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         unsigned long           expires;
  40         struct dst_entry        *path;
  41         struct dst_entry        *from;
  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 sock *sk, struct sk_buff *skb);
  49
  50         unsigned short          flags;
  51 #define DST_HOST                0x0001
  52 #define DST_NOXFRM              0x0002
  53 #define DST_NOPOLICY            0x0004
  54 #define DST_NOHASH              0x0008
  55 #define DST_NOCACHE             0x0010
  56 #define DST_NOCOUNT             0x0020
  57 #define DST_FAKE_RTABLE         0x0040
  58 #define DST_XFRM_TUNNEL         0x0080
  59 #define DST_XFRM_QUEUE          0x0100
  60 #define DST_METADATA            0x0200
  61
  62         unsigned short          pending_confirm;
  63
  64         short                   error;
  65
  66         /* A non-zero value of dst->obsolete forces by-hand validation
  67          * of the route entry.  Positive values are set by the generic
  68          * dst layer to indicate that the entry has been forcefully
  69          * destroyed.
  70          *
  71          * Negative values are used by the implementation layer code to
  72          * force invocation of the dst_ops->check() method.
  73          */
  74         short                   obsolete;
  75 #define DST_OBSOLETE_NONE       0
  76 #define DST_OBSOLETE_DEAD       2
  77 #define DST_OBSOLETE_FORCE_CHK  -1
  78 #define DST_OBSOLETE_KILL       -2
  79         unsigned short          header_len;     /* more space at head required */
  80         unsigned short          trailer_len;    /* space to reserve at tail */
  81 #ifdef CONFIG_IP_ROUTE_CLASSID
  82         __u32                   tclassid;
  83 #else
  84         __u32                   __pad2;
  85 #endif
  86
  87 #ifdef CONFIG_64BIT
  88         struct lwtunnel_state   *lwtstate;
  89         /*
  90          * Align __refcnt to a 64 bytes alignment
  91          * (L1_CACHE_SIZE would be too much)
  92          */
  93         long                    __pad_to_align_refcnt[1];
  94 #endif
  95         /*
  96          * __refcnt wants to be on a different cache line from
  97          * input/output/ops or performance tanks badly
  98          */
  99         atomic_t                __refcnt;       /* client references    */
 100         int                     __use;
 101         unsigned long           lastuse;
 102 #ifndef CONFIG_64BIT
 103         struct lwtunnel_state   *lwtstate;
 104 #endif
 105         union {
 106                 struct dst_entry        *next;
 107                 struct rtable __rcu     *rt_next;
 108                 struct rt6_info         *rt6_next;
 109                 struct dn_route __rcu   *dn_next;
 110         };
 111 };
 112
 113 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
 114 extern const u32 dst_default_metrics[];
 115
 116 #define DST_METRICS_READ_ONLY           0x1UL
 117 #define DST_METRICS_FLAGS               0x3UL
 118 #define __DST_METRICS_PTR(Y)    \
 119         ((u32 *)((Y) & ~DST_METRICS_FLAGS))
 120 #define DST_METRICS_PTR(X)      __DST_METRICS_PTR((X)->_metrics)
 121
 122 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
 123 {
 124         return dst->_metrics & DST_METRICS_READ_ONLY;
 125 }
 126
 127 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
 128
 129 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
 130 {
 131         unsigned long val = dst->_metrics;
 132         if (!(val & DST_METRICS_READ_ONLY))
 133                 __dst_destroy_metrics_generic(dst, val);
 134 }
 135
 136 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
 137 {
 138         unsigned long p = dst->_metrics;
 139
 140         BUG_ON(!p);
 141
 142         if (p & DST_METRICS_READ_ONLY)
 143                 return dst->ops->cow_metrics(dst, p);
 144         return __DST_METRICS_PTR(p);
 145 }
 146
 147 /* This may only be invoked before the entry has reached global
 148  * visibility.
