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/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
  31 struct sk_buff;
  32
  33 struct 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         void                    *__pad0;
  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         unsigned short          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         unsigned short          pending_confirm;
  66
  67         short                   error;
  68         short                   obsolete;
  69         unsigned short          header_len;     /* more space at head required */
  70         unsigned short          trailer_len;    /* space to reserve at tail */
  71 #ifdef CONFIG_IP_ROUTE_CLASSID
  72         __u32                   tclassid;
  73 #else
  74         __u32                   __pad2;
  75 #endif
  76
  77         /*
  78          * Align __refcnt to a 64 bytes alignment
  79          * (L1_CACHE_SIZE would be too much)
  80          */
  81 #ifdef CONFIG_64BIT
  82         long                    __pad_to_align_refcnt[2];
  83 #endif
  84         /*
  85          * __refcnt wants to be on a different cache line from
  86          * input/output/ops or performance tanks badly
  87          */
  88         atomic_t                __refcnt;       /* client references    */
  89         int                     __use;
  90         unsigned long           lastuse;
  91         union {
  92                 struct dst_entry        *next;
  93                 struct rtable __rcu     *rt_next;
  94                 struct rt6_info         *rt6_next;
  95                 struct dn_route __rcu   *dn_next;
  96         };
  97 };
  98
  99 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
 100 extern const u32 dst_default_metrics[RTAX_MAX];
 101
 102 #define DST_METRICS_READ_ONLY   0x1UL
 103 #define __DST_METRICS_PTR(Y)    \
 104         ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
 105 #define DST_METRICS_PTR(X)      __DST_METRICS_PTR((X)->_metrics)
 106
 107 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
 108 {
 109         return dst->_metrics & DST_METRICS_READ_ONLY;
 110 }
 111
 112 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
 113
 114 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
 115 {
 116         unsigned long val = dst->_metrics;
 117         if (!(val & DST_METRICS_READ_ONLY))
 118                 __dst_destroy_metrics_generic(dst, val);
 119 }
 120
 121 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
 122 {
 123         unsigned long p = dst->_metrics;
 124
 125         BUG_ON(!p);
 126
 127         if (p & DST_METRICS_READ_ONLY)
 128                 return dst->ops->cow_metrics(dst, p);
 129         return __DST_METRICS_PTR(p);
 130 }
 131
 132 /* This may only be invoked before the entry has reached global
 133  * visibility.
 134  */
 135 static inline void dst_init_metrics(struct dst_entry *dst,
 136                                     const u32 *src_metrics,
 137                                     bool read_only)
 138 {
 139         dst->_metrics = ((unsigned long) src_metrics) |
 140                 (read_only ? DST_METRICS_READ_ONLY : 0);
 141 }
 142
 143 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
 144 {
 145         u32 *dst_metrics = dst_metrics_write_ptr(dest);
 146
 147         if (dst_metrics) {
 148                 u32 *src_metrics = DST_METRICS_PTR(src);
 149
 150                 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
 151         }
 152 }
 153
 154 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
 155 {
 156         return DST_METRICS_PTR(dst);
 157 }
 158
 159 static inline u32
 160 dst_metric_raw(const struct dst_entry *dst, const int metric)
 161 {
 162         u32 *p = DST_METRICS_PTR(dst);
 163
 164         return p[metric-1];
 165 }
 166
 167 static inline u32
 168 dst_metric(const struct dst_entry *dst, const int metric)
 169 {
 170         WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
 171                      metric == RTAX_ADVMSS ||
 172                      metric == RTAX_MTU);
 173         return dst_metric_raw(dst, metric);
 174 }
 175
 176 static inline u32
 177 dst_metric_advmss(const struct dst_entry *dst)
 178 {
 179         u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
 180
 181         if (!advmss)
 182                 advmss = dst->ops->default_advmss(dst);
 183
 184         return advmss;
 185 }
 186
 187 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 188 {
 189         u32 *p = dst_metrics_write_ptr(dst);
