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 #define DST_GC_MIN      (HZ/10)
  20 #define DST_GC_INC      (HZ/2)
  21 #define DST_GC_MAX      (120*HZ)
  22
  23 /* Each dst_entry has reference count and sits in some parent list(s).
  24  * When it is removed from parent list, it is "freed" (dst_free).
  25  * After this it enters dead state (dst->obsolete > 0) and if its refcnt
  26  * is zero, it can be destroyed immediately, otherwise it is added
  27  * to gc list and garbage collector periodically checks the refcnt.
  28  */
  29
  30 struct sk_buff;
  31
  32 struct dst_entry {
  33         struct rcu_head         rcu_head;
  34         struct dst_entry        *child;
  35         struct net_device       *dev;
  36         struct  dst_ops         *ops;
  37         unsigned long           _metrics;
  38         unsigned long           expires;
  39         struct dst_entry        *path;
  40         struct neighbour __rcu  *_neighbour;
  41         struct hh_cache         *hh;
  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 sk_buff*);
  49
  50         short                   error;
  51         short                   obsolete;
  52         unsigned short          header_len;     /* more space at head required */
  53         unsigned short          trailer_len;    /* space to reserve at tail */
  54 #ifdef CONFIG_IP_ROUTE_CLASSID
  55         __u32                   tclassid;
  56 #else
  57         __u32                   __pad2;
  58 #endif
  59
  60         /*
  61          * Align __refcnt to a 64 bytes alignment
  62          * (L1_CACHE_SIZE would be too much)
  63          */
  64 #ifdef CONFIG_64BIT
  65         long                    __pad_to_align_refcnt[1];
  66 #endif
  67         /*
  68          * __refcnt wants to be on a different cache line from
  69          * input/output/ops or performance tanks badly
  70          */
  71         atomic_t                __refcnt;       /* client references    */
  72         int                     __use;
  73         unsigned long           lastuse;
  74         int                     flags;
  75 #define DST_HOST                0x0001
  76 #define DST_NOXFRM              0x0002
  77 #define DST_NOPOLICY            0x0004
  78 #define DST_NOHASH              0x0008
  79 #define DST_NOCACHE             0x0010
  80 #define DST_NOCOUNT             0x0020
  81 #define DST_XFRM_TUNNEL         0x0100
  82         union {
  83                 struct dst_entry        *next;
  84                 struct rtable __rcu     *rt_next;
  85                 struct rt6_info         *rt6_next;
  86                 struct dn_route __rcu   *dn_next;
  87         };
  88 };
  89
  90 static inline struct neighbour *dst_get_neighbour(struct dst_entry *dst)
  91 {
  92         return rcu_dereference(dst->_neighbour);
  93 }
  94
  95 static inline struct neighbour *dst_get_neighbour_raw(struct dst_entry *dst)
  96 {
  97         return rcu_dereference_raw(dst->_neighbour);
  98 }
  99
 100 static inline void dst_set_neighbour(struct dst_entry *dst, struct neighbour *neigh)
 101 {
 102         rcu_assign_pointer(dst->_neighbour, neigh);
 103 }
 104
 105 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
 106 extern const u32 dst_default_metrics[RTAX_MAX];
 107
 108 #define DST_METRICS_READ_ONLY   0x1UL
 109 #define __DST_METRICS_PTR(Y)    \
 110         ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
 111 #define DST_METRICS_PTR(X)      __DST_METRICS_PTR((X)->_metrics)
 112
 113 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
 114 {
 115         return dst->_metrics & DST_METRICS_READ_ONLY;
 116 }
 117
 118 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
 119
 120 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
 121 {
 122         unsigned long val = dst->_metrics;
 123         if (!(val & DST_METRICS_READ_ONLY))
 124                 __dst_destroy_metrics_generic(dst, val);
 125 }
 126
 127 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
 128 {
 129         unsigned long p = dst->_metrics;
 130
 131         BUG_ON(!p);
 132
 133         if (p & DST_METRICS_READ_ONLY)
 134                 return dst->ops->cow_metrics(dst, p);
 135         return __DST_METRICS_PTR(p);
 136 }
 137
 138 /* This may only be invoked before the entry has reached global
 139  * visibility.