 149  */
 150 static inline void dst_init_metrics(struct dst_entry *dst,
 151                                     const u32 *src_metrics,
 152                                     bool read_only)
 153 {
 154         dst->_metrics = ((unsigned long) src_metrics) |
 155                 (read_only ? DST_METRICS_READ_ONLY : 0);
 156 }
 157
 158 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
 159 {
 160         u32 *dst_metrics = dst_metrics_write_ptr(dest);
 161
 162         if (dst_metrics) {
 163                 u32 *src_metrics = DST_METRICS_PTR(src);
 164
 165                 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
 166         }
 167 }
 168
 169 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
 170 {
 171         return DST_METRICS_PTR(dst);
 172 }
 173
 174 static inline u32
 175 dst_metric_raw(const struct dst_entry *dst, const int metric)
 176 {
 177         u32 *p = DST_METRICS_PTR(dst);
 178
 179         return p[metric-1];
 180 }
 181
 182 static inline u32
 183 dst_metric(const struct dst_entry *dst, const int metric)
 184 {
 185         WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
 186                      metric == RTAX_ADVMSS ||
 187                      metric == RTAX_MTU);
 188         return dst_metric_raw(dst, metric);
 189 }
 190
 191 static inline u32
 192 dst_metric_advmss(const struct dst_entry *dst)
 193 {
 194         u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
 195
 196         if (!advmss)
 197                 advmss = dst->ops->default_advmss(dst);
 198
 199         return advmss;
 200 }
 201
 202 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 203 {
 204         u32 *p = dst_metrics_write_ptr(dst);
 205
 206         if (p)
 207                 p[metric-1] = val;
 208 }
 209
 210 static inline u32
 211 dst_feature(const struct dst_entry *dst, u32 feature)
 212 {
 213         return dst_metric(dst, RTAX_FEATURES) & feature;
 214 }
 215
 216 static inline u32 dst_mtu(const struct dst_entry *dst)
 217 {
 218         return dst->ops->mtu(dst);
 219 }
 220
 221 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
 222 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
 223 {
 224         return msecs_to_jiffies(dst_metric(dst, metric));
 225 }
 226
 227 static inline u32
 228 dst_allfrag(const struct dst_entry *dst)
 229 {
 230         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 231         return ret;
 232 }
 233
 234 static inline int
 235 dst_metric_locked(const struct dst_entry *dst, int metric)
 236 {
 237         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 238 }
 239
 240 static inline void dst_hold(struct dst_entry *dst)
 241 {
 242         /*
 243          * If your kernel compilation stops here, please check
 244          * __pad_to_align_refcnt declaration in struct dst_entry
 245          */
 246         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 247         atomic_inc(&dst->__refcnt);
 248 }
 249
 250 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 251 {
 252         dst_hold(dst);
 253         dst->__use++;
 254         dst->lastuse = time;
 255 }
 256
 257 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 258 {
 259         dst->__use++;
 260         dst->lastuse = time;
 261 }
 262
 263 static inline struct dst_entry *dst_clone(struct dst_entry *dst)
 264 {
 265         if (dst)
 266                 atomic_inc(&dst->__refcnt);
 267         return dst;
 268 }
 269
 270 void dst_release(struct dst_entry *dst);
 271
 272 static inline void refdst_drop(unsigned long refdst)
 273 {
 274         if (!(refdst & SKB_DST_NOREF))
 275                 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
 276 }
 277
 278 /**
 279  * skb_dst_drop - drops skb dst
 280  * @skb: buffer
 281  *
 282  * Drops dst reference count if a reference was taken.
 283  */
 284 static inline void skb_dst_drop(struct sk_buff *skb)
 285 {
 286         if (skb->_skb_refdst) {
 287                 refdst_drop(skb->_skb_refdst);
 288                 skb->_skb_refdst = 0UL;
 289         }
 290 }
 291
 292 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 293 {
 294         nskb->_skb_refdst = oskb->_skb_refdst;
 295         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 296                 dst_clone(skb_dst(nskb));
 297 }
 298
 299 /**
 300  * skb_dst_force - makes sure skb dst is refcounted
 301  * @skb: buffer
 302  *
 303  * If dst is not yet refcounted, let's do it
 304  */
 305 static inline void skb_dst_force(struct sk_buff *skb)
 306 {
 307         if (skb_dst_is_noref(skb)) {
 308                 WARN_ON(!rcu_read_lock_held());
 309                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 310                 dst_clone(skb_dst(skb));
 311         }
 312 }
 313
 314
 315 /**
 316  *      __skb_tunnel_rx - prepare skb for rx reinsert
 317  *      @skb: buffer
 318  *      @dev: tunnel device
 319  *      @net: netns for packet i/o
 320  *
 321  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 322  *      so make some cleanups. (no accounting done)
 323  */
 324 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
 325                                    struct net *net)
 326 {
 327         skb->dev = dev;
 328
 329         /*
 330          * Clear hash so that we can recalulate the hash for the
 331          * encapsulated packet, unless we have already determine the hash
 332          * over the L4 4-tuple.
 333          */
 334         skb_clear_hash_if_not_l4(skb);
 335         skb_set_queue_mapping(skb, 0);
 336         skb_scrub_packet(skb, !net_eq(net, dev_net(dev)));
 337 }
 338
 339 /**
 340  *      skb_tunnel_rx - prepare skb for rx reinsert
 341  *      @skb: buffer
 342  *      @dev: tunnel device
 343  *
 344  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 345  *      so make some cleanups, and perform accounting.
 346  *      Note: this accounting is not SMP safe.