 190
 191         if (p)
 192                 p[metric-1] = val;
 193 }
 194
 195 static inline u32
 196 dst_feature(const struct dst_entry *dst, u32 feature)
 197 {
 198         return dst_metric(dst, RTAX_FEATURES) & feature;
 199 }
 200
 201 static inline u32 dst_mtu(const struct dst_entry *dst)
 202 {
 203         return dst->ops->mtu(dst);
 204 }
 205
 206 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
 207 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
 208 {
 209         return msecs_to_jiffies(dst_metric(dst, metric));
 210 }
 211
 212 static inline u32
 213 dst_allfrag(const struct dst_entry *dst)
 214 {
 215         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 216         return ret;
 217 }
 218
 219 static inline int
 220 dst_metric_locked(const struct dst_entry *dst, int metric)
 221 {
 222         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 223 }
 224
 225 static inline void dst_hold(struct dst_entry *dst)
 226 {
 227         /*
 228          * If your kernel compilation stops here, please check
 229          * __pad_to_align_refcnt declaration in struct dst_entry
 230          */
 231         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 232         atomic_inc(&dst->__refcnt);
 233 }
 234
 235 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 236 {
 237         dst_hold(dst);
 238         dst->__use++;
 239         dst->lastuse = time;
 240 }
 241
 242 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 243 {
 244         dst->__use++;
 245         dst->lastuse = time;
 246 }
 247
 248 static inline struct dst_entry *dst_clone(struct dst_entry *dst)
 249 {
 250         if (dst)
 251                 atomic_inc(&dst->__refcnt);
 252         return dst;
 253 }
 254
 255 extern void dst_release(struct dst_entry *dst);
 256
 257 static inline void refdst_drop(unsigned long refdst)
 258 {
 259         if (!(refdst & SKB_DST_NOREF))
 260                 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
 261 }
 262
 263 /**
 264  * skb_dst_drop - drops skb dst
 265  * @skb: buffer
 266  *
 267  * Drops dst reference count if a reference was taken.
 268  */
 269 static inline void skb_dst_drop(struct sk_buff *skb)
 270 {
 271         if (skb->_skb_refdst) {
 272                 refdst_drop(skb->_skb_refdst);
 273                 skb->_skb_refdst = 0UL;
 274         }
 275 }
 276
 277 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 278 {
 279         nskb->_skb_refdst = oskb->_skb_refdst;
 280         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 281                 dst_clone(skb_dst(nskb));
 282 }
 283
 284 /**
 285  * skb_dst_force - makes sure skb dst is refcounted
 286  * @skb: buffer
 287  *
 288  * If dst is not yet refcounted, let's do it
 289  */
 290 static inline void skb_dst_force(struct sk_buff *skb)
 291 {
 292         if (skb_dst_is_noref(skb)) {
 293                 WARN_ON(!rcu_read_lock_held());
 294                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 295                 dst_clone(skb_dst(skb));
 296         }
 297 }
 298
 299
 300 /**
 301  *      __skb_tunnel_rx - prepare skb for rx reinsert
 302  *      @skb: buffer
 303  *      @dev: tunnel device
 304  *
 305  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 306  *      so make some cleanups. (no accounting done)
 307  */
 308 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 309 {
 310         skb->dev = dev;
 311
 312         /*
 313          * Clear rxhash so that we can recalulate the hash for the
 314          * encapsulated packet, unless we have already determine the hash
 315          * over the L4 4-tuple.
 316          */
 317         if (!skb->l4_rxhash)
 318                 skb->rxhash = 0;
 319         skb_set_queue_mapping(skb, 0);
 320         skb_dst_drop(skb);
 321         nf_reset(skb);
 322 }
 323
 324 /**
 325  *      skb_tunnel_rx - prepare skb for rx reinsert
 326  *      @skb: buffer
 327  *      @dev: tunnel device
 328  *
 329  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 330  *      so make some cleanups, and perform accounting.
 331  *      Note: this accounting is not SMP safe.
 332  */
 333 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 334 {
 335         /* TODO : stats should be SMP safe */
 336         dev->stats.rx_packets++;
 337         dev->stats.rx_bytes += skb->len;
 338         __skb_tunnel_rx(skb, dev);
 339 }
 340
 341 /* Children define the path of the packet through the
 342  * Linux networking.  Thus, destinations are stackable.
