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 #define DST_GC_MIN (HZ/10) 20 #define DST_GC_INC (HZ/2) 21 #define DST_GC_MAX (120*HZ) 22 23 /* Each dst_entry has reference count and sits in some parent list(s). 24 * When it is removed from parent list, it is "freed" (dst_free). 25 * After this it enters dead state (dst->obsolete > 0) and if its refcnt 26 * is zero, it can be destroyed immediately, otherwise it is added 27 * to gc list and garbage collector periodically checks the refcnt. 28 */ 29 30 struct sk_buff; 31 32 struct dst_entry { 33 struct rcu_head rcu_head; 34 struct dst_entry *child; 35 struct net_device *dev; 36 struct dst_ops *ops; 37 unsigned long _metrics; 38 unsigned long expires; 39 struct dst_entry *path; 40 struct neighbour *neighbour; 41 struct hh_cache *hh; 42 #ifdef CONFIG_XFRM 43 struct xfrm_state *xfrm; 44 #else 45 void *__pad1; 46 #endif 47 int (*input)(struct sk_buff*); 48 int (*output)(struct sk_buff*); 49 50 short error; 51 short obsolete; 52 unsigned short header_len; /* more space at head required */ 53 unsigned short trailer_len; /* space to reserve at tail */ 54 #ifdef CONFIG_IP_ROUTE_CLASSID 55 __u32 tclassid; 56 #else 57 __u32 __pad2; 58 #endif 59 60 /* 61 * Align __refcnt to a 64 bytes alignment 62 * (L1_CACHE_SIZE would be too much) 63 */ 64 #ifdef CONFIG_64BIT 65 long __pad_to_align_refcnt[1]; 66 #endif 67 /* 68 * __refcnt wants to be on a different cache line from 69 * input/output/ops or performance tanks badly 70 */ 71 atomic_t __refcnt; /* client references */ 72 int __use; 73 unsigned long lastuse; 74 int flags; 75 #define DST_HOST 0x0001 76 #define DST_NOXFRM 0x0002 77 #define DST_NOPOLICY 0x0004 78 #define DST_NOHASH 0x0008 79 #define DST_NOCACHE 0x0010 80 union { 81 struct dst_entry *next; 82 struct rtable __rcu *rt_next; 83 struct rt6_info *rt6_next; 84 struct dn_route __rcu *dn_next; 85 }; 86 }; 87 88 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old); 89 extern const u32 dst_default_metrics[RTAX_MAX]; 90 91 #define DST_METRICS_READ_ONLY 0x1UL 92 #define __DST_METRICS_PTR(Y) \ 93 ((u32 *)((Y) & ~DST_METRICS_READ_ONLY)) 94 #define DST_METRICS_PTR(X) __DST_METRICS_PTR((X)->_metrics) 95 96 static inline bool dst_metrics_read_only(const struct dst_entry *dst) 97 { 98 return dst->_metrics & DST_METRICS_READ_ONLY; 99 } 100 101 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old); 102 103 static inline void dst_destroy_metrics_generic(struct dst_entry *dst) 104 { 105 unsigned long val = dst->_metrics; 106 if (!(val & DST_METRICS_READ_ONLY)) 107 __dst_destroy_metrics_generic(dst, val); 108 } 109 110 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst) 111 { 112 unsigned long p = dst->_metrics; 113 114 if (p & DST_METRICS_READ_ONLY) 115 return dst->ops->cow_metrics(dst, p); 116 return __DST_METRICS_PTR(p); 117 } 118 119 /* This may only be invoked before the entry has reached global 120 * visibility. 121 */ 122 static inline void dst_init_metrics(struct dst_entry *dst, 123 const u32 *src_metrics, 124 bool read_only) 125 { 126 dst->_metrics = ((unsigned long) src_metrics) | 127 (read_only ? DST_METRICS_READ_ONLY : 0); 128 } 129 130 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src) 131 { 132 u32 *dst_metrics = dst_metrics_write_ptr(dest); 133 134 if (dst_metrics) { 135 u32 *src_metrics = DST_METRICS_PTR(src); 136 137 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32)); 138 } 139 } 140 141 static inline u32 *dst_metrics_ptr(struct dst_entry *dst) 142 { 143 return DST_METRICS_PTR(dst); 144 } 145 146 static inline u32 147 dst_metric_raw(const