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         struct neighbour __rcu  *_neighbour;
  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 static inline struct neighbour *dst_get_neighbour_noref(struct dst_entry *dst)
 100 {
 101         return rcu_dereference(dst->_neighbour);
 102 }
 103
 104 static inline struct neighbour *dst_get_neighbour_noref_raw(struct dst_entry *dst)
 105 {
 106         return rcu_dereference_raw(dst->_neighbour);
 107 }
 108
 109 static inline void dst_set_neighbour(struct dst_entry *dst, struct neighbour *neigh)
 110 {
 111         rcu_assign_pointer(dst->_neighbour, neigh);
 112 }
 113
 114 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
 115 extern const u32 dst_default_metrics[RTAX_MAX];
 116
 117 #define DST_METRICS_READ_ONLY   0x1UL
 118 #define __DST_METRICS_PTR(Y)    \
 119         ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
 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 extern 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 void set_dst_metric_rtt(struct dst_entry *dst, int metric,
 228                                       unsigned long rtt)
 229 {
 230         dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
 231 }
 232
 233 static inline u32
 234 dst_allfrag(const struct dst_entry *dst)
 235 {
 236         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 237         return ret;
 238 }
 239
 240 static inline int
 241 dst_metric_locked(const struct dst_entry *dst, int metric)
 242 {
 243         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 244 }
 245
 246 static inline void dst_hold(struct dst_entry *dst)
 247 {
 248         /*
 249          * If your kernel compilation stops here, please check
 250          * __pad_to_align_refcnt declaration in struct dst_entry
 251          */
 252         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 253         atomic_inc(&dst->__refcnt);
 254 }
 255
 256 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 257 {
 258         dst_hold(dst);
 259         dst->__use++;
 260         dst->lastuse = time;
 261 }
 262
 263 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 264 {
 265         dst->__use++;
 266         dst->lastuse = time;
 267 }
 268
 269 static inline struct dst_entry *dst_clone(struct dst_entry *dst)
 270 {
 271         if (dst)
 272                 atomic_inc(&dst->__refcnt);
 273         return dst;
 274 }
 275
 276 extern void dst_release(struct dst_entry *dst);
 277
 278 static inline void refdst_drop(unsigned long refdst)
 279 {
 280         if (!(refdst & SKB_DST_NOREF))
 281                 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
 282 }
 283
 284 /**
 285  * skb_dst_drop - drops skb dst
 286  * @skb: buffer
 287  *
 288  * Drops dst reference count if a reference was taken.
 289  */
 290 static inline void skb_dst_drop(struct sk_buff *skb)
 291 {
 292         if (skb->_skb_refdst) {
 293                 refdst_drop(skb->_skb_refdst);
 294                 skb->_skb_refdst = 0UL;
 295         }
 296 }
 297
 298 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 299 {
 300         nskb->_skb_refdst = oskb->_skb_refdst;
 301         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 302                 dst_clone(skb_dst(nskb));
 303 }
 304
 305 /**
 306  * skb_dst_force - makes sure skb dst is refcounted
 307  * @skb: buffer
 308  *
 309  * If dst is not yet refcounted, let's do it
 310  */
 311 static inline void skb_dst_force(struct sk_buff *skb)
 312 {
 313         if (skb_dst_is_noref(skb)) {
 314                 WARN_ON(!rcu_read_lock_held());
 315                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 316                 dst_clone(skb_dst(skb));
 317         }
 318 }
 319
 320
 321 /**
 322  *      __skb_tunnel_rx - prepare skb for rx reinsert
 323  *      @skb: buffer
 324  *      @dev: tunnel device
 325  *
 326  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 327  *      so make some cleanups. (no accounting done)
 328  */
 329 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 330 {
 331         skb->dev = dev;
 332
 333         /*
 334          * Clear rxhash so that we can recalulate the hash for the
 335          * encapsulated packet, unless we have already determine the hash
 336          * over the L4 4-tuple.
 337          */
 338         if (!skb->l4_rxhash)
 339                 skb->rxhash = 0;
 340         skb_set_queue_mapping(skb, 0);
 341         skb_dst_drop(skb);
 342         nf_reset(skb);
 343 }
 344
 345 /**
 346  *      skb_tunnel_rx - prepare skb for rx reinsert
 347  *      @skb: buffer
 348  *      @dev: tunnel device
 349  *
 350  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 351  *      so make some cleanups, and perform accounting.
