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 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
  96 extern const u32 dst_default_metrics[RTAX_MAX];
  97
  98 #define DST_METRICS_READ_ONLY   0x1UL
  99 #define __DST_METRICS_PTR(Y)    \
 100         ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
 101 #define DST_METRICS_PTR(X)      __DST_METRICS_PTR((X)->_metrics)
 102
 103 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
 104 {
 105         return dst->_metrics & DST_METRICS_READ_ONLY;
 106 }
 107
 108 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
 109
 110 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
 111 {
 112         unsigned long val = dst->_metrics;
 113         if (!(val & DST_METRICS_READ_ONLY))
 114                 __dst_destroy_metrics_generic(dst, val);
 115 }
 116
 117 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
 118 {
 119         unsigned long p = dst->_metrics;
 120
 121         if (p & DST_METRICS_READ_ONLY)
 122                 return dst->ops->cow_metrics(dst, p);
 123         return __DST_METRICS_PTR(p);
 124 }
 125
 126 /* This may only be invoked before the entry has reached global
 127  * visibility.
 128  */
 129 static inline void dst_init_metrics(struct dst_entry *dst,
 130                                     const u32 *src_metrics,
 131                                     bool read_only)
 132 {
 133         dst->_metrics = ((unsigned long) src_metrics) |
 134                 (read_only ? DST_METRICS_READ_ONLY : 0);
 135 }
 136
 137 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
 138 {
 139         u32 *dst_metrics = dst_metrics_write_ptr(dest);
 140
 141         if (dst_metrics) {
 142                 u32 *src_metrics = DST_METRICS_PTR(src);
 143
 144                 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
 145         }
 146 }
 147
 148 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
 149 {
 150         return DST_METRICS_PTR(dst);
 151 }
 152
 153 static inline u32
 154 dst_metric_raw(const struct dst_entry *dst, const int metric)
 155 {
 156         u32 *p = DST_METRICS_PTR(dst);
 157
 158         return p[metric-1];
 159 }
 160
 161 static inline u32
 162 dst_metric(const struct dst_entry *dst, const int metric)
 163 {
 164         WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
 165                      metric == RTAX_ADVMSS ||
 166                      metric == RTAX_MTU);
 167         return dst_metric_raw(dst, metric);
 168 }
 169
 170 static inline u32
 171 dst_metric_advmss(const struct dst_entry *dst)
 172 {
 173         u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
 174
 175         if (!advmss)
 176                 advmss = dst->ops->default_advmss(dst);
 177
 178         return advmss;
 179 }
 180
 181 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 182 {
 183         u32 *p = dst_metrics_write_ptr(dst);
 184
 185         if (p)
 186                 p[metric-1] = val;
 187 }
 188
 189 static inline u32
 190 dst_feature(const struct dst_entry *dst, u32 feature)
 191 {
 192         return dst_metric(dst, RTAX_FEATURES) & feature;
 193 }
 194
 195 static inline u32 dst_mtu(const struct dst_entry *dst)
 196 {
 197         u32 mtu = dst_metric_raw(dst, RTAX_MTU);
 198
 199         if (!mtu)
 200                 mtu = dst->ops->default_mtu(dst);
 201
 202         return mtu;
 203 }
 204
 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)
 207 {
 208         return msecs_to_jiffies(dst_metric(dst, metric));
 209 }
 210
 211 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
 212                                       unsigned long rtt)
 213 {
 214         dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
 215 }
 216
 217 static inline u32
 218 dst_allfrag(const struct dst_entry *dst)
 219 {
 220         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 221         return ret;
 222 }
 223
 224 static inline int
 225 dst_metric_locked(const struct dst_entry *dst, int metric)
 226 {
 227         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 228 }
 229
 230 static inline void dst_hold(struct dst_entry * dst)
 231 {
 232         /*
 233          * If your kernel compilation stops here, please check
 234          * __pad_to_align_refcnt declaration in struct dst_entry
 235          */
 236         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 237         atomic_inc(&dst->__refcnt);
 238 }
 239
 240 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 241 {
 242         dst_hold(dst);
 243         dst->__use++;
 244         dst->lastuse = time;
 245 }
 246
 247 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 248 {
 249         dst->__use++;
 250         dst->lastuse = time;
 251 }
 252
 253 static inline
 254 struct dst_entry * dst_clone(struct dst_entry * dst)
 255 {
 256         if (dst)
 257                 atomic_inc(&dst->__refcnt);
 258         return dst;
 259 }
 260
 261 extern void dst_release(struct dst_entry *dst);
 262
 263 static 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  */
 275 static 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
 283 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 284 {
 285         nskb->_skb_refdst = oskb->_skb_refdst;
 286         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 287                 dst_clone(skb_dst(nskb));
 288 }
 289
 290 /**
 291  * skb_dst_force - makes sure skb dst is refcounted
 292  * @skb: buffer
 293  *
 294  * If dst is not yet refcounted, let's do it
 295  */
 296 static inline void skb_dst_force(struct sk_buff *skb)
 297 {
 298         if (skb_dst_is_noref(skb)) {
 299                 WARN_ON(!rcu_read_lock_held());
 300                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 301                 dst_clone(skb_dst(skb));
 302         }
 303 }
 304
 305
 306 /**
 307  *      __skb_tunnel_rx - prepare skb for rx reinsert
 308  *      @skb: buffer
 309  *      @dev: tunnel device
 310  *
 311  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 312  *      so make some cleanups. (no accounting done)
 313  */
 314 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 315 {
 316         skb->dev = dev;
 317         skb->rxhash = 0;
 318         skb_set_queue_mapping(skb, 0);
 319         skb_dst_drop(skb);
 320         nf_reset(skb);
 321 }
 322
 323 /**
 324  *      skb_tunnel_rx - prepare skb for rx reinsert
 325  *      @skb: buffer
 326  *      @dev: tunnel device
 327  *
 328  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 329  *      so make some cleanups, and perform accounting.