 140  */
 141 static inline void dst_init_metrics(struct dst_entry *dst,
 142                                     const u32 *src_metrics,
 143                                     bool read_only)
 144 {
 145         dst->_metrics = ((unsigned long) src_metrics) |
 146                 (read_only ? DST_METRICS_READ_ONLY : 0);
 147 }
 148
 149 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
 150 {
 151         u32 *dst_metrics = dst_metrics_write_ptr(dest);
 152
 153         if (dst_metrics) {
 154                 u32 *src_metrics = DST_METRICS_PTR(src);
 155
 156                 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
 157         }
 158 }
 159
 160 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
 161 {
 162         return DST_METRICS_PTR(dst);
 163 }
 164
 165 static inline u32
 166 dst_metric_raw(const struct dst_entry *dst, const int metric)
 167 {
 168         u32 *p = DST_METRICS_PTR(dst);
 169
 170         return p[metric-1];
 171 }
 172
 173 static inline u32
 174 dst_metric(const struct dst_entry *dst, const int metric)
 175 {
 176         WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
 177                      metric == RTAX_ADVMSS ||
 178                      metric == RTAX_MTU);
 179         return dst_metric_raw(dst, metric);
 180 }
 181
 182 static inline u32
 183 dst_metric_advmss(const struct dst_entry *dst)
 184 {
 185         u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
 186
 187         if (!advmss)
 188                 advmss = dst->ops->default_advmss(dst);
 189
 190         return advmss;
 191 }
 192
 193 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 194 {
 195         u32 *p = dst_metrics_write_ptr(dst);
 196
 197         if (p)
 198                 p[metric-1] = val;
 199 }
 200
 201 static inline u32
 202 dst_feature(const struct dst_entry *dst, u32 feature)
 203 {
 204         return dst_metric(dst, RTAX_FEATURES) & feature;
 205 }
 206
 207 static inline u32 dst_mtu(const struct dst_entry *dst)
 208 {
 209         u32 mtu = dst_metric_raw(dst, RTAX_MTU);
 210
 211         if (!mtu)
 212                 mtu = dst->ops->default_mtu(dst);
 213
 214         return mtu;
 215 }
 216
 217 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
 218 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
 219 {
 220         return msecs_to_jiffies(dst_metric(dst, metric));
 221 }
 222
 223 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
 224                                       unsigned long rtt)
 225 {
 226         dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
 227 }
 228
 229 static inline u32
 230 dst_allfrag(const struct dst_entry *dst)
 231 {
 232         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 233         return ret;
 234 }
 235
 236 static inline int
 237 dst_metric_locked(const struct dst_entry *dst, int metric)
 238 {
 239         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 240 }
 241
 242 static inline void dst_hold(struct dst_entry * dst)
 243 {
 244         /*
 245          * If your kernel compilation stops here, please check
 246          * __pad_to_align_refcnt declaration in struct dst_entry
 247          */
 248         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 249         atomic_inc(&dst->__refcnt);
 250 }
 251
 252 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 253 {
 254         dst_hold(dst);
 255         dst->__use++;
 256         dst->lastuse = time;
 257 }
 258
 259 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 260 {
 261         dst->__use++;
 262         dst->lastuse = time;
 263 }
 264
 265 static inline
 266 struct dst_entry * dst_clone(struct dst_entry * dst)
 267 {
 268         if (dst)
 269                 atomic_inc(&dst->__refcnt);
 270         return dst;
 271 }
 272
 273 extern void dst_release(struct dst_entry *dst);
 274
 275 static inline void refdst_drop(unsigned long refdst)
 276 {
 277         if (!(refdst & SKB_DST_NOREF))
 278                 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
 279 }
 280
 281 /**
 282  * skb_dst_drop - drops skb dst
 283  * @skb: buffer
 284  *
 285  * Drops dst reference count if a reference was taken.