struct dst_entry *dst, const int metric) 148 { 149 u32 *p = DST_METRICS_PTR(dst); 150 151 return p[metric-1]; 152 } 153 154 static inline u32 155 dst_metric(const struct dst_entry *dst, const int metric) 156 { 157 WARN_ON_ONCE(metric == RTAX_HOPLIMIT || 158 metric == RTAX_ADVMSS || 159 metric == RTAX_MTU); 160 return dst_metric_raw(dst, metric); 161 } 162 163 static inline u32 164 dst_metric_advmss(const struct dst_entry *dst) 165 { 166 u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS); 167 168 if (!advmss) 169 advmss = dst->ops->default_advmss(dst); 170 171 return advmss; 172 } 173 174 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val) 175 { 176 u32 *p = dst_metrics_write_ptr(dst); 177 178 if (p) 179 p[metric-1] = val; 180 } 181 182 static inline u32 183 dst_feature(const struct dst_entry *dst, u32 feature) 184 { 185 return dst_metric(dst, RTAX_FEATURES) & feature; 186 } 187 188 static inline u32 dst_mtu(const struct dst_entry *dst) 189 { 190 u32 mtu = dst_metric_raw(dst, RTAX_MTU); 191 192 if (!mtu) 193 mtu = dst->ops->default_mtu(dst); 194 195 return mtu; 196 } 197 198 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */ 199 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric) 200 { 201 return msecs_to_jiffies(dst_metric(dst, metric)); 202 } 203 204 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric, 205 unsigned long rtt) 206 { 207 dst_metric_set(dst, metric, jiffies_to_msecs(rtt)); 208 } 209 210 static inline u32 211 dst_allfrag(const struct dst_entry *dst) 212 { 213 int ret = dst_feature(dst, RTAX_FEATURE_ALLFRAG); 214 return ret; 215 } 216 217 static inline int 218 dst_metric_locked(const struct dst_entry *dst, int metric) 219 { 220 return dst_metric(dst, RTAX_LOCK) & (1<<metric); 221 } 222 223 static inline void dst_hold(struct dst_entry * dst) 224 { 225 /* 226 * If your kernel compilation stops here, please check 227 * __pad_to_align_refcnt declaration in struct dst_entry 228 */ 229 BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63); 230 atomic_inc(&dst->__refcnt); 231 } 232 233 static inline void dst_use(struct dst_entry *dst, unsigned long time) 234 { 235 dst_hold(dst); 236 dst->__use++; 237 dst->lastuse = time; 238 } 239 240 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time) 241 { 242 dst->__use++; 243 dst->lastuse = time; 244 } 245 246 static inline 247 struct dst_entry * dst_clone(struct dst_entry * dst) 248 { 249 if (dst) 250 atomic_inc(&dst->__refcnt); 251 return dst; 252 } 253 254 extern void dst_release(struct dst_entry *dst); 255 256 static inline void refdst_drop(unsigned long refdst) 257 { 258 if (!(refdst & SKB_DST_NOREF)) 259 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK)); 260 } 261 262 /** 263 * skb_dst_drop - drops skb dst 264 * @skb: buffer 265 * 266 * Drops dst reference count if a reference was taken. 267 */ 268 static inline void skb_dst_drop(struct sk_buff *skb) 269 { 270 if (skb->_skb_refdst) { 271 refdst_drop(skb->_skb_refdst); 272 skb->_skb_refdst = 0UL; 273 } 274 } 275 276 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb) 277 { 278 nskb->_skb_refdst = oskb->_skb_refdst; 279 if (!