 330  *      Note: this accounting is not SMP safe.
 331  */
 332 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 333 {
 334         /* TODO : stats should be SMP safe */
 335         dev->stats.rx_packets++;
 336         dev->stats.rx_bytes += skb->len;
 337         __skb_tunnel_rx(skb, dev);
 338 }
 339
 340 /* Children define the path of the packet through the
 341  * Linux networking.  Thus, destinations are stackable.
 342  */
 343
 344 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
 345 {
 346         struct dst_entry *child = dst_clone(skb_dst(skb)->child);
 347
 348         skb_dst_drop(skb);
 349         return child;
 350 }
 351
 352 extern int dst_discard(struct sk_buff *skb);
 353 extern void *dst_alloc(struct dst_ops * ops, int initial_ref);
 354 extern void __dst_free(struct dst_entry * dst);
 355 extern struct dst_entry *dst_destroy(struct dst_entry * dst);
 356
 357 static inline void dst_free(struct dst_entry * dst)
 358 {
 359         if (dst->obsolete > 1)
 360                 return;
 361         if (!atomic_read(&dst->__refcnt)) {
 362                 dst = dst_destroy(dst);
 363                 if (!dst)
 364                         return;
 365         }
 366         __dst_free(dst);
 367 }
 368
 369 static inline void dst_rcu_free(struct rcu_head *head)
 370 {
 371         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 372         dst_free(dst);
 373 }
 374
 375 static inline void dst_confirm(struct dst_entry *dst)
 376 {
 377         if (dst)
 378                 neigh_confirm(dst->neighbour);
 379 }
 380
 381 static inline void dst_link_failure(struct sk_buff *skb)
 382 {
 383         struct dst_entry *dst = skb_dst(skb);
 384         if (dst && dst->ops && dst->ops->link_failure)
 385                 dst->ops->link_failure(skb);
 386 }
 387
 388 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 389 {
 390         unsigned long expires = jiffies + timeout;
 391
 392         if (expires == 0)
 393                 expires = 1;
 394
 395         if (dst->expires == 0 || time_before(expires, dst->expires))
 396                 dst->expires = expires;
 397 }
 398
 399 /* Output packet to network from transport.  */
 400 static inline int dst_output(struct sk_buff *skb)
 401 {
 402         return skb_dst(skb)->output(skb);
 403 }
 404
 405 /* Input packet from network to transport.  */
 406 static inline int dst_input(struct sk_buff *skb)
 407 {
 408         return skb_dst(skb)->input(skb);
 409 }
 410
 411 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 412 {
 413         if (dst->obsolete)
 414                 dst = dst->ops->check(dst, cookie);
 415         return dst;
 416 }
 417
 418 extern void             dst_init(void);
 419
 420 /* Flags for xfrm_lookup flags argument. */
 421 enum {
 422         XFRM_LOOKUP_ICMP = 1 << 0,
 423 };
 424
 425 struct flowi;
 426 #ifndef CONFIG_XFRM
 427 static inline struct dst_entry *xfrm_lookup(struct net *net,
 428                                             struct dst_entry *dst_orig,
 429                                             const struct flowi *fl, struct sock *sk,
 430                                             int flags)
 431 {
 432         return dst_orig;
 433 }
 434 #else
 435 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 436                                      const struct flowi *fl, struct sock *sk,
 437                                      int flags);
 438 #endif
 439
 440 #endif /* _NET_DST_H */