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/bug.h> 16#include <linux/jiffies.h> 17#include <net/neighbour.h> 18#include <asm/processor.h> 19 20#define DST_GC_MIN (HZ/10) 21#define DST_GC_INC (HZ/2) 22#define DST_GC_MAX (120*HZ) 23 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. 29 */ 30 31struct sk_buff; 32 33struct dst_entry { 34 struct rcu_head rcu_head; 35 struct dst_entry *child; 36 struct net_device *dev; 37 struct dst_ops *ops; 38 unsigned long _metrics; 39 union { 40 unsigned long expires; 41 /* point to where the dst_entry copied from */ 42 struct dst_entry *from; 43 }; 44 struct dst_entry *path; 45 struct neighbour __rcu *_neighbour; 46#ifdef CONFIG_XFRM 47 struct xfrm_state *xfrm; 48#else 49 void *__pad1; 50#endif 51 int (*input)(struct sk_buff*); 52 int (*output)(struct sk_buff*); 53 54 int flags; 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 63#define DST_XFRM_TUNNEL 0x0100 64 65 short error; 66 short obsolete; 67 unsigned short header_len; /* more space at head required */ 68 unsigned short trailer_len; /* space to reserve at tail */ 69#ifdef CONFIG_IP_ROUTE_CLASSID 70 __u32 tclassid; 71#else 72 __u32 __pad2; 73#endif 74 75 /* 76 * Align __refcnt to a 64 bytes alignment 77 * (L1_CACHE_SIZE would be too much) 78 */ 79#ifdef CONFIG_64BIT 80 long __pad_to_align_refcnt[2]; 81#endif 82 /* 83 * __refcnt wants to be on a different cache line from 84 * input/output/ops or performance tanks badly 85 */ 86 atomic_t __refcnt; /* client references */ 87 int __use; 88 unsigned long lastuse; 89 union { 90 struct dst_entry *next; 91 struct rtable __rcu *rt_next; 92 struct rt6_info *rt6_next; 93 struct dn_route __rcu *dn_next; 94 }; 95}; 96 97static inline struct neighbour *dst_get_neighbour_noref(struct dst_entry *dst) 98{ 99 return rcu_dereference(dst->_neighbour); 100} 101 102static inline struct neighbour *dst_get_neighbour_noref_raw(struct dst_entry *dst) 103{ 104 return rcu_dereference_raw(dst->_neighbour); 105} 106 107static inline void dst_set_neighbour(struct dst_entry *dst, struct neighbour *neigh) 108{ 109 rcu_assign_pointer(dst->_neighbour, neigh); 110} 111 112extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old); 113extern const u32 dst_default_metrics[RTAX_MAX]; 114 115#define DST_METRICS_READ_ONLY 0x1UL 116#define __DST_METRICS_PTR(Y) \ 117 ((u32 *)((Y) & ~DST_METRICS_READ_ONLY)) 118#define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics) 119 120static inline bool dst_metrics_read_only(const struct dst_entry *dst) 121{ 122 return dst->_metrics & DST_METRICS_READ_ONLY; 123} 124 125extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old); 126 127static inline void dst_destroy_metrics_generic(struct dst_entry *dst) 128{ 129 unsigned long val = dst->_metrics; 130 if (!(val & DST_METRICS_READ_ONLY)) 131 __dst_destroy_metrics_generic(dst, val); 132} 133 134static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst) 135{ 136 unsigned long p = dst->_metrics; 137 138 BUG_ON(!p); 139 140 if (p & DST_METRICS_READ_ONLY) 141 return dst->ops->cow_metrics(dst, p); 142 return __DST_METRICS_PTR(p); 143} 144 145/* This may only be invoked before the entry has reached global 146 * visibility. 147 */ 148static inline void dst_init_metrics(struct dst_entry *dst, 149 const u32 *src_metrics, 150 bool read_only) 151{ 152 dst->_metrics = ((unsigned long) src_metrics) | 153 (read_only ? DST_METRICS_READ_ONLY : 0); 154} 155 156static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src) 157{ 158 u32 *dst_metrics = dst_metrics_write_ptr(dest); 159 160 if (dst_metrics) { 161 u32 *src_metrics = DST_METRICS_PTR(src); 162 163 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32)); 164 } 165} 166 167static inline u32 *dst_metrics_ptr(struct dst_entry *dst) 168{ 