dst.h

  1/* SPDX-License-Identifier: GPL-2.0 */
  2/*
  3 * net/dst.h	Protocol independent destination cache definitions.
  4 *
  5 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  6 *
  7 */
  8
  9#ifndef _NET_DST_H
 10#define _NET_DST_H
 11
 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 <linux/rcuref.h>
 20#include <net/neighbour.h>
 21#include <asm/processor.h>
 22#include <linux/indirect_call_wrapper.h>
 23
 24struct sk_buff;
 25
 26struct dst_entry {
 27	union {
 28		struct net_device       *dev;
 29		struct net_device __rcu *dev_rcu;
 30	};
 31	struct  dst_ops	        *ops;
 32	unsigned long		_metrics;
 33	unsigned long           expires;
 34#ifdef CONFIG_XFRM
 35	struct xfrm_state	*xfrm;
 36#else
 37	void			*__pad1;
 38#endif
 39	int			(*input)(struct sk_buff *);
 40	int			(*output)(struct net *net, struct sock *sk, struct sk_buff *skb);
 41
 42	unsigned short		flags;
 43#define DST_NOXFRM		0x0002
 44#define DST_NOPOLICY		0x0004
 45#define DST_NOCOUNT		0x0008
 46#define DST_FAKE_RTABLE		0x0010
 47#define DST_XFRM_TUNNEL		0x0020
 48#define DST_XFRM_QUEUE		0x0040
 49#define DST_METADATA		0x0080
 50
 51	/* A non-zero value of dst->obsolete forces by-hand validation
 52	 * of the route entry.  Positive values are set by the generic
 53	 * dst layer to indicate that the entry has been forcefully
 54	 * destroyed.
 55	 *
 56	 * Negative values are used by the implementation layer code to
 57	 * force invocation of the dst_ops->check() method.
 58	 */
 59	short			obsolete;
 60#define DST_OBSOLETE_NONE	0
 61#define DST_OBSOLETE_DEAD	2
 62#define DST_OBSOLETE_FORCE_CHK	-1
 63#define DST_OBSOLETE_KILL	-2
 64	unsigned short		header_len;	/* more space at head required */
 65	unsigned short		trailer_len;	/* space to reserve at tail */
 66
 67	/*
 68	 * __rcuref wants to be on a different cache line from
 69	 * input/output/ops or performance tanks badly
 70	 */
 71#ifdef CONFIG_64BIT
 72	rcuref_t		__rcuref;	/* 64-bit offset 64 */
 73#endif
 74	int			__use;
 75	unsigned long		lastuse;
 76	struct rcu_head		rcu_head;
 77	short			error;
 78	short			__pad;
 79	__u32			tclassid;
 80#ifndef CONFIG_64BIT
 81	struct lwtunnel_state   *lwtstate;
 82	rcuref_t		__rcuref;	/* 32-bit offset 64 */
 83#endif
 84	netdevice_tracker	dev_tracker;
 85
 86	/*
 87	 * Used by rtable and rt6_info. Moves lwtstate into the next cache
 88	 * line on 64bit so that lwtstate does not cause false sharing with
 89	 * __rcuref under contention of __rcuref. This also puts the
 90	 * frequently accessed members of rtable and rt6_info out of the
 91	 * __rcuref cache line.
