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 EXPORT_SYMBOL(dst_default_metrics);
  62
  63 void dst_init(struct dst_entry *dst, struct dst_ops *ops,
  64               struct net_device *dev, int initial_ref, int initial_obsolete,
  65               unsigned short flags)
  66 {
  67         dst->dev = dev;
  68         if (dev)
  69                 dev_hold(dev);
  70         dst->ops = ops;
  71         dst_init_metrics(dst, dst_default_metrics.metrics, true);
  72         dst->expires = 0UL;
  73 #ifdef CONFIG_XFRM
  74         dst->xfrm = NULL;
  75 #endif
  76         dst->input = dst_discard;
  77         dst->output = dst_discard_out;
  78         dst->error = 0;
  79         dst->obsolete = initial_obsolete;
  80         dst->header_len = 0;
  81         dst->trailer_len = 0;
  82 #ifdef CONFIG_IP_ROUTE_CLASSID
  83         dst->tclassid = 0;
  84 #endif
  85         dst->lwtstate = NULL;
  86         atomic_set(&dst->__refcnt, initial_ref);
  87         dst->__use = 0;
  88         dst->lastuse = jiffies;
  89         dst->flags = flags;
  90         if (!(flags & DST_NOCOUNT))
  91                 dst_entries_add(ops, 1);
  92 }
  93 EXPORT_SYMBOL(dst_init);
  94
  95 void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
  96                 int initial_ref, int initial_obsolete, unsigned short flags)
  97 {
  98         struct dst_entry *dst;
  99
 100         if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
 101                 if (ops->gc(ops))
 102                         return NULL;
 103         }
 104
 105         dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
 106         if (!dst)
 107                 return NULL;
 108
 109         dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
 110
 111         return dst;
 112 }
 113 EXPORT_SYMBOL(dst_alloc);
 114
 115 struct dst_entry *dst_destroy(struct dst_entry * dst)
 116 {
 117         struct dst_entry *child = NULL;
 118
 119         smp_rmb();
 120
 121 #ifdef CONFIG_XFRM
 122         if (dst->xfrm) {
 123                 struct xfrm_dst *xdst = (struct xfrm_dst *) dst;
 124
 125                 child = xdst->child;
 126         }
 127 #endif
 128         if (!(dst->flags & DST_NOCOUNT))
 129                 dst_entries_add(dst->ops, -1);
 130
 131         if (dst->ops->destroy)
 132                 dst->ops->destroy(dst);
 133         if (dst->dev)
 134                 dev_put(dst->dev);
 135
 136         lwtstate_put(dst->lwtstate);
 137
 138         if (dst->flags & DST_METADATA)
 139                 metadata_dst_free((struct metadata_dst *)dst);
 140         else
 141                 kmem_cache_free(dst->ops->kmem_cachep, dst);
 142
 143         dst = child;
 144         if (dst)
 145                 dst_release_immediate(dst);
 146         return NULL;
 147 }
 148 EXPORT_SYMBOL(dst_destroy);
 149
 150 static void dst_destroy_rcu(struct rcu_head *head)
 151 {
 152         struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
 153
 154         dst = dst_destroy(dst);
 155 }
 156
 157 /* Operations to mark dst as DEAD and clean up the net device referenced
 158  * by dst:
 159  * 1. put the dst under loopback interface and discard all tx/rx packets
 160  *    on this route.
 161  * 2. release the net_device
 162  * This function should be called when removing routes from the fib tree
 163  * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to
 164  * make the next dst_ops->check() fail.
