif
79 
80 
81 	/*
82 	 * Align __refcnt to a 64 bytes alignment
83 	 * (L1_CACHE_SIZE would be too much)
84 	 */
85 #ifdef CONFIG_64BIT
86 	long			__pad_to_align_refcnt[1];
87 #endif
88 	/*
89 	 * __refcnt wants to be on a different cache line from
90 	 * input/output/ops or performance tanks badly
91 	 */
92 	atomic_t		__refcnt;	/* client references	*/
93 	int			__use;
94 	unsigned long		lastuse;
95 	union {
96 		struct dst_entry *next;
97 		struct rtable    *rt_next;
98 		struct rt6_info   *rt6_next;
99 		struct dn_route  *dn_next;
100 	};
101 };
102 
103 #ifdef __KERNEL__
104 
105 static inline u32
106 dst_metric(const struct dst_entry *dst, int metric)
107 {
108 	return dst->metrics[metric-1];
109 }
110 
111 static inline u32
112 dst_feature(const struct dst_entry *dst, u32 feature)
113 {
114 	return dst_metric(dst, RTAX_FEATURES) & feature;
115 }
116 
117 static inline u32 dst_mtu(const struct dst_entry *dst)
118 {
119 	u32 mtu = dst_metric(dst, RTAX_MTU);
120 	/*
121 	 * Alexey put it here, so ask him about it :)
122 	 */
123 	barrier();
124 	return mtu;
125 }
126 
127 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
128 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
129 {
130 	return msecs_to_jiffies(dst_metric(dst, metric));
131 }
132 
133 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
134 				      unsigned long rtt)
135 {
136 	dst->metrics[metric-1] = jiffies_to_msecs(rtt);
137 }
138 
139 static inline u32
140 dst_allfrag(const struct dst_entry *dst)
141 {
142 	int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
143 	/* Yes, _exactly_. This is paranoia. */
144 	barrier();
145 	return ret;
146 }
147 
148 static inline int
149 dst_metric_locked(struct dst_entry *dst, int metric)
150 {
151 	return dst_metric(dst, RTAX_LOCK) & (1<<metric);
152 }
153 
154 static inline void dst_hold(struct dst_entry * dst)
155 {
156 	/*
157 	 * If your kernel compilation stops here, please check
158 	 * __pad_to_align_refcnt declaration in struct dst_entry
159 	 */
160 	BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
161 	atomic_inc(&dst->__refcnt);
162 }
163 
164 static inline void dst_use(struct dst_entry *dst, unsigned long time)
165 {
166 	dst_hold(dst);
167 	dst->__use++;
168 	dst->lastuse = time;
169 }
170 
171 static inline
172 struct dst_entry * dst_clone(struct dst_entry * dst)
173 {
174 	if (dst)
175 		atomic_inc(&dst->__refcnt);
176 	return dst;
177 }
178 
179 extern void dst_release(struct dst_entry *dst);
180 static inline void skb_dst_drop(struct sk_buff *skb)
181 {
182 	if (skb->_skb_dst)
183 		dst_release(skb_dst(skb));
184 	skb->_skb_dst = 0UL;
185 }
186 
187 /* Children define the path of the packet through the
188  * Linux networking.  Thus, destinations are stackable.
189  */
190 
191 static inline struct dst_entry *dst_pop(struct dst_entry *dst)
192 {
193 	struct dst_entry *child = dst_clone(dst->child);
194 
195 	dst_release(dst);
196 	return child;
197 }
198 
199 extern int dst_discard(struct sk_buff *skb);
200 extern void * dst_alloc(struct dst_ops * ops);
201 extern void __dst_free(struct dst_entry * dst);
202 extern struct dst_entry *dst_destroy(struct dst_entry * dst);
203 
204 static inline void dst_free(struct dst_entry * dst)
205 {
206 	if (dst->obsolete > 1)
207 		return;
208 	if (!atomic_read(&dst->__refcnt)) {
209 		dst = dst_destroy(dst);
210 		if (!dst)
211 			return;
212 	}
213 	__dst_free(dst);
214 }
215 
216 static inline void dst_rcu_free(struct rcu_head *head)
217 {
218 	struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
219 	dst_free(dst);
220 }
221 
222 static inline void dst_confirm(struct dst_entry *dst)
223 {
224 	if (dst)
225 		neigh_confirm(dst->neighbour);
226 }
227 
228 static inline void dst_negative_advice(struct dst_entry **dst_p,
229 				       struct sock *sk)
230 {
231 	struct dst_entry * dst = *dst_p;
232 	if (dst && dst->ops->negative_advice) {
233 		*dst_p = dst->ops->negative_advice(dst);
234 
235 		if (dst != *dst_p) {
236 			extern void sk_reset_txq(struct sock *sk);
237 
238 			sk_reset_txq(sk);
239 		}
240 	}
241 }
242 
243 static inline void dst_link_failure(struct sk_buff *skb)
244 {
245 	struct dst_entry *dst = skb_dst(skb);
246 	if (dst && dst->ops && dst->ops->link_failure)
247 		dst->ops->link_failure(skb);
248 }
249 
250 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
251 {
252 	unsigned long expires = jiffies + timeout;
253 
254 	if (expires == 0)
255 		expires = 1;
256 
257 	if (dst->expires == 0 || time_before(expires, dst->expires))
258 		dst->expires = expires;
259 }
260 
261 /* Output packet to network from transport.  */
262 static inline int dst_output(struct sk_buff *skb)
263 {
264 	return skb_dst(skb)->output(skb);
265 }
266 
267 /* Input packet from network to transport.  */
268 static inline int dst_input(struct sk_buff *skb)
269 {
270 	return skb_dst(skb)->input(skb);
271 }
272 
273 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
274 {
275 	if (dst->obsolete)
276 		dst = dst->ops->check(dst, cookie);
277 	return dst;
278 }
279 
280 extern void		dst_init(void);
281 
282 /* Flags for xfrm_lookup flags argument. */
283 enum {
284 	XFRM_LOOKUP_WAIT = 1 << 0,
285 	XFRM_LOOKUP_ICMP = 1 << 1,
286 };
287 
288 struct flowi;
289 #ifndef CONFIG_XFRM
290 static inline int xfrm_lookup(struct net *net, struct dst_entry **dst_p,
291 			      struct flowi *fl, struct sock *sk, int flags)
292 {
293 	return 0;
294 }
295 static inline int __xfrm_lookup(struct net *net, struct dst_entry **dst_p,
296 				struct flowi *fl, struct sock *sk, int flags)
297 {
298 	return 0;
299 }
300 #else
301 extern int xfrm_lookup(struct net *net, struct dst_entry **dst_p,
302 		       struct flowi *fl, struct sock *sk, int flags);
303 extern int __xfrm_lookup(struct net *net, struct dst_entry **dst_p,
304 			 struct flowi *fl, struct sock *sk, int flags);
305 #endif
306 #endif
307 
308 #endif /* _NET_DST_H */
309