 352  *      Note: this accounting is not SMP safe.
 353  */
 354 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 355 {
 356         /* TODO : stats should be SMP safe */
 357         dev->stats.rx_packets++;
 358         dev->stats.rx_bytes += skb->len;
 359         __skb_tunnel_rx(skb, dev);
 360 }
 361
 362 /* Children define the path of the packet through the
 363  * Linux networking.  Thus, destinations are stackable.
 364  */
 365
 366 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
 367 {
 368         struct dst_entry *child = dst_clone(skb_dst(skb)->child);
 369
 370         skb_dst_drop(skb);
 371         return child;
 372 }
 373
 374 extern int dst_discard(struct sk_buff *skb);
 375 extern void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
 376                        int initial_ref, int initial_obsolete,
 377                        unsigned short flags);
 378 extern void __dst_free(struct dst_entry *dst);
 379 extern struct dst_entry *dst_destroy(struct dst_entry *dst);
 380
 381 static inline void dst_free(struct dst_entry *dst)
 382 {
 383         if (dst->obsolete > 1)
 384                 return;
 385         if (!atomic_read(&dst->__refcnt)) {
 386                 dst = dst_destroy(dst);
 387                 if (!dst)
 388                         return;
 389         }
 390         __dst_free(dst);
 391 }
 392
 393 static inline void dst_rcu_free(struct rcu_head *head)
 394 {
 395         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 396         dst_free(dst);
 397 }
 398
 399 static inline void dst_confirm(struct dst_entry *dst)
 400 {
 401         dst->pending_confirm = 1;
 402 }
 403
 404 static inline int dst_neigh_output(struct dst_entry *dst, struct neighbour *n,
 405                                    struct sk_buff *skb)
 406 {
 407         struct hh_cache *hh;
 408
 409         if (unlikely(dst->pending_confirm)) {
 410                 n->confirmed = jiffies;
 411                 dst->pending_confirm = 0;
 412         }
 413
 414         hh = &n->hh;
 415         if ((n->nud_state & NUD_CONNECTED) && hh->hh_len)
 416                 return neigh_hh_output(hh, skb);
 417         else
 418                 return n->output(n, skb);
 419 }
 420
 421 static inline struct neighbour *dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
 422 {
 423         return dst->ops->neigh_lookup(dst, daddr);
 424 }
 425
 426 static inline void dst_link_failure(struct sk_buff *skb)
 427 {
 428         struct dst_entry *dst = skb_dst(skb);
 429         if (dst && dst->ops && dst->ops->link_failure)
 430                 dst->ops->link_failure(skb);
 431 }
 432
 433 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 434 {
 435         unsigned long expires = jiffies + timeout;
 436
 437         if (expires == 0)
 438                 expires = 1;
 439
 440         if (dst->expires == 0 || time_before(expires, dst->expires))
 441                 dst->expires = expires;
 442 }
 443
 444 /* Output packet to network from transport.  */
 445 static inline int dst_output(struct sk_buff *skb)
 446 {
 447         return skb_dst(skb)->output(skb);
 448 }
 449
 450 /* Input packet from network to transport.  */
 451 static inline int dst_input(struct sk_buff *skb)
 452 {
 453         return skb_dst(skb)->input(skb);
 454 }
 455
 456 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 457 {
 458         if (dst->obsolete)
 459                 dst = dst->ops->check(dst, cookie);
 460         return dst;
 461 }
 462
 463 extern void             dst_init(void);
 464
 465 /* Flags for xfrm_lookup flags argument. */
 466 enum {
 467         XFRM_LOOKUP_ICMP = 1 << 0,
 468 };
 469
 470 struct flowi;
 471 #ifndef CONFIG_XFRM
 472 static inline struct dst_entry *xfrm_lookup(struct net *net,
 473                                             struct dst_entry *dst_orig,
 474                                             const struct flowi *fl, struct sock *sk,
 475                                             int flags)
 476 {
 477         return dst_orig;
 478 }
 479 #else
 480 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 481                                      const struct flowi *fl, struct sock *sk,
 482                                      int flags);
 483 #endif
 484
 485 #endif /* _NET_DST_H */