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 /*
  20  * 0 - no debugging messages
  21  * 1 - rare events and bugs (default)
  22  * 2 - trace mode.
  23  */
  24 #define RT_CACHE_DEBUG          0
  25
  26 #define DST_GC_MIN      (HZ/10)
  27 #define DST_GC_INC      (HZ/2)
  28 #define DST_GC_MAX      (120*HZ)
  29
  30 /* Each dst_entry has reference count and sits in some parent list(s).
  31  * When it is removed from parent list, it is "freed" (dst_free).
  32  * After this it enters dead state (dst->obsolete > 0) and if its refcnt
  33  * is zero, it can be destroyed immediately, otherwise it is added
  34  * to gc list and garbage collector periodically checks the refcnt.
  35  */
  36
  37 struct sk_buff;
  38
  39 struct dst_entry {
  40         struct rcu_head         rcu_head;
  41         struct dst_entry        *child;
  42         struct net_device       *dev;
  43         struct  dst_ops         *ops;
  44         unsigned long           _metrics;
  45         unsigned long           expires;
  46         struct dst_entry        *path;
  47         struct neighbour        *neighbour;
  48         struct hh_cache         *hh;
  49 #ifdef CONFIG_XFRM
  50         struct xfrm_state       *xfrm;
  51 #else
  52         void                    *__pad1;
  53 #endif
  54         int                     (*input)(struct sk_buff*);
  55         int                     (*output)(struct sk_buff*);
  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[1];
  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         int                     flags;
  82 #define DST_HOST                0x0001
  83 #define DST_NOXFRM              0x0002
  84 #define DST_NOPOLICY            0x0004
  85 #define DST_NOHASH              0x0008
  86 #define DST_NOCACHE             0x0010
  87         union {
  88                 struct dst_entry        *next;
  89                 struct rtable __rcu     *rt_next;
  90                 struct rt6_info         *rt6_next;
  91                 struct dn_route __rcu   *dn_next;
  92         };
  93 };
  94
  95 #ifdef __KERNEL__
  96
  97 extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
  98 extern const u32 dst_default_metrics[RTAX_MAX];
  99
 100 #define DST_METRICS_READ_ONLY   0x1UL
 101 #define __DST_METRICS_PTR(Y)    \
 102         ((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
 103 #define DST_METRICS_PTR(X)      __DST_METRICS_PTR((X)->_metrics)
 104
 105 static inline bool dst_metrics_read_only(const struct dst_entry *dst)
 106 {
 107         return dst->_metrics & DST_METRICS_READ_ONLY;
 108 }
 109
 110 extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
 111
 112 static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
 113 {
 114         unsigned long val = dst->_metrics;
 115         if (!(val & DST_METRICS_READ_ONLY))
 116                 __dst_destroy_metrics_generic(dst, val);
 117 }
 118
 119 static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
 120 {
 121         unsigned long p = dst->_metrics;
 122
 123         BUG_ON(!p);
 124
 125         if (p & DST_METRICS_READ_ONLY)
 126                 return dst->ops->cow_metrics(dst, p);
 127         return __DST_METRICS_PTR(p);
 128 }
 129
 130 /* This may only be invoked before the entry has reached global
 131  * visibility.
