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 int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb) 30 { 31 kfree_skb(skb); 32 return 0; 33 } 34 EXPORT_SYMBOL(dst_discard_out); 35 36 const struct dst_metrics dst_default_metrics = { 37 /* This initializer is needed to force linker to place this variable 38 * into const section. Otherwise it might end into bss section. 39 * We really want to avoid false sharing on this variable, and catch 40 * any writes on it. 41 */ 42 .refcnt = REFCOUNT_INIT(1), 43 }; 44 EXPORT_SYMBOL(dst_default_metrics); 45 46 void dst_init(struct dst_entry *dst, struct dst_ops *ops, 47 struct net_device *dev, int initial_ref, int initial_obsolete, 48 unsigned short flags) 49 { 50 dst->dev = dev; 51 if (dev) 52 dev_hold(dev); 53 dst->ops = ops; 54 dst_init_metrics(dst, dst_default_metrics.metrics, true); 55 dst->expires = 0UL; 56 #ifdef CONFIG_XFRM 57 dst->xfrm = NULL; 58 #endif 59 dst->input = dst_discard; 60 dst->output = dst_discard_out; 61 dst->error = 0; 62 dst->obsolete = initial_obsolete; 63 dst->header_len = 0; 64 dst->trailer_len = 0; 65 #ifdef CONFIG_IP_ROUTE_CLASSID 66 dst->tclassid = 0; 67 #endif 68 dst->lwtstate = NULL; 69 atomic_set(&dst->__refcnt, initial_ref); 70 dst->__use = 0; 71 dst->lastuse = jiffies; 72 dst->flags = flags; 73 if (!(flags & DST_NOCOUNT)) 74 dst_entries_add(ops, 1); 75 } 76 EXPORT_SYMBOL(dst_init); 77 78 void *dst_alloc(struct dst_ops *ops, struct net_device *dev, 79 int initial_ref, int initial_obsolete, unsigned short flags) 80 { 81 struct dst_entry *dst; 82 83 if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) { 84 if (ops->gc(ops)) { 85 printk_ratelimited(KERN_NOTICE "Route cache is full: " 86 "consider increasing sysctl " 87 "net.ipv[4|6].route.max_size.\n"); 88 return NULL; 89 } 90 } 91 92 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC); 93 if (!dst) 94 return NULL; 95 96 dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags); 97 98 return dst; 99 } 100 EXPORT_SYMBOL(dst_alloc); 101 102 struct dst_entry *dst_destroy(struct dst_entry * dst) 103 { 104 struct dst_entry *child = NULL; 105 106 smp_rmb(); 107 108 #ifdef CONFIG_XFRM 109 if (dst->xfrm) { 110 struct xfrm_dst *xdst = (struct xfrm_dst *) dst; 111 112 child = xdst->child; 113 } 114 #endif 115 if (!(dst->flags & DST_NOCOUNT)) 116 dst_entries_add(dst->ops, -1); 117 118 if (dst->ops->destroy) 119 dst->ops->destroy(dst); 120 if (dst->dev) 121 dev_put(dst->dev); 122 123 lwtstate_put(dst->lwtstate); 124 125 if (dst->flags & DST_METADATA) 126 metadata_dst_free((struct metadata_dst *)dst); 127 else 128 kmem_cache_free(dst->ops->kmem_cachep, dst); 129 130 dst = child; 131 if (dst) 132 dst_release_immediate(dst); 133 return NULL; 134 } 135 EXPORT_SYMBOL(dst_destroy); 136 137 static void dst_destroy_rcu(struct rcu_head *head) 138 { 139 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head); 140 141 dst = dst_destroy(dst); 142 } 143 144 /* Operations to mark dst as DEAD and clean up the net device referenced 145 * by dst: 146 * 1. put the dst under loopback interface and discard all tx/rx packets 147 * on this route. 148 * 2. release the net_device 149 * This function should be called when removing routes from the fib tree 150 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to 151 * make the next dst_ops->check() fail. 152 */ 153 void dst_dev_put(struct dst_entry *dst) 154 { 155 struct net_device *dev = dst->dev; 156 157 dst->obsolete = DST_OBSOLETE_DEAD; 158 if (dst->ops->ifdown) 159 dst->ops->ifdown(dst, dev, true); 160 dst->input = dst_discard; 161 dst->output = dst_discard_out; 162 dst->dev = dev_net(dst->dev)->loopback_dev; 163 dev_hold(dst->dev); 164 dev_put(dev); 165 } 166 EXPORT_SYMBOL(dst_dev_put); 167 168 void dst_release(struct dst_entry *dst) 169 { 170 if (dst) { 171 int newrefcnt; 172 173 newrefcnt = atomic_dec_return(&dst->__refcnt); 174 if (unlikely(newrefcnt < 0)) 175 net_warn_ratelimited("%s: dst:%p refcnt:%d\n", 176 __func__, dst, newrefcnt); 177 if (!newrefcnt) 178 call_rcu(&dst->rcu_head, dst_destroy_rcu); 179 } 