net/core/dst.c

   1 /*
   2  * net/core/dst.c       Protocol independent destination cache.
   3  *
   4  * Authors:             Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   5  *
   6  */
   7
   8 #include <linux/bitops.h>
   9 #include <linux/errno.h>
  10 #include <linux/init.h>
  11 #include <linux/kernel.h>
  12 #include <linux/workqueue.h>
  13 #include <linux/mm.h>
  14 #include <linux/module.h>
  15 #include <linux/slab.h>
  16 #include <linux/netdevice.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/string.h>
  19 #include <linux/types.h>
  20 #include <net/net_namespace.h>
  21 #include <linux/sched.h>
  22 #include <linux/prefetch.h>
  23 #include <net/lwtunnel.h>
  24
  25 #include <net/dst.h>
  26 #include <net/dst_metadata.h>
  27
  28 /*
  29  * Theory of operations:
  30  * 1) We use a list, protected by a spinlock, to add
  31  *    new entries from both BH and non-BH context.
  32  * 2) In order to keep spinlock held for a small delay,
  33  *    we use a second list where are stored long lived
  34  *    entries, that are handled by the garbage collect thread
  35  *    fired by a workqueue.
  36  * 3) This list is guarded by a mutex,
  37  *    so that the gc_task and dst_dev_event() can be synchronized.
  38  */
  39
  40 /*
  41  * We want to keep lock & list close together
  42  * to dirty as few cache lines as possible in __dst_free().
  43  * As this is not a very strong hint, we dont force an alignment on SMP.
  44  */
  45 int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
  46 {
  47         kfree_skb(skb);
  48         return 0;
  49 }
  50 EXPORT_SYMBOL(dst_discard_out);
  51
  52 const struct dst_metrics dst_default_metrics = {
  53         /* This initializer is needed to force linker to place this variable
  54          * into const section. Otherwise it might end into bss section.
  55          * We really want to avoid false sharing on this variable, and catch
  56          * any writes on it.
  57          */
  58         .refcnt = REFCOUNT_INIT(1),
  59 };
  60
  61 void dst_init(struct dst_entry *dst, struct dst_ops *ops,
  62               struct net_device *dev, int initial_ref, int initial_obsolete,
  63               unsigned short flags)
  64 {
  65         dst->child = NULL;
  66         dst->dev = dev;
  67         if (dev)
  68                 dev_hold(dev);
  69         dst->ops = ops;
  70         dst_init_metrics(dst, dst_default_metrics.metrics, true);
  71         dst->expires = 0UL;
  72         dst->path = dst;
  73         dst->from = NULL;
  74 #ifdef CONFIG_XFRM
  75         dst->xfrm = NULL;
  76 #endif
  77         dst->input = dst_discard;
  78         dst->output = dst_discard_out;
  79         dst->error = 0;
  80         dst->obsolete = initial_obsolete;
  81         dst->header_len = 0;
  82         dst->trailer_len = 0;
  83 #ifdef CONFIG_IP_ROUTE_CLASSID
  84         dst->tclassid = 0;
  85 #endif
  86         dst->lwtstate = NULL;
  87         atomic_set(&dst->__refcnt, initial_ref);
  88         dst->__use = 0;
  89         dst->lastuse = jiffies;
  90         dst->flags = flags;
  91         dst->next = NULL;
  92         if (!(flags & DST_NOCOUNT))
  93                 dst_entries_add(ops, 1);
  94 }
  95 EXPORT_SYMBOL(dst_init);
  96
  97 void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
  98                 int initial_ref, int initial_obsolete, unsigned short flags)
  99 {
 100         struct dst_entry *dst;
 101
 102         if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
 103                 if (ops->gc(ops))
 104                         return NULL;
 105         }
 106
 107         dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
 108         if (!dst)
 109                 return NULL;
 110
 111         dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
 112
 113         return dst;
 114 }
 115 EXPORT_SYMBOL(dst_alloc);
 116
 117 struct dst_entry *dst_destroy(struct dst_entry * dst)
 118 {
 119         struct dst_entry *child;
 120
 121         smp_rmb();
 122
 123         child = dst->child;
 124
 125         if (!(dst->flags & DST_NOCOUNT))
 126                 dst_entries_add(dst->ops, -1);
 127
 128         if (dst->ops->destroy)
 129                 dst->ops->destroy(dst);
 130         if (dst->dev)
 131                 dev_put(dst->dev);
 132
 133         lwtstate_put(dst->lwtstate);
 134
 135         if (dst->flags & DST_METADATA)
 136                 metadata_dst_free((struct metadata_dst *)dst);
 137         else
 138                 kmem_cache_free(dst->ops->kmem_cachep, dst);
 139
 140         dst = child;
 141         if (dst)
 142                 dst_release_immediate(dst);
 143         return NULL;
 144 }
 145 EXPORT_SYMBOL(dst_destroy);
 146
 147 static void dst_destroy_rcu(struct rcu_head *head)
 148 {
 149         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 150
 151         dst = dst_destroy(dst);
 152 }
 153
 154 /* Operations to mark dst as DEAD and clean up the net device referenced
 155  * by dst:
 156  * 1. put the dst under loopback interface and discard all tx/rx packets
 157  *    on this route.
 158  * 2. release the net_device
 159  * This function should be called when removing routes from the fib tree
 160  * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
 161  * make the next dst_ops->check() fail.