 132  */
 133 static inline void dst_init_metrics(struct dst_entry *dst,
 134                                     const u32 *src_metrics,
 135                                     bool read_only)
 136 {
 137         dst->_metrics = ((unsigned long) src_metrics) |
 138                 (read_only ? DST_METRICS_READ_ONLY : 0);
 139 }
 140
 141 static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
 142 {
 143         u32 *dst_metrics = dst_metrics_write_ptr(dest);
 144
 145         if (dst_metrics) {
 146                 u32 *src_metrics = DST_METRICS_PTR(src);
 147
 148                 memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
 149         }
 150 }
 151
 152 static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
 153 {
 154         return DST_METRICS_PTR(dst);
 155 }
 156
 157 static inline u32
 158 dst_metric_raw(const struct dst_entry *dst, const int metric)
 159 {
 160         u32 *p = DST_METRICS_PTR(dst);
 161
 162         return p[metric-1];
 163 }
 164
 165 static inline u32
 166 dst_metric(const struct dst_entry *dst, const int metric)
 167 {
 168         WARN_ON_ONCE(metric == RTAX_HOPLIMIT ||
 169                      metric == RTAX_ADVMSS ||
 170                      metric == RTAX_MTU);
 171         return dst_metric_raw(dst, metric);
 172 }
 173
 174 static inline u32
 175 dst_metric_advmss(const struct dst_entry *dst)
 176 {
 177         u32 advmss = dst_metric_raw(dst, RTAX_ADVMSS);
 178
 179         if (!advmss)
 180                 advmss = dst->ops->default_advmss(dst);
 181
 182         return advmss;
 183 }
 184
 185 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 186 {
 187         u32 *p = dst_metrics_write_ptr(dst);
 188
 189         if (p)
 190                 p[metric-1] = val;
 191 }
 192
 193 static inline u32
 194 dst_feature(const struct dst_entry *dst, u32 feature)
 195 {
 196         return dst_metric(dst, RTAX_FEATURES) & feature;
 197 }
 198
 199 static inline u32 dst_mtu(const struct dst_entry *dst)
 200 {
 201         u32 mtu = dst_metric_raw(dst, RTAX_MTU);
 202
 203         if (!mtu)
 204                 mtu = dst->ops->default_mtu(dst);
 205
 206         return mtu;
 207 }
 208
 209 /* RTT metrics are stored in milliseconds for user ABI, but used as jiffies */
 210 static inline unsigned long dst_metric_rtt(const struct dst_entry *dst, int metric)
 211 {
 212         return msecs_to_jiffies(dst_metric(dst, metric));
 213 }
 214
 215 static inline void set_dst_metric_rtt(struct dst_entry *dst, int metric,
 216                                       unsigned long rtt)
 217 {
 218         dst_metric_set(dst, metric, jiffies_to_msecs(rtt));
 219 }
 220
 221 static inline u32
 222 dst_allfrag(const struct dst_entry *dst)
 223 {
 224         int ret = dst_feature(dst,  RTAX_FEATURE_ALLFRAG);
 225         return ret;
 226 }
 227
 228 static inline int
 229 dst_metric_locked(const struct dst_entry *dst, int metric)
 230 {
 231         return dst_metric(dst, RTAX_LOCK) & (1<<metric);
 232 }
 233
 234 static inline void dst_hold(struct dst_entry * dst)
 235 {
 236         /*
 237          * If your kernel compilation stops here, please check
 238          * __pad_to_align_refcnt declaration in struct dst_entry
 239          */
 240         BUILD_BUG_ON(offsetof(struct dst_entry, __refcnt) & 63);
 241         atomic_inc(&dst->__refcnt);
 242 }
 243
 244 static inline void dst_use(struct dst_entry *dst, unsigned long time)
 245 {
 246         dst_hold(dst);
 247         dst->__use++;
 248         dst->lastuse = time;
 249 }
 250
 251 static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
 252 {
 253         dst->__use++;
 254         dst->lastuse = time;
 255 }
 256
 257 static inline
 258 struct dst_entry * dst_clone(struct dst_entry * dst)
 259 {
 260         if (dst)
 261                 atomic_inc(&dst->__refcnt);
 262         return dst;
 263 }
 264
 265 extern void dst_release(struct dst_entry *dst);
 266
 267 static inline void refdst_drop(unsigned long refdst)
 268 {
 269         if (!(refdst & SKB_DST_NOREF))
 270                 dst_release((struct dst_entry *)(refdst & SKB_DST_PTRMASK));
 271 }
 272
 273 /**
 274  * skb_dst_drop - drops skb dst
 275  * @skb: buffer
 276  *
 277  * Drops dst reference count if a reference was taken.
 278  */
 279 static inline void skb_dst_drop(struct sk_buff *skb)
 280 {
 281         if (skb->_skb_refdst) {
 282                 refdst_drop(skb->_skb_refdst);
 283                 skb->_skb_refdst = 0UL;
 284         }
 285 }
 286
 287 static inline void skb_dst_copy(struct sk_buff *nskb, const struct sk_buff *oskb)
 288 {
 289         nskb->_skb_refdst = oskb->_skb_refdst;
 290         if (!(nskb->_skb_refdst & SKB_DST_NOREF))
 291                 dst_clone(skb_dst(nskb));
 292 }
 293
 294 /**
 295  * skb_dst_force - makes sure skb dst is refcounted
 296  * @skb: buffer
 297  *
 298  * If dst is not yet refcounted, let's do it
 299  */
 300 static inline void skb_dst_force(struct sk_buff *skb)
 301 {
 302         if (skb_dst_is_noref(skb)) {
 303                 WARN_ON(!rcu_read_lock_held());
 304                 skb->_skb_refdst &= ~SKB_DST_NOREF;
 305                 dst_clone(skb_dst(skb));
 306         }
 307 }
 308
 309
 310 /**
 311  *      __skb_tunnel_rx - prepare skb for rx reinsert
 312  *      @skb: buffer
 313  *      @dev: tunnel device
 314  *
 315  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 316  *      so make some cleanups. (no accounting done)
 317  */
 318 static inline void __skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 319 {
 320         skb->dev = dev;
 321         skb->rxhash = 0;
 322         skb_set_queue_mapping(skb, 0);
 323         skb_dst_drop(skb);
 324         nf_reset(skb);
 325 }
 326
 327 /**
 328  *      skb_tunnel_rx - prepare skb for rx reinsert
 329  *      @skb: buffer
 330  *      @dev: tunnel device
 331  *
 332  *      After decapsulation, packet is going to re-enter (netif_rx()) our stack,
 333  *      so make some cleanups, and perform accounting.