180 } 181 EXPORT_SYMBOL(dst_release); 182 183 void dst_release_immediate(struct dst_entry *dst) 184 { 185 if (dst) { 186 int newrefcnt; 187 188 newrefcnt = atomic_dec_return(&dst->__refcnt); 189 if (unlikely(newrefcnt < 0)) 190 net_warn_ratelimited("%s: dst:%p refcnt:%d\n", 191 __func__, dst, newrefcnt); 192 if (!newrefcnt) 193 dst_destroy(dst); 194 } 195 } 196 EXPORT_SYMBOL(dst_release_immediate); 197 198 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old) 199 { 200 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC); 201 202 if (p) { 203 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old); 204 unsigned long prev, new; 205 206 refcount_set(&p->refcnt, 1); 207 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics)); 208 209 new = (unsigned long) p; 210 prev = cmpxchg(&dst->_metrics, old, new); 211 212 if (prev != old) { 213 kfree(p); 214 p = (struct dst_metrics *)__DST_METRICS_PTR(prev); 215 if (prev & DST_METRICS_READ_ONLY) 216 p = NULL; 217 } else if (prev & DST_METRICS_REFCOUNTED) { 218 if (refcount_dec_and_test(&old_p->refcnt)) 219 kfree(old_p); 220 } 221 } 222 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0); 223 return (u32 *)p; 224 } 225 EXPORT_SYMBOL(dst_cow_metrics_generic); 226 227 /* Caller asserts that dst_metrics_read_only(dst) is false. */ 228 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old) 229 { 230 unsigned long prev, new; 231 232 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY; 233 prev = cmpxchg(&dst->_metrics, old, new); 234 if (prev == old) 235 kfree(__DST_METRICS_PTR(old)); 236 } 237 EXPORT_SYMBOL(__dst_destroy_metrics_generic); 238 239 static struct dst_ops md_dst_ops = { 240 .family = AF_UNSPEC, 241 }; 242 243 static int dst_md_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb) 244 { 245 WARN_ONCE(1, "Attempting to call output on metadata dst\n"); 246 kfree_skb(skb); 247 return 0; 248 } 249 250 static int dst_md_discard(struct sk_buff *skb) 251 { 252 WARN_ONCE(1, "Attempting to call input on metadata dst\n"); 253 kfree_skb(skb); 254 return 0; 255 } 256 257 static void __metadata_dst_init(struct metadata_dst *md_dst, 258 enum metadata_type type, u8 optslen) 259 260 { 261 struct dst_entry *dst; 262 263 dst = &md_dst->dst; 264 dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE, 265 DST_METADATA | DST_NOCOUNT); 266 267 dst->input = dst_md_discard; 268 dst->output = dst_md_discard_out; 269 270 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst)); 271 md_dst->type = type; 272 } 273 274 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type, 275 gfp_t flags) 276 { 277 struct metadata_dst *md_dst; 278 279 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags); 280 if (!md_dst) 281 return NULL; 282 283 __metadata_dst_init(md_dst, type, optslen); 284 285 return md_dst; 286 } 287 EXPORT_SYMBOL_GPL(metadata_dst_alloc); 288 289 void metadata_dst_free(struct metadata_dst *md_dst) 290 { 291 #ifdef CONFIG_DST_CACHE 292 if (md_dst->type == METADATA_IP_TUNNEL) 293 dst_cache_destroy(&md_dst->u.tun_info.dst_cache); 294 #endif 295 kfree(md_dst); 296 } 297 EXPORT_SYMBOL_GPL(metadata_dst_free); 298 299 struct metadata_dst __percpu * 300 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags) 301 { 302 int cpu; 303 struct metadata_dst __percpu *md_dst; 304 305 md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen, 306 __alignof__(struct metadata_dst), flags); 307 if (!md_dst) 308 return NULL; 309 310 for_each_possible_cpu(cpu) 311 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen); 312 313 return md_dst; 314 } 315 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu); 316 317 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst) 318 { 319 #ifdef CONFIG_DST_CACHE 320 int cpu; 321 322 for_each_possible_cpu(cpu) { 323 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu); 324 325 if (one_md_dst->type == METADATA_IP_TUNNEL) 326 dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache); 327 } 328 #endif 329 free_percpu(md_dst); 330 } 331 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu); 332