 162  */
 163 void dst_dev_put(struct dst_entry *dst)
 164 {
 165         struct net_device *dev = dst->dev;
 166
 167         dst->obsolete = DST_OBSOLETE_DEAD;
 168         if (dst->ops->ifdown)
 169                 dst->ops->ifdown(dst, dev, true);
 170         dst->input = dst_discard;
 171         dst->output = dst_discard_out;
 172         dst->dev = dev_net(dst->dev)->loopback_dev;
 173         dev_hold(dst->dev);
 174         dev_put(dev);
 175 }
 176 EXPORT_SYMBOL(dst_dev_put);
 177
 178 void dst_release(struct dst_entry *dst)
 179 {
 180         if (dst) {
 181                 int newrefcnt;
 182
 183                 newrefcnt = atomic_dec_return(&dst->__refcnt);
 184                 if (unlikely(newrefcnt < 0))
 185                         net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
 186                                              __func__, dst, newrefcnt);
 187                 if (!newrefcnt)
 188                         call_rcu(&dst->rcu_head, dst_destroy_rcu);
 189         }
 190 }
 191 EXPORT_SYMBOL(dst_release);
 192
 193 void dst_release_immediate(struct dst_entry *dst)
 194 {
 195         if (dst) {
 196                 int newrefcnt;
 197
 198                 newrefcnt = atomic_dec_return(&dst->__refcnt);
 199                 if (unlikely(newrefcnt < 0))
 200                         net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
 201                                              __func__, dst, newrefcnt);
 202                 if (!newrefcnt)
 203                         dst_destroy(dst);
 204         }
 205 }
 206 EXPORT_SYMBOL(dst_release_immediate);
 207
 208 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
 209 {
 210         struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
 211
 212         if (p) {
 213                 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
 214                 unsigned long prev, new;
 215
 216                 refcount_set(&p->refcnt, 1);
 217                 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
 218
 219                 new = (unsigned long) p;
 220                 prev = cmpxchg(&dst->_metrics, old, new);
 221
 222                 if (prev != old) {
 223                         kfree(p);
 224                         p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
 225                         if (prev & DST_METRICS_READ_ONLY)
 226                                 p = NULL;
 227                 } else if (prev & DST_METRICS_REFCOUNTED) {
 228                         if (refcount_dec_and_test(&old_p->refcnt))
 229                                 kfree(old_p);
 230                 }
 231         }
 232         BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
 233         return (u32 *)p;
 234 }
 235 EXPORT_SYMBOL(dst_cow_metrics_generic);
 236
 237 /* Caller asserts that dst_metrics_read_only(dst) is false.  */
 238 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
 239 {
 240         unsigned long prev, new;
 241
 242         new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
 243         prev = cmpxchg(&dst->_metrics, old, new);
 244         if (prev == old)
 245                 kfree(__DST_METRICS_PTR(old));
 246 }
 247 EXPORT_SYMBOL(__dst_destroy_metrics_generic);
 248
 249 static struct dst_ops md_dst_ops = {
 250         .family =               AF_UNSPEC,
 251 };
 252
 253 static int dst_md_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 254 {
 255         WARN_ONCE(1, "Attempting to call output on metadata dst\n");
 256         kfree_skb(skb);
 257         return 0;
 258 }
 259
 260 static int dst_md_discard(struct sk_buff *skb)
 261 {
 262         WARN_ONCE(1, "Attempting to call input on metadata dst\n");
 263         kfree_skb(skb);
 264         return 0;
 265 }
 266
 267 static void __metadata_dst_init(struct metadata_dst *md_dst,
 268                                 enum metadata_type type, u8 optslen)
 269
 270 {
 271         struct dst_entry *dst;
 272
 273         dst = &md_dst->dst;
 274         dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE,
 275                  DST_METADATA | DST_NOCOUNT);
 276
 277         dst->input = dst_md_discard;
 278         dst->output = dst_md_discard_out;
 279
 280         memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
 281         md_dst->type = type;
 282 }
 283
 284 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
 285                                         gfp_t flags)
 286 {
 287         struct metadata_dst *md_dst;
 288
 289         md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
 290         if (!md_dst)
 291                 return NULL;
 292
 293         __metadata_dst_init(md_dst, type, optslen);
 294
 295         return md_dst;
 296 }
 297 EXPORT_SYMBOL_GPL(metadata_dst_alloc);
 298
 299 void metadata_dst_free(struct metadata_dst *md_dst)
 300 {
 301 #ifdef CONFIG_DST_CACHE
 302         if (md_dst->type == METADATA_IP_TUNNEL)
 303                 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
 304 #endif
 305         kfree(md_dst);
 306 }
 307
 308 struct metadata_dst __percpu *
 309 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
 310 {
 311         int cpu;
 312         struct metadata_dst __percpu *md_dst;
 313
 314         md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
 315                                     __alignof__(struct metadata_dst), flags);
 316         if (!md_dst)
 317                 return NULL;
 318
 319         for_each_possible_cpu(cpu)
 320                 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
 321
 322         return md_dst;
 323 }
 324 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
 325
 326 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
 327 {
 328 #ifdef CONFIG_DST_CACHE
 329         int cpu;
 330
 331         for_each_possible_cpu(cpu) {
 332                 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
 333
 334                 if (one_md_dst->type == METADATA_IP_TUNNEL)
 335                         dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
 336         }
 337 #endif
 338         free_percpu(md_dst);
 339 }
 340 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);