 334  *      Note: this accounting is not SMP safe.
 335  */
 336 static inline void skb_tunnel_rx(struct sk_buff *skb, struct net_device *dev)
 337 {
 338         /* TODO : stats should be SMP safe */
 339         dev->stats.rx_packets++;
 340         dev->stats.rx_bytes += skb->len;
 341         __skb_tunnel_rx(skb, dev);
 342 }
 343
 344 /* Children define the path of the packet through the
 345  * Linux networking.  Thus, destinations are stackable.
 346  */
 347
 348 static inline struct dst_entry *skb_dst_pop(struct sk_buff *skb)
 349 {
 350         struct dst_entry *child = dst_clone(skb_dst(skb)->child);
 351
 352         skb_dst_drop(skb);
 353         return child;
 354 }
 355
 356 extern int dst_discard(struct sk_buff *skb);
 357 extern void *dst_alloc(struct dst_ops * ops, int initial_ref);
 358 extern void __dst_free(struct dst_entry * dst);
 359 extern struct dst_entry *dst_destroy(struct dst_entry * dst);
 360
 361 static inline void dst_free(struct dst_entry * dst)
 362 {
 363         if (dst->obsolete > 1)
 364                 return;
 365         if (!atomic_read(&dst->__refcnt)) {
 366                 dst = dst_destroy(dst);
 367                 if (!dst)
 368                         return;
 369         }
 370         __dst_free(dst);
 371 }
 372
 373 static inline void dst_rcu_free(struct rcu_head *head)
 374 {
 375         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 376         dst_free(dst);
 377 }
 378
 379 static inline void dst_confirm(struct dst_entry *dst)
 380 {
 381         if (dst)
 382                 neigh_confirm(dst->neighbour);
 383 }
 384
 385 static inline void dst_link_failure(struct sk_buff *skb)
 386 {
 387         struct dst_entry *dst = skb_dst(skb);
 388         if (dst && dst->ops && dst->ops->link_failure)
 389                 dst->ops->link_failure(skb);
 390 }
 391
 392 static inline void dst_set_expires(struct dst_entry *dst, int timeout)
 393 {
 394         unsigned long expires = jiffies + timeout;
 395
 396         if (expires == 0)
 397                 expires = 1;
 398
 399         if (dst->expires == 0 || time_before(expires, dst->expires))
 400                 dst->expires = expires;
 401 }
 402
 403 /* Output packet to network from transport.  */
 404 static inline int dst_output(struct sk_buff *skb)
 405 {
 406         return skb_dst(skb)->output(skb);
 407 }
 408
 409 /* Input packet from network to transport.  */
 410 static inline int dst_input(struct sk_buff *skb)
 411 {
 412         return skb_dst(skb)->input(skb);
 413 }
 414
 415 static inline struct dst_entry *dst_check(struct dst_entry *dst, u32 cookie)
 416 {
 417         if (dst->obsolete)
 418                 dst = dst->ops->check(dst, cookie);
 419         return dst;
 420 }
 421
 422 extern void             dst_init(void);
 423
 424 /* Flags for xfrm_lookup flags argument. */
 425 enum {
 426         XFRM_LOOKUP_ICMP = 1 << 0,
 427 };
 428
 429 struct flowi;
 430 #ifndef CONFIG_XFRM
 431 static inline struct dst_entry *xfrm_lookup(struct net *net,
 432                                             struct dst_entry *dst_orig,
 433                                             const struct flowi *fl, struct sock *sk,
 434                                             int flags)
 435 {
 436         return dst_orig;
 437 }
 438 #else
 439 extern struct dst_entry *xfrm_lookup(struct net *net, struct dst_entry *dst_orig,
 440                                      const struct flowi *fl, struct sock *sk,
 441                                      int flags);
 442 #endif
 443 #endif
 444
 445 #endif /* _NET_DST_H */