 286  */
 287 static inline void skb_dst_drop(struct sk_buff *skb)
 288 {
 289         if (skb->_skb_refdst) {
 290                 refdst_drop(skb->_skb_refdst);
 291                 skb->_skb_refdst = 0UL;
 292         }
 293 }
 294
 295 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 296 {
 297         nskb->_skb_refdst = oskb->_skb_refdst;
 298         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 299                 dst_clone(skb_dst(nskb));
 300 }
 301
 302 /**
 303  * skb_dst_force - makes sure skb dst is refcounted
 304  * @skb: buffer
 305  *
 306  * If dst is not yet refcounted, let's do it
 307  */
 308 static inline void skb_dst_force(struct sk_buff *skb)
 309 {
 310         if (skb_dst_is_noref(skb)) {
 311                 WARN_ON(!rcu_read_lock_held());
 312                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 313                 dst_clone(skb_dst(skb));
 314         }
 315 }
 316
 317
 318 /**
 319  *      __skb_tunnel_rx - prepare skb for rx reinsert
 320  *      @skb: buffer
 321  *      @dev: tunnel device
 322  *
 323  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 324  *      so make some cleanups. (no accounting done)
 325  */
 326 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 327 {
 328         skb->dev = dev;
 329         skb->rxhash = 0;
 330         skb_set_queue_mapping(skb, 0);
 331         skb_dst_drop(skb);
 332         nf_reset(skb);
 333 }
 334
 335 /**
 336  *      skb_tunnel_rx - prepare skb for rx reinsert
 337  *      @skb: buffer
 338  *      @dev: tunnel device
 339  *
 340  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 341  *      so make some cleanups, and perform accounting.
 342  *      Note: this accounting is not SMP safe.
 343  */
 344 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 345 {
 346         /* TODO : stats should be SMP safe */
 347         dev->stats.rx_packets++;
 348         dev->stats.rx_bytes += skb->len;
 349         __skb_tunnel_rx(skb, dev);
 350 }
 351
 352 /* Children define the path of the packet through the
 353  * Linux networking.  Thus, destinations are stackable.
 354  */
 355
 356 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
 357 {
 358         struct dst_entry *child = dst_clone(skb_dst(skb)->child);
 359
 360         skb_dst_drop(skb);
 361         return child;
 362 }
 363
 364 extern int dst_discard(struct sk_buff *skb);
 365 extern void *dst_alloc(struct dst_ops * ops, struct net_device *dev,
 366                        int initial_ref, int initial_obsolete, int flags);
 367 extern void __dst_free(struct dst_entry * dst);
 368 extern 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 > 1)
 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         if (dst) {
 391                 struct neighbour *n;
 392
 393                 rcu_read_lock();
 394                 n = dst_get_neighbour(dst);
 395                 neigh_confirm(n);
 396                 rcu_read_unlock();
 397         }
 398 }
 399
 400 static inline void dst_link_failure(struct sk_buff *skb)
 401 {
 402         struct dst_entry *dst = skb_dst(skb);
 403         if (dst && dst->ops && dst->ops->link_failure)
 404                 dst->ops->link_failure(skb);
 405 }
 406
 407 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 408 {
 409         unsigned long expires = jiffies + timeout;
 410
 411         if (expires == 0)
 412                 expires = 1;
 413
 414         if (dst->expires == 0 || time_before(expires, dst->expires))
 415                 dst->expires = expires;
 416 }
 417
 418 /* Output packet to network from transport.  */
 419 static inline int dst_output(struct sk_buff *skb)
 420 {
 421         return skb_dst(skb)->output(skb);
 422 }
 423
 424 /* Input packet from network to transport.  */
 425 static inline int dst_input(struct sk_buff *skb)
 426 {
 427         return skb_dst(skb)->input(skb);
 428 }
 429
 430 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 431 {
 432         if (dst->obsolete)
 433                 dst = dst->ops->check(dst, cookie);
 434         return dst;
 435 }
 436
 437 extern void             dst_init(void);
 438
 439 /* Flags for xfrm_lookup flags argument. */
 440 enum {
 441         XFRM_LOOKUP_ICMP = 1 << 0,
 442 };
 443
 444 struct flowi;
 445 #ifndef CONFIG_XFRM
 446 static inline struct dst_entry *xfrm_lookup(struct net *net,
 447                                             struct dst_entry *dst_orig,
 448                                             const struct flowi *fl, struct sock *sk,
 449                                             int flags)
 450 {
 451         return dst_orig;
 452 }
 453 #else
 454 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 455                                      const struct flowi *fl, struct sock *sk,
 456                                      int flags);
 457 #endif
 458
 459 #endif /* _NET_DST_H */