169 return DST_METRICS_PTR(dst); 170} 171 172static inline u32 173dst_metric_raw(const struct dst_entry *dst, const int metric) 174{ 175 u32 *p = DST_METRICS_PTR(dst); 176 177 return p[metric-1]; 178} 179 180static inline u32 181dst_metric(const struct dst_entry *dst, const int metric) 182{ 183 WARN_ON_ONCE(metric == RTAX_HOPLIMIT || 184 metric == RTAX_ADVMSS || 185 metric == RTAX_MTU); 186 return dst_metric_raw(dst, metric); 187} 188 189static inline u32 190dst_metric_advmss(const struct dst_entry *dst) 191{ 192 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS); 193 194 if (!advmss) 195 advmss = dst->ops->default_advmss(dst); 196 197 return advmss; 198} 199 200static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val) 201{ 202 u32 *p = dst_metrics_write_ptr(dst); 203 204 if (p) 205 p[metric-1] = val; 206} 207 208static inline u32 209dst_feature(const struct dst_entry *dst, u32 feature) 210{ 211 return dst_metric(dst, RTAX_FEATURES) & feature; 212} 213 214static inline u32 dst_mtu(const struct dst_entry *dst) 215{ 216 return dst->ops->mtu(dst); 217} 218 219/* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */ 220static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric) 221{ 222 return msecs_to_jiffies(dst_metric(dst, metric)); 223} 224 225static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric, 226 unsigned long rtt) 227{ 228 dst_metric_set(dst, metric, jiffies_to_msecs(rtt)); 229} 230 231static inline u32 232dst_allfrag(const struct dst_entry *dst) 233{ 234 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG); 235 return ret; 236} 237 238static inline int 239dst_metric_locked(const struct dst_entry *dst, int metric) 240{ 241 return dst_metric(dst, RTAX_LOCK) & (1<<metric); 242} 243 244static inline void dst_hold(struct dst_entry * dst) 245{ 246 /* 247 * If your kernel compilation stops here, please check 248 * __pad_to_align_refcnt declaration in struct dst_entry 249 */ 250 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63); 251 atomic_inc(&dst->__refcnt); 252} 253 254static inline void dst_use(struct dst_entry *dst, unsigned long time) 255{ 256 dst_hold(dst); 257 dst->__use++; 258 dst->lastuse = time; 259} 260 261static inline void dst_use_noref(struct dst_entry *dst, unsigned long time) 262{ 263 dst->__use++; 264 dst->lastuse = time; 265} 266 267static inline 268struct dst_entry * dst_clone(struct dst_entry * dst) 269{ 270 if (dst) 271 atomic_inc(&dst->__refcnt); 272 return dst; 273} 274 275extern void dst_release(struct dst_entry *dst); 276 277static inline void refdst_drop(unsigned long refdst) 278{ 279 if (!(refdst & SKB_DST_NOREF)) 280 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK)); 281} 282 283/** 284 * skb_dst_drop - drops skb dst 285 * @skb: buffer 286 * 287 * Drops dst reference count if a reference was taken. 288 */ 289static inline void skb_dst_drop(struct sk_buff *skb) 290{ 291 if (skb->_skb_refdst) { 292 refdst_drop(skb->_skb_refdst); 293 skb->_skb_refdst = 0UL; 294 } 295} 296 297static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb) 298{ 299 nskb->_skb_refdst = oskb->_skb_refdst; 300 if (!(nskb->_skb_refdst & SKB_DST_NOREF)) 301 dst_clone(skb_dst(nskb)); 302} 303 304/** 305 * skb_dst_force - makes sure skb dst is refcounted 306 * @skb: buffer 307 * 308 * If dst is not yet refcounted, let's do it 309 */ 310static inline void skb_dst_force(struct sk_buff *skb) 311{ 312 if (skb_dst_is_noref(skb)) { 313 WARN_ON(!rcu_read_lock_held()); 314 skb->_skb_refdst &= ~SKB_DST_NOREF; 315 dst_clone(skb_dst(skb)); 316 } 317} 318 319 320/** 321 * __skb_tunnel_rx - prepare skb for rx reinsert 322 * @skb: buffer 323 * @dev: tunnel device 324 * 325 * After decapsulation, packet is going to re-enter (netif_rx()) our stack, 326 * so make some cleanups. (no accounting done) 327 */ 328static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev) 329{ 330 skb->dev = dev; 331 332 /* 333 * Clear rxhash so that we can recalulate the hash for the 334 * encapsulated packet, unless we have already determine the hash 335 * over the L4 4-tuple. 