 92	 */
 93	struct list_head	rt_uncached;
 94	struct uncached_list	*rt_uncached_list;
 95#ifdef CONFIG_64BIT
 96	struct lwtunnel_state   *lwtstate;
 97#endif
 98};
 99
100struct dst_metrics {
101	u32		metrics[RTAX_MAX];
102	refcount_t	refcnt;
103} __aligned(4);		/* Low pointer bits contain DST_METRICS_FLAGS */
104extern const struct dst_metrics dst_default_metrics;
105
106u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
107
108#define DST_METRICS_READ_ONLY		0x1UL
109#define DST_METRICS_REFCOUNTED		0x2UL
110#define DST_METRICS_FLAGS		0x3UL
111#define __DST_METRICS_PTR(Y)	\
112	((u32 *)((Y) & ~DST_METRICS_FLAGS))
113#define DST_METRICS_PTR(X)	__DST_METRICS_PTR((X)->_metrics)
114
115static inline bool dst_metrics_read_only(const struct dst_entry *dst)
116{
117	return dst->_metrics & DST_METRICS_READ_ONLY;
118}
119
120void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
121
122static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
123{
124	unsigned long val = dst->_metrics;
125	if (!(val & DST_METRICS_READ_ONLY))
126		__dst_destroy_metrics_generic(dst, val);
127}
128
129static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
130{
131	unsigned long p = dst->_metrics;
132
133	BUG_ON(!p);
134
135	if (p & DST_METRICS_READ_ONLY)
136		return dst->ops->cow_metrics(dst, p);
137	return __DST_METRICS_PTR(p);
138}
139
140/* This may only be invoked before the entry has reached global
141 * visibility.
142 */
143static inline void dst_init_metrics(struct dst_entry *dst,
144				    const u32 *src_metrics,
145				    bool read_only)
146{
147	dst->_metrics = ((unsigned long) src_metrics) |
148		(read_only ? DST_METRICS_READ_ONLY : 0);
149}
150
151static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
152{
153	u32 *dst_metrics = dst_metrics_write_ptr(dest);
154
155	if (dst_metrics) {
156		u32 *src_metrics = DST_METRICS_PTR(src);
157
158		memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
159	}
160}
161
162static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
163{
164	return DST_METRICS_PTR(dst);
165}
166
167static inline u32
168dst_metric_raw(const struct dst_entry *dst, const int metric)
169{
170	u32 *p = DST_METRICS_PTR(dst);
171
172	return p[metric-1];
173}
174
175static inline u32
176dst_metric(const struct dst_entry *dst, const int metric)
177{
178	WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
179		     metric == RTAX_ADVMSS ||
180		     metric == RTAX_MTU);
181	return dst_metric_raw(dst, metric);
182}
183
184static inline u32
185dst_metric_advmss(const struct dst_entry *dst)
186{
187	u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
188
189	if (!advmss)
190		advmss = dst->ops->default_advmss(dst);
191
192	return advmss;
193}
194
195static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
196{
197	u32 *p = dst_metrics_write_ptr(dst);
198
199	if (p)
200		p[metric-1] = val;
201}
202
203/* Kernel-internal feature bits that are unallocated in user space. */
204#define DST_FEATURE_ECN_CA	(1U << 31)
205
206#define DST_FEATURE_MASK	(DST_FEATURE_ECN_CA)
207#define DST_FEATURE_ECN_MASK	(DST_FEATURE_ECN_CA | RTAX_FEATURE_ECN)
208
209static inline u32
210dst_feature(const struct dst_entry *dst, u32 feature)
211{
212	return dst_metric(dst, RTAX_FEATURES) & feature;
213}
214
215INDIRECT_CALLABLE_DECLARE(unsigned int ip6_mtu(const struct dst_entry *));
216INDIRECT_CALLABLE_DECLARE(unsigned int ipv4_mtu(const struct dst_entry *));
217static inline u32 dst_mtu(const struct dst_entry *dst)
218{
219	return INDIRECT_CALL_INET(dst->ops->mtu, ip6_mtu, ipv4_mtu, dst);
220}
221
222/* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
223static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
224{
225	return msecs_to_jiffies(dst_metric(dst, metric));
226}
227
228static inline int
229dst_metric_locked(const struct dst_entry *dst, int metric)
230{
231	return dst_metric(dst, RTAX_LOCK) & (1 << metric);
232}
233
234static inline void dst_hold(struct dst_entry *dst)
235{
236	/*
237	 * If your kernel compilation stops here, please check
238	 * the placement of __rcuref in struct dst_entry
239	 */
240	BUILD_BUG_ON(offsetof(struct dst_entry, __rcuref) & 63);
241	WARN_ON(!rcuref_get(&dst->__rcuref));
242}
243
244static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
245{
246	if (unlikely(time != READ_ONCE(dst->lastuse))) {
247		dst->__use++;
248		WRITE_ONCE(dst->lastuse, time);
249	}
250}
251
252static inline struct dst_entry *dst_clone(struct dst_entry *dst)
253{
254	if (dst)
255		dst_hold(dst);
256	return dst;
257}
258
259void dst_release(struct dst_entry *dst);
260
261void dst_release_immediate(struct dst_entry *dst);
262
263static inline void refdst_drop(unsigned long refdst)
264{
265	if (!(refdst & SKB_DST_NOREF))
266		dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
267}
268
269/**
270 * skb_dst_drop - drops skb dst
271 * @skb: buffer
272 *
273 * Drops dst reference count if a reference was taken.
