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        *neighbour;
  41 #ifdef CONFIG_XFRM
  42         struct xfrm_state       *xfrm;
  43 #else
  44         void                    *__pad1;
  45 #endif
  46         int                     (*input)(struct sk_buff*);
  47         int                     (*output)(struct sk_buff*);
  48
  49         int                     flags;
  50 #define DST_HOST                0x0001
  51 #define DST_NOXFRM              0x0002
  52 #define DST_NOPOLICY            0x0004
  53 #define DST_NOHASH              0x0008
  54 #define DST_NOCACHE             0x0010
  55 #define DST_NOCOUNT             0x0020
  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[2];
  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         union {
  82                 struct dst_entry        *next;
  83                 struct rtable __rcu     *rt_next;
  84                 struct rt6_info         *rt6_next;
  85                 struct dn_route __rcu   *dn_next;
  86         };
  87 };
  88
  89 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
  90 extern const u32 dst_default_metrics[RTAX_MAX];
  91
  92 #define DST_METRICS_READ_ONLY   0x1UL
  93 #define __DST_METRICS_PTR(Y)    \
  94         ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
  95 #define DST_METRICS_PTR(X)      __DST_METRICS_PTR((X)->_metrics)
  96
  97 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
  98 {
  99         return dst->_metrics & DST_METRICS_READ_ONLY;
 100 }
 101
 102 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
 103
 104 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
 105 {
 106         unsigned long val = dst->_metrics;
 107         if (!(val & DST_METRICS_READ_ONLY))
 108                 __dst_destroy_metrics_generic(dst, val);
 109 }
 110
 111 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
 112 {
 113         unsigned long p = dst->_metrics;
 114
 115         BUG_ON(!p);
 116
 117         if (p & DST_METRICS_READ_ONLY)
 118                 return dst->ops->cow_metrics(dst, p);
 119         return __DST_METRICS_PTR(p);
 120 }
 121
 122 /* This may only be invoked before the entry has reached global
 123  * visibility.
 124  */
 125 static inline void dst_init_metrics(struct dst_entry *dst,
 126                                     const u32 *src_metrics,
 127                                     bool read_only)
 128 {
 129         dst->_metrics = ((unsigned long) src_metrics) |
 130                 (read_only ? DST_METRICS_READ_ONLY : 0);
 131 }
 132
 133 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
 134 {
 135         u32 *dst_metrics = dst_metrics_write_ptr(dest);
 136
 137         if (dst_metrics) {
 138                 u32 *src_metrics = DST_METRICS_PTR(src);
 139
 140                 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
 141         }
 142 }
 143
 144 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
 145 {
 146         return DST_METRICS_PTR(dst);
 147 }
 148
 149 static inline u32
 150 dst_metric_raw(const struct dst_entry *dst, const int metric)
 151 {
 152         u32 *p = DST_METRICS_PTR(dst);
 153
 154         return p[metric-1];
 155 }
 156
 157 static inline u32
 158 dst_metric(const struct dst_entry *dst, const int metric)
 159 {
 160         WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
 161                      metric == RTAX_ADVMSS ||
 162                      metric == RTAX_MTU);
 163         return dst_metric_raw(dst, metric);
 164 }
 165
 166 static inline u32
 167 dst_metric_advmss(const struct dst_entry *dst)
 168 {
 169         u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
 170
 171         if (!advmss)
 172                 advmss = dst->ops->default_advmss(dst);
 173
 174         return advmss;
 175 }
 176
 177 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 178 {
 179         u32 *p = dst_metrics_write_ptr(dst);
 180
 181         if (p)
 182                 p[metric-1] = val;
 183 }
 184
 185 static inline u32
 186 dst_feature(const struct dst_entry *dst, u32 feature)
 187 {
 188         return dst_metric(dst, RTAX_FEATURES) & feature;
 189 }
 190
 191 static inline u32 dst_mtu(const struct dst_entry *dst)
 192 {
 193         u32 mtu = dst_metric_raw(dst, RTAX_MTU);
 194
 195         if (!mtu)
 196                 mtu = dst->ops->default_mtu(dst);
 197
 198         return mtu;
 199 }
 200
 201 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
 202 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
 203 {
 204         return msecs_to_jiffies(dst_metric(dst, metric));
 205 }
 206
 207 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
 208                                       unsigned long rtt)
 209 {
 210         dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
 211 }
 212
 213 static inline u32
 214 dst_allfrag(const struct dst_entry *dst)
 215 {
 216         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 217         return ret;
 218 }
 219
 220 static inline int
 221 dst_metric_locked(const struct dst_entry *dst, int metric)
 222 {
 223         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 224 }
 225
 226 static inline void dst_hold(struct dst_entry * dst)
 227 {
 228         /*
 229          * If your kernel compilation stops here, please check
 230          * __pad_to_align_refcnt declaration in struct dst_entry
 231          */
 232         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 233         atomic_inc(&dst->__refcnt);
 234 }
 235
 236 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 237 {
 238         dst_hold(dst);
 239         dst->__use++;
 240         dst->lastuse = time;
 241 }
 242
 243 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 244 {
 245         dst->__use++;
 246         dst->lastuse = time;
 247 }
 248
 249 static inline
 250 struct dst_entry * dst_clone(struct dst_entry * dst)
 251 {
 252         if (dst)
 253                 atomic_inc(&dst->__refcnt);
 254         return dst;
 255 }
 256
 257 extern void dst_release(struct dst_entry *dst);
 258
 259 static inline void refdst_drop(unsigned long refdst)
 260 {
 261         if (!(refdst & SKB_DST_NOREF))
 262                 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
 263 }
 264
 265 /**
 266  * skb_dst_drop - drops skb dst
 267  * @skb: buffer
 268  *
 269  * Drops dst reference count if a reference was taken.
