1 /* SPDX-License-Identifier: GPL-2.0 */
3 * net/dst.h Protocol independent destination cache definitions.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
12 #include <net/dst_ops.h>
13 #include <linux/netdevice.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/rcupdate.h>
16 #include <linux/bug.h>
17 #include <linux/jiffies.h>
18 #include <linux/refcount.h>
19 #include <net/neighbour.h>
20 #include <asm/processor.h>
21 #include <linux/indirect_call_wrapper.h>
26 struct net_device *dev;
28 unsigned long _metrics;
29 unsigned long expires;
31 struct xfrm_state *xfrm;
35 int (*input)(struct sk_buff *);
36 int (*output)(struct net *net, struct sock *sk, struct sk_buff *skb);
39 #define DST_NOXFRM 0x0002
40 #define DST_NOPOLICY 0x0004
41 #define DST_NOCOUNT 0x0008
42 #define DST_FAKE_RTABLE 0x0010
43 #define DST_XFRM_TUNNEL 0x0020
44 #define DST_XFRM_QUEUE 0x0040
45 #define DST_METADATA 0x0080
47 /* A non-zero value of dst->obsolete forces by-hand validation
48 * of the route entry. Positive values are set by the generic
49 * dst layer to indicate that the entry has been forcefully
52 * Negative values are used by the implementation layer code to
53 * force invocation of the dst_ops->check() method.
56 #define DST_OBSOLETE_NONE 0
57 #define DST_OBSOLETE_DEAD 2
58 #define DST_OBSOLETE_FORCE_CHK -1
59 #define DST_OBSOLETE_KILL -2
60 unsigned short header_len; /* more space at head required */
61 unsigned short trailer_len; /* space to reserve at tail */
64 * __refcnt wants to be on a different cache line from
65 * input/output/ops or performance tanks badly
68 atomic_t __refcnt; /* 64-bit offset 64 */
71 unsigned long lastuse;
72 struct lwtunnel_state *lwtstate;
73 struct rcu_head rcu_head;
78 atomic_t __refcnt; /* 32-bit offset 64 */
80 netdevice_tracker dev_tracker;
84 u32 metrics[RTAX_MAX];
86 } __aligned(4); /* Low pointer bits contain DST_METRICS_FLAGS */
87 extern const struct dst_metrics dst_default_metrics;
89 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
91 #define DST_METRICS_READ_ONLY 0x1UL
92 #define DST_METRICS_REFCOUNTED 0x2UL
93 #define DST_METRICS_FLAGS 0x3UL
94 #define __DST_METRICS_PTR(Y) \
95 ((u32 *)((Y) & ~DST_METRICS_FLAGS))
96 #define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics)
98 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
100 return dst->_metrics & DST_METRICS_READ_ONLY;
103 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
105 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
107 unsigned long val = dst->_metrics;
108 if (!(val & DST_METRICS_READ_ONLY))
109 __dst_destroy_metrics_generic(dst, val);
112 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
114 unsigned long p = dst->_metrics;
118 if (p & DST_METRICS_READ_ONLY)
119 return dst->ops->cow_metrics(dst, p);
120 return __DST_METRICS_PTR(p);
123 /* This may only be invoked before the entry has reached global
126 static inline void dst_init_metrics(struct dst_entry *dst,
127 const u32 *src_metrics,
130 dst->_metrics = ((unsigned long) src_metrics) |
131 (read_only ? DST_METRICS_READ_ONLY : 0);
134 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
136 u32 *dst_metrics = dst_metrics_write_ptr(dest);
139 u32 *src_metrics = DST_METRICS_PTR(src);
141 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
145 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
147 return DST_METRICS_PTR(dst);
151 dst_metric_raw(const struct dst_entry *dst, const int metric)
153 u32 *p = DST_METRICS_PTR(dst);
159 dst_metric(const struct dst_entry *dst, const int metric)
161 WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
162 metric == RTAX_ADVMSS ||
164 return dst_metric_raw(dst, metric);
168 dst_metric_advmss(const struct dst_entry *dst)
170 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
173 advmss = dst->ops->default_advmss(dst);
178 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
180 u32 *p = dst_metrics_write_ptr(dst);