274 */
275static inline void skb_dst_drop(struct sk_buff *skb)
276{
277	if (skb->_skb_refdst) {
278		refdst_drop(skb->_skb_refdst);
279		skb->_skb_refdst = 0UL;
280	}
281}
282
283static inline void __skb_dst_copy(struct sk_buff *nskb, unsigned long refdst)
284{
285	nskb->slow_gro |= !!refdst;
286	nskb->_skb_refdst = refdst;
287	if (!(nskb->_skb_refdst & SKB_DST_NOREF))
288		dst_clone(skb_dst(nskb));
289}
290
291static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
292{
293	__skb_dst_copy(nskb, oskb->_skb_refdst);
294}
295
296/**
297 * dst_hold_safe - Take a reference on a dst if possible
298 * @dst: pointer to dst entry
299 *
300 * This helper returns false if it could not safely
301 * take a reference on a dst.
302 */
303static inline bool dst_hold_safe(struct dst_entry *dst)
304{
305	return rcuref_get(&dst->__rcuref);
306}
307
308/**
309 * skb_dst_force - makes sure skb dst is refcounted
310 * @skb: buffer
311 *
312 * If dst is not yet refcounted and not destroyed, grab a ref on it.
313 * Returns: true if dst is refcounted.
314 */
315static inline bool skb_dst_force(struct sk_buff *skb)
316{
317	if (skb_dst_is_noref(skb)) {
318		struct dst_entry *dst = skb_dst(skb);
319
320		WARN_ON(!rcu_read_lock_held());
321		if (!dst_hold_safe(dst))
322			dst = NULL;
323
324		skb->_skb_refdst = (unsigned long)dst;
325		skb->slow_gro |= !!dst;
326	}
327
328	return skb->_skb_refdst != 0UL;
329}
330
331
332/**
333 *	__skb_tunnel_rx - prepare skb for rx reinsert
334 *	@skb: buffer
335 *	@dev: tunnel device
336 *	@net: netns for packet i/o
337 *
338 *	After decapsulation, packet is going to re-enter (netif_rx()) our stack,
339 *	so make some cleanups. (no accounting done)
340 */
341static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
342				   struct net *net)
343{
344	skb->dev = dev;
345
346	/*
347	 * Clear hash so that we can recalculate the hash for the
348	 * encapsulated packet, unless we have already determine the hash
349	 * over the L4 4-tuple.
350	 */
351	skb_clear_hash_if_not_l4(skb);
352	skb_set_queue_mapping(skb, 0);
353	skb_scrub_packet(skb, !net_eq(net, dev_net(dev)));
354}
355
356/**
357 *	skb_tunnel_rx - prepare skb for rx reinsert
358 *	@skb: buffer
359 *	@dev: tunnel device
360 *	@net: netns for packet i/o
361 *
362 *	After decapsulation, packet is going to re-enter (netif_rx()) our stack,
363 *	so make some cleanups, and perform accounting.
364 *	Note: this accounting is not SMP safe.