 343  */
 344
 345 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
 346 {
 347         struct dst_entry *child = dst_clone(skb_dst(skb)->child);
 348
 349         skb_dst_drop(skb);
 350         return child;
 351 }
 352
 353 extern int dst_discard(struct sk_buff *skb);
 354 extern void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
 355                        int initial_ref, int initial_obsolete,
 356                        unsigned short flags);
 357 extern void __dst_free(struct dst_entry *dst);
 358 extern struct dst_entry *dst_destroy(struct dst_entry *dst);
 359
 360 static inline void dst_free(struct dst_entry *dst)
 361 {
 362         if (dst->obsolete > 1)
 363                 return;
 364         if (!atomic_read(&dst->__refcnt)) {
 365                 dst = dst_destroy(dst);
 366                 if (!dst)
 367                         return;
 368         }
 369         __dst_free(dst);
 370 }
 371
 372 static inline void dst_rcu_free(struct rcu_head *head)
 373 {
 374         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 375         dst_free(dst);
 376 }
 377
 378 static inline void dst_confirm(struct dst_entry *dst)
 379 {
 380         dst->pending_confirm = 1;
 381 }
 382
 383 static inline int dst_neigh_output(struct dst_entry *dst, struct neighbour *n,
 384                                    struct sk_buff *skb)
 385 {
 386         struct hh_cache *hh;
 387
 388         if (unlikely(dst->pending_confirm)) {
 389                 n->confirmed = jiffies;
 390                 dst->pending_confirm = 0;
 391         }
 392
 393         hh = &n->hh;
 394         if ((n->nud_state & NUD_CONNECTED) && hh->hh_len)
 395                 return neigh_hh_output(hh, skb);
 396         else
 397                 return n->output(n, skb);
 398 }
 399
 400 static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
 401 {
 402         return dst->ops->neigh_lookup(dst, NULL, daddr);
 403 }
 404
 405 static inline struct neighbour *dst_neigh_lookup_skb(const struct dst_entry *dst,
 406                                                      struct sk_buff *skb)
 407 {
 408         return dst->ops->neigh_lookup(dst, skb, NULL);
 409 }
 410
 411 static inline void dst_link_failure(struct sk_buff *skb)
 412 {
 413         struct dst_entry *dst = skb_dst(skb);
 414         if (dst && dst->ops && dst->ops->link_failure)
 415                 dst->ops->link_failure(skb);
 416 }
 417
 418 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 419 {
 420         unsigned long expires = jiffies + timeout;
 421
 422         if (expires == 0)
 423                 expires = 1;
 424
 425         if (dst->expires == 0 || time_before(expires, dst->expires))
 426                 dst->expires = expires;
 427 }
 428
 429 /* Output packet to network from transport.  */
 430 static inline int dst_output(struct sk_buff *skb)
 431 {
 432         return skb_dst(skb)->output(skb);
 433 }
 434
 435 /* Input packet from network to transport.  */
 436 static inline int dst_input(struct sk_buff *skb)
 437 {
 438         return skb_dst(skb)->input(skb);
 439 }
 440
 441 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 442 {
 443         if (dst->obsolete)
 444                 dst = dst->ops->check(dst, cookie);
 445         return dst;
 446 }
 447
 448 extern void             dst_init(void);
 449
 450 /* Flags for xfrm_lookup flags argument. */
 451 enum {
 452         XFRM_LOOKUP_ICMP = 1 << 0,
 453 };
 454
 455 struct flowi;
 456 #ifndef CONFIG_XFRM
 457 static inline struct dst_entry *xfrm_lookup(struct net *net,
 458                                             struct dst_entry *dst_orig,
 459                                             const struct flowi *fl, struct sock *sk,
 460                                             int flags)
 461 {
 462         return dst_orig;
 463 }
 464 #else
 465 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 466                                      const struct flowi *fl, struct sock *sk,
 467                                      int flags);
 468 #endif
 469
 470 #endif /* _NET_DST_H */