186 /* Kernel-internal feature bits that are unallocated in user space. */
187 #define DST_FEATURE_ECN_CA (1U << 31)
189 #define DST_FEATURE_MASK (DST_FEATURE_ECN_CA)
190 #define DST_FEATURE_ECN_MASK (DST_FEATURE_ECN_CA | RTAX_FEATURE_ECN)
193 dst_feature(const struct dst_entry *dst, u32 feature)
195 return dst_metric(dst, RTAX_FEATURES) & feature;
198 INDIRECT_CALLABLE_DECLARE(unsigned int ip6_mtu(const struct dst_entry *));
199 INDIRECT_CALLABLE_DECLARE(unsigned int ipv4_mtu(const struct dst_entry *));
200 static inline u32 dst_mtu(const struct dst_entry *dst)
202 return INDIRECT_CALL_INET(dst->ops->mtu, ip6_mtu, ipv4_mtu, dst);
205 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
206 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
208 return msecs_to_jiffies(dst_metric(dst, metric));
212 dst_allfrag(const struct dst_entry *dst)
214 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG);
219 dst_metric_locked(const struct dst_entry *dst, int metric)
221 return dst_metric(dst, RTAX_LOCK) & (1 << metric);
224 static inline void dst_hold(struct dst_entry *dst)
227 * If your kernel compilation stops here, please check
228 * the placement of __refcnt in struct dst_entry
230 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
231 WARN_ON(atomic_inc_not_zero(&dst->__refcnt) == 0);
234 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
236 if (unlikely(time != dst->lastuse)) {
242 static inline struct dst_entry *dst_clone(struct dst_entry *dst)
249 void dst_release(struct dst_entry *dst);
251 void dst_release_immediate(struct dst_entry *dst);
253 static inline void refdst_drop(unsigned long refdst)
255 if (!(refdst & SKB_DST_NOREF))
256 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
260 * skb_dst_drop - drops skb dst
263 * Drops dst reference count if a reference was taken.
265 static inline void skb_dst_drop(struct sk_buff *skb)
267 if (skb->_skb_refdst) {
268 refdst_drop(skb->_skb_refdst);
269 skb->_skb_refdst = 0UL;
273 static inline void __skb_dst_copy(struct sk_buff *nskb, unsigned long refdst)
275 nskb->slow_gro |= !!refdst;
276 nskb->_skb_refdst = refdst;
277 if (!(nskb->_skb_refdst & SKB_DST_NOREF))
278 dst_clone(skb_dst(nskb));
281 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
283 __skb_dst_copy(nskb, oskb->_skb_refdst);
287 * dst_hold_safe - Take a reference on a dst if possible
288 * @dst: pointer to dst entry
290 * This helper returns false if it could not safely
291 * take a reference on a dst.
293 static inline bool dst_hold_safe(struct dst_entry *dst)
295 return atomic_inc_not_zero(&dst->__refcnt);
299 * skb_dst_force - makes sure skb dst is refcounted
302 * If dst is not yet refcounted and not destroyed, grab a ref on it.
303 * Returns true if dst is refcounted.
305 static inline bool skb_dst_force(struct sk_buff *skb)
307 if (skb_dst_is_noref(skb)) {
308 struct dst_entry *dst = skb_dst(skb);
310 WARN_ON(!rcu_read_lock_held());
311 if (!dst_hold_safe(dst))
314 skb->_skb_refdst = (unsigned long)dst;
315 skb->slow_gro |= !!dst;
318 return skb->_skb_refdst != 0UL;
323 * __skb_tunnel_rx - prepare skb for rx reinsert
325 * @dev: tunnel device
326 * @net: netns for packet i/o
328 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
329 * so make some cleanups. (no accounting done)
331 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
337 * Clear hash so that we can recalulate the hash for the
338 * encapsulated packet, unless we have already determine the hash
339 * over the L4 4-tuple.
341 skb_clear_hash_if_not_l4(skb);
342 skb_set_queue_mapping(skb, 0);
343 skb_scrub_packet(skb, !net_eq(net, dev_net(dev)));
347 * skb_tunnel_rx - prepare skb for rx reinsert
349 * @dev: tunnel device
350 * @net: netns for packet i/o
352 * After decapsulation, packet is going to re-enter (netif_rx()) our stack,
353 * so make some cleanups, and perform accounting.
354 * Note: this accounting is not SMP safe.