365 */
366static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
367				 struct net *net)
368{
369	DEV_STATS_INC(dev, rx_packets);
370	DEV_STATS_ADD(dev, rx_bytes, skb->len);
371	__skb_tunnel_rx(skb, dev, net);
372}
373
374static inline u32 dst_tclassid(const struct sk_buff *skb)
375{
376#ifdef CONFIG_IP_ROUTE_CLASSID
377	const struct dst_entry *dst;
378
379	dst = skb_dst(skb);
380	if (dst)
381		return dst->tclassid;
382#endif
383	return 0;
384}
385
386int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
387static inline int dst_discard(struct sk_buff *skb)
388{
389	return dst_discard_out(&init_net, skb->sk, skb);
390}
391void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
392		int initial_obsolete, unsigned short flags);
393void dst_init(struct dst_entry *dst, struct dst_ops *ops,
394	      struct net_device *dev, int initial_obsolete,
395	      unsigned short flags);
396void dst_dev_put(struct dst_entry *dst);
397
398static inline void dst_confirm(struct dst_entry *dst)
399{
400}
401
402static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
403{
404	struct neighbour *n = dst->ops->neigh_lookup(dst, NULL, daddr);
405	return IS_ERR(n) ? NULL : n;
406}
407
408static inline struct neighbour *dst_neigh_lookup_skb(const struct dst_entry *dst,
409						     struct sk_buff *skb)
410{
411	struct neighbour *n;
412
413	if (WARN_ON_ONCE(!dst->ops->neigh_lookup))
414		return NULL;
415
416	n = dst->ops->neigh_lookup(dst, skb, NULL);
417
418	return IS_ERR(n) ? NULL : n;
419}
420
421static inline void dst_confirm_neigh(const struct dst_entry *dst,
422				     const void *daddr)
423{
424	if (dst->ops->confirm_neigh)
425		dst->ops->confirm_neigh(dst, daddr);
426}
427
428static inline void dst_link_failure(struct sk_buff *skb)
429{
430	struct dst_entry *dst = skb_dst(skb);
431	if (dst && dst->ops && dst->ops->link_failure)
432		dst->ops->link_failure(skb);
433}
434
435static inline void dst_set_expires(struct dst_entry *dst, int timeout)
436{
437	unsigned long old, expires = jiffies + timeout;
438
439	if (expires == 0)
440		expires = 1;
441
442	old = READ_ONCE(dst->expires);
443
444	if (!old || time_before(expires, old))
445		WRITE_ONCE(dst->expires, expires);
446}
447
448static inline unsigned int dst_dev_overhead(struct dst_entry *dst,
449					    struct sk_buff *skb)
450{
451	if (likely(dst))
452		return LL_RESERVED_SPACE(dst->dev);
453
454	return skb->mac_len;
455}
456
457INDIRECT_CALLABLE_DECLARE(int ip6_output(struct net *, struct sock *,
458					 struct sk_buff *));
459INDIRECT_CALLABLE_DECLARE(int ip_output(struct net *, struct sock *,
460					 struct sk_buff *));
461/* Output packet to network from transport.  */
462static inline int dst_output(struct net *net, struct sock *sk, struct sk_buff *skb)
463{
464	return INDIRECT_CALL_INET(READ_ONCE(skb_dst(skb)->output),
465				  ip6_output, ip_output,
466				  net, sk, skb);
467}
468
469INDIRECT_CALLABLE_DECLARE(int ip6_input(struct sk_buff *));
470INDIRECT_CALLABLE_DECLARE(int ip_local_deliver(struct sk_buff *));
471/* Input packet from network to transport.  */
472static inline int dst_input(struct sk_buff *skb)
473{
474	return INDIRECT_CALL_INET(READ_ONCE(skb_dst(skb)->input),
475				  ip6_input, ip_local_deliver, skb);
476}
477
478INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
479							  u32));
480INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
481							   u32));
482static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
483{
484	if (READ_ONCE(dst->obsolete))
485		dst = INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check,
486					 ipv4_dst_check, dst, cookie);
487	return dst;
488}
489
490/* Flags for xfrm_lookup flags argument. */