336 */ 337 if (!skb->l4_rxhash) 338 skb->rxhash = 0; 339 skb_set_queue_mapping(skb, 0); 340 skb_dst_drop(skb); 341 nf_reset(skb); 342} 343 344/** 345 * skb_tunnel_rx - prepare skb for rx reinsert 346 * @skb: buffer 347 * @dev: tunnel device 348 * 349 * After decapsulation, packet is going to re-enter (netif_rx()) our stack, 350 * so make some cleanups, and perform accounting. 351 * Note: this accounting is not SMP safe. 352 */ 353static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev) 354{ 355 /* TODO : stats should be SMP safe */ 356 dev->stats.rx_packets++; 357 dev->stats.rx_bytes += skb->len; 358 __skb_tunnel_rx(skb, dev); 359} 360 361/* Children define the path of the packet through the 362 * Linux networking. Thus, destinations are stackable. 363 */ 364 365static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb) 366{ 367 struct dst_entry *child = dst_clone(skb_dst(skb)->child); 368 369 skb_dst_drop(skb); 370 return child; 371} 372 373extern int dst_discard(struct sk_buff *skb); 374extern void *dst_alloc(struct dst_ops * ops, struct net_device *dev, 375 int initial_ref, int initial_obsolete, int flags); 376extern void __dst_free(struct dst_entry * dst); 377extern struct dst_entry *dst_destroy(struct dst_entry * dst); 378 379static inline void dst_free(struct dst_entry * dst) 380{ 381 if (dst->obsolete > 1) 382 return; 383 if (!atomic_read(&dst->__refcnt)) { 384 dst = dst_destroy(dst); 385 if (!dst) 386 return; 387 } 388 __dst_free(dst); 389} 390 391static inline void dst_rcu_free(struct rcu_head *head) 392{ 393 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head); 394 dst_free(dst); 395} 396 397static inline void dst_confirm(struct dst_entry *dst) 398{ 399 if (dst) { 400 struct neighbour *n; 401 402 rcu_read_lock(); 403 n = dst_get_neighbour_noref(dst); 404 neigh_confirm(n); 405 rcu_read_unlock(); 406 } 407} 408 409static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr) 410{ 411 return dst->ops->neigh_lookup(dst, daddr); 412} 413 414static inline void dst_link_failure(struct sk_buff *skb) 415{ 416 struct dst_entry *dst = skb_dst(skb); 417 if (dst && dst->ops && dst->ops->link_failure) 418 dst->ops->link_failure(skb); 419} 420 421static inline void dst_set_expires(struct dst_entry *dst, int timeout) 422{ 423 unsigned long expires = jiffies + timeout; 424 425 if (expires == 0) 426 expires = 1; 427 428 if (dst->expires == 0 || time_before(expires, dst->expires)) 429 dst->expires = expires; 430} 431 432/* Output packet to network from transport. */ 433static inline int dst_output(struct sk_buff *skb) 434{ 435 return skb_dst(skb)->output(skb); 436} 437 438/* Input packet from network to transport. */ 439static inline int dst_input(struct sk_buff *skb) 440{ 441 return skb_dst(skb)->input(skb); 442} 443 444static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie) 445{ 446 if (dst->obsolete) 447 dst = dst->ops->check(dst, cookie); 448 return dst; 449} 450 451extern void dst_init(void); 452 453/* Flags for xfrm_lookup flags argument. */ 454enum { 455 XFRM_LOOKUP_ICMP = 1 << 0, 456}; 457 458struct flowi; 459#ifndef CONFIG_XFRM 460static inline struct dst_entry *xfrm_lookup(struct net *net, 461 struct dst_entry *dst_orig, 462 const struct flowi *fl, struct sock *sk, 463 int flags) 464{ 465 return dst_orig; 466} 467#else 468extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, 469 const struct flowi *fl, struct sock *sk, 470 int flags); 471#endif 472 473#endif /* _NET_DST_H */ 474