include/net/dst.h

   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         struct  dst_ops         *ops;
  44         unsigned long           _metrics;
  45         unsigned long           expires;
  46         struct dst_entry        *path;
  47         struct neighbour        *neighbour;
  48         struct hh_cache         *hh;
  49 #ifdef CONFIG_XFRM
  50         struct xfrm_state       *xfrm;
  51 #else
  52         void                    *__pad1;
  53 #endif
  54         int                     (*input)(struct sk_buff*);
  55         int                     (*output)(struct sk_buff*);
  56
  57         short                   error;
  58         short                   obsolete;
  59         unsigned short          header_len;     /* more space at head required */
  60         unsigned short          trailer_len;    /* space to reserve at tail */
  61 #ifdef CONFIG_IP_ROUTE_CLASSID
  62         __u32                   tclassid;
  63 #else
  64         __u32                   __pad2;
  65 #endif
  66
  67         /*
  68          * Align __refcnt to a 64 bytes alignment
  69          * (L1_CACHE_SIZE would be too much)
  70          */
  71 #ifdef CONFIG_64BIT
  72         long                    __pad_to_align_refcnt[1];
  73 #endif
  74         /*
  75          * __refcnt wants to be on a different cache line from
  76          * input/output/ops or performance tanks badly
  77          */
  78         atomic_t                __refcnt;       /* client references    */
  79         int                     __use;
  80         unsigned long           lastuse;
  81         int                     flags;
  82 #define DST_HOST                0x0001
  83 #define DST_NOXFRM              0x0002
  84 #define DST_NOPOLICY            0x0004
  85 #define DST_NOHASH              0x0008
  86 #define DST_NOCACHE             0x0010
  87         union {
  88                 struct dst_entry        *next;
  89                 struct rtable __rcu     *rt_next;
  90                 struct rt6_info         *rt6_next;
  91                 struct dn_route __rcu   *dn_next;
  92         };
  93 };
  94
  95 #ifdef __KERNEL__
  96
  97 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
  98 extern const u32 dst_default_metrics[RTAX_MAX];
  99
 100 #define DST_METRICS_READ_ONLY   0x1UL
 101 #define __DST_METRICS_PTR(Y)    \
 102         ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
 103 #define DST_METRICS_PTR(X)      __DST_METRICS_PTR((X)->_metrics)
 104
 105 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
 106 {
 107         return dst->_metrics & DST_METRICS_READ_ONLY;
 108 }
 109
 110 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
 111
 112 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
 113 {
 114         unsigned long val = dst->_metrics;
 115         if (!(val & DST_METRICS_READ_ONLY))
 116                 __dst_destroy_metrics_generic(dst, val);
 117 }
 118
 119 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
 120 {
 121         unsigned long p = dst->_metrics;
 122
 123         if (p & DST_METRICS_READ_ONLY)
 124                 return dst->ops->cow_metrics(dst, p);
 125         return __DST_METRICS_PTR(p);
 126 }
 127
 128 /* This may only be invoked before the entry has reached global
 129  * visibility.
