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 #include <net/xfrm.h>
  25
  26 #include <net/dst.h>
  27 #include <net/dst_metadata.h>
  28
  29 /*
  30  * Theory of operations:
  31  * 1) We use a list, protected by a spinlock, to add
  32  *    new entries from both BH and non-BH context.
  33  * 2) In order to keep spinlock held for a small delay,
  34  *    we use a second list where are stored long lived
  35  *    entries, that are handled by the garbage collect thread
  36  *    fired by a workqueue.
  37  * 3) This list is guarded by a mutex,
  38  *    so that the gc_task and dst_dev_event() can be synchronized.
  39  */
  40
  41 /*
  42  * We want to keep lock & list close together
  43  * to dirty as few cache lines as possible in __dst_free().
  44  * As this is not a very strong hint, we dont force an alignment on SMP.
  45  */
  46 int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
  47 {
  48         kfree_skb(skb);
  49         return 0;
  50 }
  51 EXPORT_SYMBOL(dst_discard_out);
  52
  53 const struct dst_metrics dst_default_metrics = {
  54         /* This initializer is needed to force linker to place this variable
  55          * into const section. Otherwise it might end into bss section.
  56          * We really want to avoid false sharing on this variable, and catch
  57          * any writes on it.
  58          */
  59         .refcnt = REFCOUNT_INIT(1),
  60 };
  61
  62 void dst_init(struct dst_entry *dst, struct dst_ops *ops,
  63               struct net_device *dev, int initial_ref, int initial_obsolete,
  64               unsigned short flags)
  65 {
  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 #ifdef CONFIG_XFRM
  73         dst->xfrm = NULL;
  74 #endif
  75         dst->input = dst_discard;
  76         dst->output = dst_discard_out;
  77         dst->error = 0;
  78         dst->obsolete = initial_obsolete;
  79         dst->header_len = 0;
  80         dst->trailer_len = 0;
  81 #ifdef CONFIG_IP_ROUTE_CLASSID
  82         dst->tclassid = 0;
  83 #endif
  84         dst->lwtstate = NULL;
  85         atomic_set(&dst->__refcnt, initial_ref);
  86         dst->__use = 0;
  87         dst->lastuse = jiffies;
  88         dst->flags = flags;
  89         if (!(flags & DST_NOCOUNT))
  90                 dst_entries_add(ops, 1);
  91 }
  92 EXPORT_SYMBOL(dst_init);
  93
  94 void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
  95                 int initial_ref, int initial_obsolete, unsigned short flags)
  96 {
  97         struct dst_entry *dst;
  98
  99         if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
 100                 if (ops->gc(ops))
 101                         return NULL;
 102         }
 103
 104         dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
 105         if (!dst)
 106                 return NULL;
 107
 108         dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
 109
 110         return dst;
 111 }
 112 EXPORT_SYMBOL(dst_alloc);
 113
 114 struct dst_entry *dst_destroy(struct dst_entry * dst)
 115 {
 116         struct dst_entry *child = NULL;
 117
 118         smp_rmb();
 119
 120 #ifdef CONFIG_XFRM
 121         if (dst->xfrm) {
 122                 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
 123
 124                 child = xdst->child;
 125         }
 126 #endif
 127         if (!(dst->flags & DST_NOCOUNT))
 128                 dst_entries_add(dst->ops, -1);
 129
 130         if (dst->ops->destroy)
 131                 dst->ops->destroy(dst);
 132         if (dst->dev)
 133                 dev_put(dst->dev);
 134
 135         lwtstate_put(dst->lwtstate);
 136
 137         if (dst->flags & DST_METADATA)
 138                 metadata_dst_free((struct metadata_dst *)dst);
 139         else
 140                 kmem_cache_free(dst->ops->kmem_cachep, dst);
 141
 142         dst = child;
 143         if (dst)
 144                 dst_release_immediate(dst);
 145         return NULL;
 146 }
 147 EXPORT_SYMBOL(dst_destroy);
 148
 149 static void dst_destroy_rcu(struct rcu_head *head)
 150 {
 151         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 152
 153         dst = dst_destroy(dst);
 154 }
 155
 156 /* Operations to mark dst as DEAD and clean up the net device referenced
 157  * by dst:
 158  * 1. put the dst under loopback interface and discard all tx/rx packets
 159  *    on this route.
 160  * 2. release the net_device
 161  * This function should be called when removing routes from the fib tree
 162  * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
 163  * make the next dst_ops->check() fail.