491enum {
492	XFRM_LOOKUP_ICMP = 1 << 0,
493	XFRM_LOOKUP_QUEUE = 1 << 1,
494	XFRM_LOOKUP_KEEP_DST_REF = 1 << 2,
495};
496
497struct flowi;
498#ifndef CONFIG_XFRM
499static inline struct dst_entry *xfrm_lookup(struct net *net,
500					    struct dst_entry *dst_orig,
501					    const struct flowi *fl,
502					    const struct sock *sk,
503					    int flags)
504{
505	return dst_orig;
506}
507
508static inline struct dst_entry *
509xfrm_lookup_with_ifid(struct net *net, struct dst_entry *dst_orig,
510		      const struct flowi *fl, const struct sock *sk,
511		      int flags, u32 if_id)
512{
513	return dst_orig;
514}
515
516static inline struct dst_entry *xfrm_lookup_route(struct net *net,
517						  struct dst_entry *dst_orig,
518						  const struct flowi *fl,
519						  const struct sock *sk,
520						  int flags)
521{
522	return dst_orig;
523}
524
525static inline struct xfrm_state *dst_xfrm(const struct dst_entry *dst)
526{
527	return NULL;
528}
529
530#else
531struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
532			      const struct flowi *fl, const struct sock *sk,
533			      int flags);
534
535struct dst_entry *xfrm_lookup_with_ifid(struct net *net,
536					struct dst_entry *dst_orig,
537					const struct flowi *fl,
538					const struct sock *sk, int flags,
539					u32 if_id);
540
541struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
542				    const struct flowi *fl, const struct sock *sk,
543				    int flags);
544
545/* skb attached with this dst needs transformation if dst->xfrm is valid */
546static inline struct xfrm_state *dst_xfrm(const struct dst_entry *dst)
547{
548	return dst->xfrm;
549}
550#endif
551
552static inline void skb_dst_update_pmtu(struct sk_buff *skb, u32 mtu)
553{
554	struct dst_entry *dst = skb_dst(skb);
555
556	if (dst && dst->ops->update_pmtu)
557		dst->ops->update_pmtu(dst, NULL, skb, mtu, true);
558}
559
560/* update dst pmtu but not do neighbor confirm */
561static inline void skb_dst_update_pmtu_no_confirm(struct sk_buff *skb, u32 mtu)
562{
563	struct dst_entry *dst = skb_dst(skb);
564
565	if (dst && dst->ops->update_pmtu)
566		dst->ops->update_pmtu(dst, NULL, skb, mtu, false);
567}
568
569static inline struct net_device *dst_dev(const struct dst_entry *dst)
570{
571	return READ_ONCE(dst->dev);
572}
573
574static inline struct net_device *dst_dev_rcu(const struct dst_entry *dst)
575{
576	return rcu_dereference(dst->dev_rcu);
577}
578
579static inline struct net *dst_dev_net_rcu(const struct dst_entry *dst)
580{
581	return dev_net_rcu(dst_dev_rcu(dst));
582}
583
584static inline struct net_device *skb_dst_dev(const struct sk_buff *skb)
585{
586	return dst_dev(skb_dst(skb));
587}
588
589static inline struct net_device *skb_dst_dev_rcu(const struct sk_buff *skb)
590{
591	return dst_dev_rcu(skb_dst(skb));
592}
593
594static inline struct net *skb_dst_dev_net(const struct sk_buff *skb)
595{
596	return dev_net(skb_dst_dev(skb));
597}
598
599static inline struct net *skb_dst_dev_net_rcu(const struct sk_buff *skb)
600{
601	return dev_net_rcu(skb_dst_dev_rcu(skb));
602}
603
604struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie);
605void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
606			       struct sk_buff *skb, u32 mtu, bool confirm_neigh);
607void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
608			    struct sk_buff *skb);
609u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old);
610struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst,
611					     struct sk_buff *skb,
612					     const void *daddr);
613unsigned int dst_blackhole_mtu(const struct dst_entry *dst);
614
615#endif /* _NET_DST_H */