(nskb->_skb_refdst & SKB_DST_NOREF)) 280 dst_clone(skb_dst(nskb)); 281 } 282 283 /** 284 * skb_dst_force - makes sure skb dst is refcounted 285 * @skb: buffer 286 * 287 * If dst is not yet refcounted, let's do it 288 */ 289 static inline void skb_dst_force(struct sk_buff *skb) 290 { 291 if (skb_dst_is_noref(skb)) { 292 WARN_ON(!rcu_read_lock_held()); 293 skb->_skb_refdst &= ~SKB_DST_NOREF; 294 dst_clone(skb_dst(skb)); 295 } 296 } 297 298 299 /** 300 * __skb_tunnel_rx - prepare skb for rx reinsert 301 * @skb: buffer 302 * @dev: tunnel device 303 * 304 * After decapsulation, packet is going to re-enter (netif_rx()) our stack, 305 * so make some cleanups. (no accounting done) 306 */ 307 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev) 308 { 309 skb->dev = dev; 310 skb->rxhash = 0; 311 skb_set_queue_mapping(skb, 0); 312 skb_dst_drop(skb); 313 nf_reset(skb); 314 } 315 316 /** 317 * skb_tunnel_rx - prepare skb for rx reinsert 318 * @skb: buffer 319 * @dev: tunnel device 320 * 321 * After decapsulation, packet is going to re-enter (netif_rx()) our stack, 322 * so make some cleanups, and perform accounting. 323 * Note: this accounting is not SMP safe. 324 */ 325 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev) 326 { 327 /* TODO : stats should be SMP safe */ 328 dev->stats.rx_packets++; 329 dev->stats.rx_bytes += skb->len; 330 __skb_tunnel_rx(skb, dev); 331 } 332 333 /* Children define the path of the packet through the 334 * Linux networking. Thus, destinations are stackable. 335 */ 336 337 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb) 338 { 339 struct dst_entry *child = dst_clone(skb_dst(skb)->child); 340 341 skb_dst_drop(skb); 342 return child; 343 } 344 345 extern int dst_discard(struct sk_buff *skb); 346 extern void *dst_alloc(struct dst_ops * ops, struct net_device *dev, 347 int initial_ref, int initial_obsolete, int flags); 348 extern void __dst_free(struct dst_entry * dst); 349 extern struct dst_entry *dst_destroy(struct dst_entry * dst); 350 351 static inline void dst_free(struct dst_entry * dst) 352 { 353 if (dst->obsolete > 1) 354 return; 355 if (!atomic_read(&dst->__refcnt)) { 356 dst = dst_destroy(dst); 357 if (!dst) 358 return; 359 } 360 __dst_free(dst); 361 } 362 363 static inline void dst_rcu_free(struct rcu_head *head) 364 { 365 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head); 366 dst_free(dst); 367 } 368 369 static inline void dst_confirm(struct dst_entry *dst) 370 { 371 if (dst) 372 neigh_confirm(dst->neighbour); 373 } 374 375 static inline void dst_link_failure(struct sk_buff *skb) 376 { 377 struct dst_entry *dst = skb_dst(skb); 378 if (dst && dst->ops && dst->ops->link_failure) 379 dst->ops->link_failure(skb); 380 } 381 382 static inline void dst_set_expires(struct dst_entry *dst, int timeout) 383 { 384 unsigned long expires = jiffies + timeout; 385 386 if (expires == 0) 387 expires = 1; 388 389 if (dst->expires == 0 || time_before(expires, dst->expires)) 390 dst->expires = expires; 391 } 392 393 /* Output packet to network from transport. */ 394 static inline int dst_output(struct sk_buff *skb) 395 { 396 return skb_dst(skb)->output(skb); 397 } 398 399 /* Input packet from network to transport. */ 400 static inline int dst_input(struct sk_buff *skb) 401 { 402 return skb_dst(skb)->input(skb); 403 } 404 405 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie) 406 { 407 if (dst->obsolete) 408 dst = dst->ops->check(dst, cookie); 409 return dst; 410 } 411 412 extern void dst_init(void); 413 414 /* Flags for xfrm_lookup flags argument. */ 415 enum { 416 XFRM_LOOKUP_ICMP = 1 << 0, 417 }; 418 419 struct flowi; 420 #ifndef CONFIG_XFRM 421 static inline struct dst_entry *xfrm_lookup(struct net *net, 422 struct dst_entry *dst_orig, 423 const struct flowi *fl, struct sock *sk, 424 int flags) 425 { 426 return dst_orig; 427 } 428 #else 429 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig, 430 const struct flowi *fl, struct sock *sk, 431 int flags); 432 #endif 433 434 #endif /* _NET_DST_H */ 435