 130  */
 131 static inline void dst_init_metrics(struct dst_entry *dst,
 132                                     const u32 *src_metrics,
 133                                     bool read_only)
 134 {
 135         dst->_metrics = ((unsigned long) src_metrics) |
 136                 (read_only ? DST_METRICS_READ_ONLY : 0);
 137 }
 138
 139 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
 140 {
 141         u32 *dst_metrics = dst_metrics_write_ptr(dest);
 142
 143         if (dst_metrics) {
 144                 u32 *src_metrics = DST_METRICS_PTR(src);
 145
 146                 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
 147         }
 148 }
 149
 150 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
 151 {
 152         return DST_METRICS_PTR(dst);
 153 }
 154
 155 static inline u32
 156 dst_metric_raw(const struct dst_entry *dst, const int metric)
 157 {
 158         u32 *p = DST_METRICS_PTR(dst);
 159
 160         return p[metric-1];
 161 }
 162
 163 static inline u32
 164 dst_metric(const struct dst_entry *dst, const int metric)
 165 {
 166         WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
 167                      metric == RTAX_ADVMSS ||
 168                      metric == RTAX_MTU);
 169         return dst_metric_raw(dst, metric);
 170 }
 171
 172 static inline u32
 173 dst_metric_advmss(const struct dst_entry *dst)
 174 {
 175         u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
 176
 177         if (!advmss)
 178                 advmss = dst->ops->default_advmss(dst);
 179
 180         return advmss;
 181 }
 182
 183 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 184 {
 185         u32 *p = dst_metrics_write_ptr(dst);
 186
 187         if (p)
 188                 p[metric-1] = val;
 189 }
 190
 191 static inline u32
 192 dst_feature(const struct dst_entry *dst, u32 feature)
 193 {
 194         return dst_metric(dst, RTAX_FEATURES) & feature;
 195 }
 196
 197 static inline u32 dst_mtu(const struct dst_entry *dst)
 198 {
 199         u32 mtu = dst_metric_raw(dst, RTAX_MTU);
 200
 201         if (!mtu)
 202                 mtu = dst->ops->default_mtu(dst);
 203
 204         return mtu;
 205 }
 206
 207 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
 208 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
 209 {
 210         return msecs_to_jiffies(dst_metric(dst, metric));
 211 }
 212
 213 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
 214                                       unsigned long rtt)
 215 {
 216         dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
 217 }
 218
 219 static inline u32
 220 dst_allfrag(const struct dst_entry *dst)
 221 {
 222         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 223         return ret;
 224 }
 225
 226 static inline int
 227 dst_metric_locked(const struct dst_entry *dst, int metric)
 228 {
 229         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 230 }
 231
 232 static inline void dst_hold(struct dst_entry * dst)
 233 {
 234         /*
 235          * If your kernel compilation stops here, please check
 236          * __pad_to_align_refcnt declaration in struct dst_entry
 237          */
 238         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 239         atomic_inc(&dst->__refcnt);
 240 }
 241
 242 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 243 {
 244         dst_hold(dst);
 245         dst->__use++;
 246         dst->lastuse = time;
 247 }
 248
 249 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 250 {
 251         dst->__use++;
 252         dst->lastuse = time;
 253 }
 254
 255 static inline
 256 struct dst_entry * dst_clone(struct dst_entry * dst)
 257 {
 258         if (dst)
 259                 atomic_inc(&dst->__refcnt);
 260         return dst;
 261 }
 262
 263 extern void dst_release(struct dst_entry *dst);
 264
 265 static inline void refdst_drop(unsigned long refdst)
 266 {
 267         if (!(refdst & SKB_DST_NOREF))
 268                 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
 269 }
 270
 271 /**
 272  * skb_dst_drop - drops skb dst
 273  * @skb: buffer
 274  *
 275  * Drops dst reference count if a reference was taken.
 276  */
 277 static inline void skb_dst_drop(struct sk_buff *skb)
 278 {
 279         if (skb->_skb_refdst) {
 280                 refdst_drop(skb->_skb_refdst);
 281                 skb->_skb_refdst = 0UL;
 282         }
 283 }
 284
 285 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 286 {
 287         nskb->_skb_refdst = oskb->_skb_refdst;
 288         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 289                 dst_clone(skb_dst(nskb));
 290 }
 291
 292 /**
 293  * skb_dst_force - makes sure skb dst is refcounted
 294  * @skb: buffer
 295  *
 296  * If dst is not yet refcounted, let's do it
 297  */
 298 static inline void skb_dst_force(struct sk_buff *skb)
 299 {
 300         if (skb_dst_is_noref(skb)) {
 301                 WARN_ON(!rcu_read_lock_held());
 302                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 303                 dst_clone(skb_dst(skb));
 304         }
 305 }
 306
 307
 308 /**
 309  *      __skb_tunnel_rx - prepare skb for rx reinsert
 310  *      @skb: buffer
 311  *      @dev: tunnel device
 312  *
 313  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 314  *      so make some cleanups. (no accounting done)
 315  */
 316 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 317 {
 318         skb->dev = dev;
 319         skb->rxhash = 0;
 320         skb_set_queue_mapping(skb, 0);
 321         skb_dst_drop(skb);
 322         nf_reset(skb);
 323 }
 324
 325 /**
 326  *      skb_tunnel_rx - prepare skb for rx reinsert
 327  *      @skb: buffer
 328  *      @dev: tunnel device
 329  *
 330  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 331  *      so make some cleanups, and perform accounting.