356 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
359 /* TODO : stats should be SMP safe */
360 dev->stats.rx_packets++;
361 dev->stats.rx_bytes += skb->len;
362 __skb_tunnel_rx(skb, dev, net);
365 static inline u32 dst_tclassid(const struct sk_buff *skb)
367 #ifdef CONFIG_IP_ROUTE_CLASSID
368 const struct dst_entry *dst;
372 return dst->tclassid;
377 int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
378 static inline int dst_discard(struct sk_buff *skb)
380 return dst_discard_out(&init_net, skb->sk, skb);
382 void *dst_alloc(struct dst_ops *ops, struct net_device *dev, int initial_ref,
383 int initial_obsolete, unsigned short flags);
384 void dst_init(struct dst_entry *dst, struct dst_ops *ops,
385 struct net_device *dev, int initial_ref, int initial_obsolete,
386 unsigned short flags);
387 struct dst_entry *dst_destroy(struct dst_entry *dst);
388 void dst_dev_put(struct dst_entry *dst);
390 static inline void dst_confirm(struct dst_entry *dst)
394 static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
396 struct neighbour *n = dst->ops->neigh_lookup(dst, NULL, daddr);
397 return IS_ERR(n) ? NULL : n;
400 static inline struct neighbour *dst_neigh_lookup_skb(const struct dst_entry *dst,
405 if (WARN_ON_ONCE(!dst->ops->neigh_lookup))
408 n = dst->ops->neigh_lookup(dst, skb, NULL);
410 return IS_ERR(n) ? NULL : n;
413 static inline void dst_confirm_neigh(const struct dst_entry *dst,
416 if (dst->ops->confirm_neigh)
417 dst->ops->confirm_neigh(dst, daddr);
420 static inline void dst_link_failure(struct sk_buff *skb)
422 struct dst_entry *dst = skb_dst(skb);
423 if (dst && dst->ops && dst->ops->link_failure)
424 dst->ops->link_failure(skb);
427 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
429 unsigned long expires = jiffies + timeout;
434 if (dst->expires == 0 || time_before(expires, dst->expires))
435 dst->expires = expires;
438 INDIRECT_CALLABLE_DECLARE(int ip6_output(struct net *, struct sock *,
440 INDIRECT_CALLABLE_DECLARE(int ip_output(struct net *, struct sock *,
442 /* Output packet to network from transport. */
443 static inline int dst_output(struct net *net, struct sock *sk, struct sk_buff *skb)
445 return INDIRECT_CALL_INET(skb_dst(skb)->output,
446 ip6_output, ip_output,
450 INDIRECT_CALLABLE_DECLARE(int ip6_input(struct sk_buff *));
451 INDIRECT_CALLABLE_DECLARE(int ip_local_deliver(struct sk_buff *));
452 /* Input packet from network to transport. */
453 static inline int dst_input(struct sk_buff *skb)
455 return INDIRECT_CALL_INET(skb_dst(skb)->input,
456 ip6_input, ip_local_deliver, skb);
459 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
461 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
463 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
466 dst = INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check,
467 ipv4_dst_check, dst, cookie);
471 /* Flags for xfrm_lookup flags argument. */
473 XFRM_LOOKUP_ICMP = 1 << 0,
474 XFRM_LOOKUP_QUEUE = 1 << 1,
475 XFRM_LOOKUP_KEEP_DST_REF = 1 << 2,
480 static inline struct dst_entry *xfrm_lookup(struct net *net,
481 struct dst_entry *dst_orig,
482 const struct flowi *fl,
483 const struct sock *sk,
489 static inline struct dst_entry *
490 xfrm_lookup_with_ifid(struct net *net, struct dst_entry *dst_orig,
491 const struct flowi *fl, const struct sock *sk,
492 int flags, u32 if_id)
497 static inline struct dst_entry *xfrm_lookup_route(struct net *net,
498 struct dst_entry *dst_orig,
499 const struct flowi *fl,
500 const struct sock *sk,
506 static inline struct xfrm_state *dst_xfrm(const struct dst_entry *dst)
512 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
513 const struct flowi *fl, const struct sock *sk,
516 struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
517 struct dst_entry *dst_orig,
518 const struct flowi *fl,
519 const struct sock *sk, int flags,
522 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
523 const struct flowi *fl, const struct sock *sk,
526 /* skb attached with this dst needs transformation if dst->xfrm is valid */
527 static inline struct xfrm_state *dst_xfrm(const struct dst_entry *dst)
533 static inline void skb_dst_update_pmtu(struct sk_buff *skb, u32 mtu)
535 struct dst_entry *dst = skb_dst(skb);
537 if (dst && dst->ops->update_pmtu)
538 dst->ops->update_pmtu(dst, NULL, skb, mtu, true);
541 /* update dst pmtu but not do neighbor confirm */
542 static inline void skb_dst_update_pmtu_no_confirm(struct sk_buff *skb, u32 mtu)
544 struct dst_entry *dst = skb_dst(skb);
546 if (dst && dst->ops->update_pmtu)
547 dst->ops->update_pmtu(dst, NULL, skb, mtu, false);
550 struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie);
551 void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
552 struct sk_buff *skb, u32 mtu, bool confirm_neigh);
553 void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
554 struct sk_buff *skb);
555 u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old);
556 struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
559 unsigned int dst_blackhole_mtu(const struct dst_entry *dst);
561 #endif /* _NET_DST_H */