 165  */
 166 void dst_dev_put(struct dst_entry *dst)
 167 {
 168         struct net_device *dev = dst->dev;
 169
 170         dst->obsolete = DST_OBSOLETE_DEAD;
 171         if (dst->ops->ifdown)
 172                 dst->ops->ifdown(dst, dev, true);
 173         dst->input = dst_discard;
 174         dst->output = dst_discard_out;
 175         dst->dev = dev_net(dst->dev)->loopback_dev;
 176         dev_hold(dst->dev);
 177         dev_put(dev);
 178 }
 179 EXPORT_SYMBOL(dst_dev_put);
 180
 181 void dst_release(struct dst_entry *dst)
 182 {
 183         if (dst) {
 184                 int newrefcnt;
 185
 186                 newrefcnt = atomic_dec_return(&dst->__refcnt);
 187                 if (unlikely(newrefcnt < 0))
 188                         net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
 189                                              __func__, dst, newrefcnt);
 190                 if (!newrefcnt)
 191                         call_rcu(&dst->rcu_head, dst_destroy_rcu);
 192         }
 193 }
 194 EXPORT_SYMBOL(dst_release);
 195
 196 void dst_release_immediate(struct dst_entry *dst)
 197 {
 198         if (dst) {
 199                 int newrefcnt;
 200
 201                 newrefcnt = atomic_dec_return(&dst->__refcnt);
 202                 if (unlikely(newrefcnt < 0))
 203                         net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
 204                                              __func__, dst, newrefcnt);
 205                 if (!newrefcnt)
 206                         dst_destroy(dst);
 207         }
 208 }
 209 EXPORT_SYMBOL(dst_release_immediate);
 210
 211 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
 212 {
 213         struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC);
 214
 215         if (p) {
 216                 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old);
 217                 unsigned long prev, new;
 218
 219                 refcount_set(&p->refcnt, 1);
 220                 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics));
 221
 222                 new = (unsigned long) p;
 223                 prev = cmpxchg(&dst->_metrics, old, new);
 224
 225                 if (prev != old) {
 226                         kfree(p);
 227                         p = (struct dst_metrics *)__DST_METRICS_PTR(prev);
 228                         if (prev & DST_METRICS_READ_ONLY)
 229                                 p = NULL;
 230                 } else if (prev & DST_METRICS_REFCOUNTED) {
 231                         if (refcount_dec_and_test(&old_p->refcnt))
 232                                 kfree(old_p);
 233                 }
 234         }
 235         BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0);
 236         return (u32 *)p;
 237 }
 238 EXPORT_SYMBOL(dst_cow_metrics_generic);
 239
 240 /* Caller asserts that dst_metrics_read_only(dst) is false.  */
 241 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
 242 {
 243         unsigned long prev, new;
 244
 245         new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY;
 246         prev = cmpxchg(&dst->_metrics, old, new);
 247         if (prev == old)
 248                 kfree(__DST_METRICS_PTR(old));
 249 }
 250 EXPORT_SYMBOL(__dst_destroy_metrics_generic);
 251
 252 static struct dst_ops md_dst_ops = {
 253         .family =               AF_UNSPEC,
 254 };
 255
 256 static int dst_md_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 257 {
 258         WARN_ONCE(1, "Attempting to call output on metadata dst\n");
 259         kfree_skb(skb);
 260         return 0;
 261 }
 262
 263 static int dst_md_discard(struct sk_buff *skb)
 264 {
 265         WARN_ONCE(1, "Attempting to call input on metadata dst\n");
 266         kfree_skb(skb);
 267         return 0;
 268 }
 269
 270 static void __metadata_dst_init(struct metadata_dst *md_dst,
 271                                 enum metadata_type type, u8 optslen)
 272
 273 {
 274         struct dst_entry *dst;
 275
 276         dst = &md_dst->dst;
 277         dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE,
 278                  DST_METADATA | DST_NOCOUNT);
 279
 280         dst->input = dst_md_discard;
 281         dst->output = dst_md_discard_out;
 282
 283         memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
 284         md_dst->type = type;
 285 }
 286
 287 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type,
 288                                         gfp_t flags)
 289 {
 290         struct metadata_dst *md_dst;
 291
 292         md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
 293         if (!md_dst)
 294                 return NULL;
 295
 296         __metadata_dst_init(md_dst, type, optslen);
 297
 298         return md_dst;
 299 }
 300 EXPORT_SYMBOL_GPL(metadata_dst_alloc);
 301
 302 void metadata_dst_free(struct metadata_dst *md_dst)
 303 {
 304 #ifdef CONFIG_DST_CACHE
 305         if (md_dst->type == METADATA_IP_TUNNEL)
 306                 dst_cache_destroy(&md_dst->u.tun_info.dst_cache);
 307 #endif
 308         kfree(md_dst);
 309 }
 310
 311 struct metadata_dst __percpu *
 312 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags)
 313 {
 314         int cpu;
 315         struct metadata_dst __percpu *md_dst;
 316
 317         md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen,
 318                                     __alignof__(struct metadata_dst), flags);
 319         if (!md_dst)
 320                 return NULL;
 321
 322         for_each_possible_cpu(cpu)
 323                 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen);
 324
 325         return md_dst;
 326 }
 327 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu);
 328
 329 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
 330 {
 331 #ifdef CONFIG_DST_CACHE
 332         int cpu;
 333
 334         for_each_possible_cpu(cpu) {
 335                 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu);
 336
 337                 if (one_md_dst->type == METADATA_IP_TUNNEL)
 338                         dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache);
 339         }
 340 #endif
 341         free_percpu(md_dst);
 342 }
 343 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);