 332  *      Note: this accounting is not SMP safe.
 333  */
 334 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 335 {
 336         /* TODO : stats should be SMP safe */
 337         dev->stats.rx_packets++;
 338         dev->stats.rx_bytes += skb->len;
 339         __skb_tunnel_rx(skb, dev);
 340 }
 341
 342 /* Children define the path of the packet through the
 343  * Linux networking.  Thus, destinations are stackable.
 344  */
 345
 346 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
 347 {
 348         struct dst_entry *child = skb_dst(skb)->child;
 349
 350         skb_dst_drop(skb);
 351         return child;
 352 }
 353
 354 extern int dst_discard(struct sk_buff *skb);
 355 extern void *dst_alloc(struct dst_ops * ops, int initial_ref);
 356 extern void __dst_free(struct dst_entry * dst);
 357 extern struct dst_entry *dst_destroy(struct dst_entry * dst);
 358
 359 static inline void dst_free(struct dst_entry * dst)
 360 {
 361         if (dst->obsolete > 1)
 362                 return;
 363         if (!atomic_read(&dst->__refcnt)) {
 364                 dst = dst_destroy(dst);
 365                 if (!dst)
 366                         return;
 367         }
 368         __dst_free(dst);
 369 }
 370
 371 static inline void dst_rcu_free(struct rcu_head *head)
 372 {
 373         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 374         dst_free(dst);
 375 }
 376
 377 static inline void dst_confirm(struct dst_entry *dst)
 378 {
 379         if (dst)
 380                 neigh_confirm(dst->neighbour);
 381 }
 382
 383 static inline void dst_link_failure(struct sk_buff *skb)
 384 {
 385         struct dst_entry *dst = skb_dst(skb);
 386         if (dst && dst->ops && dst->ops->link_failure)
 387                 dst->ops->link_failure(skb);
 388 }
 389
 390 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 391 {
 392         unsigned long expires = jiffies + timeout;
 393
 394         if (expires == 0)
 395                 expires = 1;
 396
 397         if (dst->expires == 0 || time_before(expires, dst->expires))
 398                 dst->expires = expires;
 399 }
 400
 401 /* Output packet to network from transport.  */
 402 static inline int dst_output(struct sk_buff *skb)
 403 {
 404         return skb_dst(skb)->output(skb);
 405 }
 406
 407 /* Input packet from network to transport.  */
 408 static inline int dst_input(struct sk_buff *skb)
 409 {
 410         return skb_dst(skb)->input(skb);
 411 }
 412
 413 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 414 {
 415         if (dst->obsolete)
 416                 dst = dst->ops->check(dst, cookie);
 417         return dst;
 418 }
 419
 420 extern void             dst_init(void);
 421
 422 /* Flags for xfrm_lookup flags argument. */
 423 enum {
 424         XFRM_LOOKUP_ICMP = 1 << 0,
 425 };
 426
 427 struct flowi;
 428 #ifndef CONFIG_XFRM
 429 static inline struct dst_entry *xfrm_lookup(struct net *net,
 430                                             struct dst_entry *dst_orig,
 431                                             const struct flowi *fl, struct sock *sk,
 432                                             int flags)
 433 {
 434         return dst_orig;
 435 }
 436 #else
 437 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 438                                      const struct flowi *fl, struct sock *sk,
 439                                      int flags);
 440 #endif
 441 #endif
 442
 443 #endif /* _NET_DST_H */