 164  */
 165 void dst_dev_put(struct dst_entry *dst)
 166 {
 167         struct net_device *dev = dst->dev;
 168
 169         dst->obsolete = DST_OBSOLETE_DEAD;
 170         if (dst->ops->ifdown)
 171                 dst->ops->ifdown(dst, dev, true);
 172         dst->input = dst_discard;
 173         dst->output = dst_discard_out;
 174         dst->dev = dev_net(dst->dev)->loopback_dev;
 175         dev_hold(dst->dev);
 176         dev_put(dev);
 177 }
 178 EXPORT_SYMBOL(dst_dev_put);
 179
 180 void dst_release(struct dst_entry *dst)
 181 {
 182         if (dst) {
 183                 int newrefcnt;
 184
 185                 newrefcnt = atomic_dec_return(&dst->__refcnt);
 186                 if (unlikely(newrefcnt < 0))
 187                         net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
 188                                              __func__, dst, newrefcnt);
 189                 if (!newrefcnt)
 190                         call_rcu(&dst->rcu_head, dst_destroy_rcu);
 191         }
 192 }
 193 EXPORT_SYMBOL(dst_release);
 194
 195 void dst_release_immediate(struct dst_entry *dst)
 196 {
 197         if (dst) {
 198                 int newrefcnt;
 199
 200                 newrefcnt = atomic_dec_return(&dst->__refcnt);
 201                 if (unlikely(newrefcnt < 0))
 202                         net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
 203                                              __func__, dst, newrefcnt);
 204                 if (!newrefcnt)
 205                         dst_destroy(dst);
 206         }
 207 }
 208 EXPORT_SYMBOL(dst_release_immediate);
 209
 210 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
 211 {
 212         struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
 213
 214         if (p) {
 215                 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
 216                 unsigned long prev, new;
 217
 218                 refcount_set(&p->refcnt, 1);
 219                 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
 220
 221                 new = (unsigned long) p;
 222                 prev = cmpxchg(&dst->_metrics, old, new);
 223
 224                 if (prev != old) {
 225                         kfree(p);
 226                         p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
 227                         if (prev & DST_METRICS_READ_ONLY)
 228                                 p = NULL;
 229                 } else if (prev & DST_METRICS_REFCOUNTED) {
 230                         if (refcount_dec_and_test(&old_p->refcnt))
 231                                 kfree(old_p);
 232                 }
 233         }
 234         BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
 235         return (u32 *)p;
 236 }
 237 EXPORT_SYMBOL(dst_cow_metrics_generic);
 238
 239 /* Caller asserts that dst_metrics_read_only(dst) is false.  */
 240 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
 241 {
 242         unsigned long prev, new;
 243
 244         new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
 245         prev = cmpxchg(&dst->_metrics, old, new);
 246         if (prev == old)
 247                 kfree(__DST_METRICS_PTR(old));
 248 }
 249 EXPORT_SYMBOL(__dst_destroy_metrics_generic);
 250
 251 static struct dst_ops md_dst_ops = {
 252         .family =               AF_UNSPEC,
 253 };
 254
 255 static int dst_md_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 256 {
 257         WARN_ONCE(1, "Attempting to call output on metadata dst\n");
 258         kfree_skb(skb);
 259         return 0;
 260 }
 261
 262 static int dst_md_discard(struct sk_buff *skb)
 263 {
 264         WARN_ONCE(1, "Attempting to call input on metadata dst\n");
 265         kfree_skb(skb);
 266         return 0;
 267 }
 268
 269 static void __metadata_dst_init(struct metadata_dst *md_dst,
 270                                 enum metadata_type type, u8 optslen)
 271
 272 {
 273         struct dst_entry *dst;
 274
 275         dst = &md_dst->dst;
 276         dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE,
 277                  DST_METADATA | DST_NOCOUNT);
 278
 279         dst->input = dst_md_discard;
 280         dst->output = dst_md_discard_out;
 281
 282         memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
 283         md_dst->type = type;
 284 }
 285
 286 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
 287                                         gfp_t flags)
 288 {
 289         struct metadata_dst *md_dst;
 290
 291         md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
 292         if (!md_dst)
 293                 return NULL;
 294
 295         __metadata_dst_init(md_dst, type, optslen);
 296
 297         return md_dst;
 298 }
 299 EXPORT_SYMBOL_GPL(metadata_dst_alloc);
 300
 301 void metadata_dst_free(struct metadata_dst *md_dst)
 302 {
 303 #ifdef CONFIG_DST_CACHE
 304         if (md_dst->type == METADATA_IP_TUNNEL)
 305                 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
 306 #endif
 307         kfree(md_dst);
 308 }
 309
 310 struct metadata_dst __percpu *
 311 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
 312 {
 313         int cpu;
 314         struct metadata_dst __percpu *md_dst;
 315
 316         md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
 317                                     __alignof__(struct metadata_dst), flags);
 318         if (!md_dst)
 319                 return NULL;
 320
 321         for_each_possible_cpu(cpu)
 322                 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
 323
 324         return md_dst;
 325 }
 326 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
 327
 328 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
 329 {
 330 #ifdef CONFIG_DST_CACHE
 331         int cpu;
 332
 333         for_each_possible_cpu(cpu) {
 334                 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
 335
 336                 if (one_md_dst->type == METADATA_IP_TUNNEL)
 337                         dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
 338         }
 339 #endif
 340         free_percpu(md_dst);
 341 }
 342 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);