 347  */
 348 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev,
 349                                  struct net *net)
 350 {
 351         /* TODO : stats should be SMP safe */
 352         dev->stats.rx_packets++;
 353         dev->stats.rx_bytes += skb->len;
 354         __skb_tunnel_rx(skb, dev, net);
 355 }
 356
 357 int dst_discard_sk(struct sock *sk, struct sk_buff *skb);
 358 static inline int dst_discard(struct sk_buff *skb)
 359 {
 360         return dst_discard_sk(skb->sk, skb);
 361 }
 362 void *dst_alloc(struct dst_ops *ops, struct net_device *dev, int initial_ref,
 363                 int initial_obsolete, unsigned short flags);
 364 void dst_init(struct dst_entry *dst, struct dst_ops *ops,
 365               struct net_device *dev, int initial_ref, int initial_obsolete,
 366               unsigned short flags);
 367 void __dst_free(struct dst_entry *dst);
 368 struct dst_entry *dst_destroy(struct dst_entry *dst);
 369
 370 static inline void dst_free(struct dst_entry *dst)
 371 {
 372         if (dst->obsolete > 0)
 373                 return;
 374         if (!atomic_read(&dst->__refcnt)) {
 375                 dst = dst_destroy(dst);
 376                 if (!dst)
 377                         return;
 378         }
 379         __dst_free(dst);
 380 }
 381
 382 static inline void dst_rcu_free(struct rcu_head *head)
 383 {
 384         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 385         dst_free(dst);
 386 }
 387
 388 static inline void dst_confirm(struct dst_entry *dst)
 389 {
 390         dst->pending_confirm = 1;
 391 }
 392
 393 static inline int dst_neigh_output(struct dst_entry *dst, struct neighbour *n,
 394                                    struct sk_buff *skb)
 395 {
 396         const struct hh_cache *hh;
 397
 398         if (dst->pending_confirm) {
 399                 unsigned long now = jiffies;
 400
 401                 dst->pending_confirm = 0;
 402                 /* avoid dirtying neighbour */
 403                 if (n->confirmed != now)
 404                         n->confirmed = now;
 405         }
 406
 407         hh = &n->hh;
 408         if ((n->nud_state & NUD_CONNECTED) && hh->hh_len)
 409                 return neigh_hh_output(hh, skb);
 410         else
 411                 return n->output(n, skb);
 412 }
 413
 414 static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
 415 {
 416         struct neighbour *n = dst->ops->neigh_lookup(dst, NULL, daddr);
 417         return IS_ERR(n) ? NULL : n;
 418 }
 419
 420 static inline struct neighbour *dst_neigh_lookup_skb(const struct dst_entry *dst,
 421                                                      struct sk_buff *skb)
 422 {
 423         struct neighbour *n =  dst->ops->neigh_lookup(dst, skb, NULL);
 424         return IS_ERR(n) ? NULL : n;
 425 }
 426
 427 static inline void dst_link_failure(struct sk_buff *skb)
 428 {
 429         struct dst_entry *dst = skb_dst(skb);
 430         if (dst && dst->ops && dst->ops->link_failure)
 431                 dst->ops->link_failure(skb);
 432 }
 433
 434 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 435 {
 436         unsigned long expires = jiffies + timeout;
 437
 438         if (expires == 0)
 439                 expires = 1;
 440
 441         if (dst->expires == 0 || time_before(expires, dst->expires))
 442                 dst->expires = expires;
 443 }
 444
 445 /* Output packet to network from transport.  */
 446 static inline int dst_output_sk(struct sock *sk, struct sk_buff *skb)
 447 {
 448         return skb_dst(skb)->output(sk, skb);
 449 }
 450 static inline int dst_output(struct sk_buff *skb)
 451 {
 452         return dst_output_sk(skb->sk, skb);
 453 }
 454
 455 /* Input packet from network to transport.  */
 456 static inline int dst_input(struct sk_buff *skb)
 457 {
 458         return skb_dst(skb)->input(skb);
 459 }
 460
 461 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 462 {
 463         if (dst->obsolete)
 464                 dst = dst->ops->check(dst, cookie);
 465         return dst;
 466 }
 467
 468 void dst_subsys_init(void);
 469
 470 /* Flags for xfrm_lookup flags argument. */
 471 enum {
 472         XFRM_LOOKUP_ICMP = 1 << 0,
 473         XFRM_LOOKUP_QUEUE = 1 << 1,
 474         XFRM_LOOKUP_KEEP_DST_REF = 1 << 2,
 475 };
 476
 477 struct flowi;
 478 #ifndef CONFIG_XFRM
 479 static inline struct dst_entry *xfrm_lookup(struct net *net,
 480                                             struct dst_entry *dst_orig,
 481                                             const struct flowi *fl, struct sock *sk,
 482                                             int flags)
 483 {
 484         return dst_orig;
 485 }
 486
 487 static inline struct dst_entry *xfrm_lookup_route(struct net *net,
 488                                                   struct dst_entry *dst_orig,
 489                                                   const struct flowi *fl,
 490                                                   struct sock *sk,
 491                                                   int flags)
 492 {
 493         return dst_orig;
 494 }
 495
 496 static inline struct xfrm_state *dst_xfrm(const struct dst_entry *dst)
 497 {
 498         return NULL;
 499 }
 500
 501 #else
 502 struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 503                               const struct flowi *fl, struct sock *sk,
 504                               int flags);
 505
 506 struct dst_entry *xfrm_lookup_route(struct net *net, struct dst_entry *dst_orig,
 507                                     const struct flowi *fl, struct sock *sk,
 508                                     int flags);
 509
 510 /* skb attached with this dst needs transformation if dst->xfrm is valid */
 511 static inline struct xfrm_state *dst_xfrm(const struct dst_entry *dst)
 512 {
 513         return dst->xfrm;
 514 }
 515 #endif
 516
 517 #endif /* _NET_DST_H */