 270  */
 271 static inline void skb_dst_drop(struct sk_buff *skb)
 272 {
 273         if (skb->_skb_refdst) {
 274                 refdst_drop(skb->_skb_refdst);
 275                 skb->_skb_refdst = 0UL;
 276         }
 277 }
 278
 279 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 280 {
 281         nskb->_skb_refdst = oskb->_skb_refdst;
 282         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 283                 dst_clone(skb_dst(nskb));
 284 }
 285
 286 /**
 287  * skb_dst_force - makes sure skb dst is refcounted
 288  * @skb: buffer
 289  *
 290  * If dst is not yet refcounted, let's do it
 291  */
 292 static inline void skb_dst_force(struct sk_buff *skb)
 293 {
 294         if (skb_dst_is_noref(skb)) {
 295                 WARN_ON(!rcu_read_lock_held());
 296                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 297                 dst_clone(skb_dst(skb));
 298         }
 299 }
 300
 301
 302 /**
 303  *      __skb_tunnel_rx - prepare skb for rx reinsert
 304  *      @skb: buffer
 305  *      @dev: tunnel device
 306  *
 307  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 308  *      so make some cleanups. (no accounting done)
 309  */
 310 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 311 {
 312         skb->dev = dev;
 313         skb->rxhash = 0;
 314         skb_set_queue_mapping(skb, 0);
 315         skb_dst_drop(skb);
 316         nf_reset(skb);
 317 }
 318
 319 /**
 320  *      skb_tunnel_rx - prepare skb for rx reinsert
 321  *      @skb: buffer
 322  *      @dev: tunnel device
 323  *
 324  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 325  *      so make some cleanups, and perform accounting.
 326  *      Note: this accounting is not SMP safe.
 327  */
 328 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 329 {
 330         /* TODO : stats should be SMP safe */
 331         dev->stats.rx_packets++;
 332         dev->stats.rx_bytes += skb->len;
 333         __skb_tunnel_rx(skb, dev);
 334 }
 335
 336 /* Children define the path of the packet through the
 337  * Linux networking.  Thus, destinations are stackable.
 338  */
 339
 340 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
 341 {
 342         struct dst_entry *child = dst_clone(skb_dst(skb)->child);
 343
 344         skb_dst_drop(skb);
 345         return child;
 346 }
 347
 348 extern int dst_discard(struct sk_buff *skb);
 349 extern void *dst_alloc(struct dst_ops * ops, struct net_device *dev,
 350                        int initial_ref, int initial_obsolete, int flags);
 351 extern void __dst_free(struct dst_entry * dst);
 352 extern struct dst_entry *dst_destroy(struct dst_entry * dst);
 353
 354 static inline void dst_free(struct dst_entry * dst)
 355 {
 356         if (dst->obsolete > 1)
 357                 return;
 358         if (!atomic_read(&dst->__refcnt)) {
 359                 dst = dst_destroy(dst);
 360                 if (!dst)
 361                         return;
 362         }
 363         __dst_free(dst);
 364 }
 365
 366 static inline void dst_rcu_free(struct rcu_head *head)
 367 {
 368         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 369         dst_free(dst);
 370 }
 371
 372 static inline void dst_confirm(struct dst_entry *dst)
 373 {
 374         if (dst)
 375                 neigh_confirm(dst->neighbour);
 376 }
 377
 378 static inline void dst_link_failure(struct sk_buff *skb)
 379 {
 380         struct dst_entry *dst = skb_dst(skb);
 381         if (dst && dst->ops && dst->ops->link_failure)
 382                 dst->ops->link_failure(skb);
 383 }
 384
 385 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 386 {
 387         unsigned long expires = jiffies + timeout;
 388
 389         if (expires == 0)
 390                 expires = 1;
 391
 392         if (dst->expires == 0 || time_before(expires, dst->expires))
 393                 dst->expires = expires;
 394 }
 395
 396 /* Output packet to network from transport.  */
 397 static inline int dst_output(struct sk_buff *skb)
 398 {
 399         return skb_dst(skb)->output(skb);
 400 }
 401
 402 /* Input packet from network to transport.  */
 403 static inline int dst_input(struct sk_buff *skb)
 404 {
 405         return skb_dst(skb)->input(skb);
 406 }
 407
 408 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 409 {
 410         if (dst->obsolete)
 411                 dst = dst->ops->check(dst, cookie);
 412         return dst;
 413 }
 414
 415 extern void             dst_init(void);
 416
 417 /* Flags for xfrm_lookup flags argument. */
 418 enum {
 419         XFRM_LOOKUP_ICMP = 1 << 0,
 420 };
 421
 422 struct flowi;
 423 #ifndef CONFIG_XFRM
 424 static inline struct dst_entry *xfrm_lookup(struct net *net,
 425                                             struct dst_entry *dst_orig,
 426                                             const struct flowi *fl, struct sock *sk,
 427                                             int flags)
 428 {
 429         return dst_orig;
 430 }
 431 #else
 432 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 433                                      const struct flowi *fl, struct sock *sk,
 434                                      int flags);
 435 #endif
 436
 437 #endif /* _NET_DST_H */