1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * net/core/dst.c Protocol independent destination cache. 4 * 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 6 * 7 */ 8 9 #include <linux/bitops.h> 10 #include <linux/errno.h> 11 #include <linux/init.h> 12 #include <linux/kernel.h> 13 #include <linux/workqueue.h> 14 #include <linux/mm.h> 15 #include <linux/module.h> 16 #include <linux/slab.h> 17 #include <linux/netdevice.h> 18 #include <linux/skbuff.h> 19 #include <linux/string.h> 20 #include <linux/types.h> 21 #include <net/net_namespace.h> 22 #include <linux/sched.h> 23 #include <linux/prefetch.h> 24 #include <net/lwtunnel.h> 25 #include <net/xfrm.h> 26 27 #include <net/dst.h> 28 #include <net/dst_metadata.h> 29 30 int dst_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb) 31 { 32 kfree_skb(skb); 33 return 0; 34 } 35 EXPORT_SYMBOL(dst_discard_out); 36 37 const struct dst_metrics dst_default_metrics = { 38 /* This initializer is needed to force linker to place this variable 39 * into const section. Otherwise it might end into bss section. 40 * We really want to avoid false sharing on this variable, and catch 41 * any writes on it. 42 */ 43 .refcnt = REFCOUNT_INIT(1), 44 }; 45 EXPORT_SYMBOL(dst_default_metrics); 46 47 void dst_init(struct dst_entry *dst, struct dst_ops *ops, 48 struct net_device *dev, int initial_ref, int initial_obsolete, 49 unsigned short flags) 50 { 51 dst->dev = dev; 52 netdev_hold(dev, &dst->dev_tracker, GFP_ATOMIC); 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 && 84 !(flags & DST_NOCOUNT) && 85 dst_entries_get_fast(ops) > ops->gc_thresh) 86 ops->gc(ops); 87 88 dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC); 89 if (!dst) 90 return NULL; 91 92 dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags); 93 94 return dst; 95 } 96 EXPORT_SYMBOL(dst_alloc); 97 98 struct dst_entry *dst_destroy(struct dst_entry * dst) 99 { 100 struct dst_entry *child = NULL; 101 102 smp_rmb(); 103 104 #ifdef CONFIG_XFRM 105 if (dst->xfrm) { 106 struct xfrm_dst *xdst = (struct xfrm_dst *) dst; 107 108 child = xdst->child; 109 } 110 #endif 111 if (!(dst->flags & DST_NOCOUNT)) 112 dst_entries_add(dst->ops, -1); 113 114 if (dst->ops->destroy) 115 dst->ops->destroy(dst); 116 netdev_put(dst->dev, &dst->dev_tracker); 117 118 lwtstate_put(dst->lwtstate); 119 120 if (dst->flags & DST_METADATA) 121 metadata_dst_free((struct metadata_dst *)dst); 122 else 123 kmem_cache_free(dst->ops->kmem_cachep, dst); 124 125 dst = child; 126 if (dst) 127 dst_release_immediate(dst); 128 return NULL; 129 } 130 EXPORT_SYMBOL(dst_destroy); 131 132 static void dst_destroy_rcu(struct rcu_head *head) 133 { 134 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head); 135 136 dst = dst_destroy(dst); 137 } 138 139 /* Operations to mark dst as DEAD and clean up the net device referenced 140 * by dst: 141 * 1. put the dst under blackhole interface and discard all tx/rx packets 142 * on this route. 143 * 2. release the net_device 144 * This function should be called when removing routes from the fib tree 145 * in preparation for a NETDEV_DOWN/NETDEV_UNREGISTER event and also to 146 * make the next dst_ops->check() fail. 147 */ 148 void dst_dev_put(struct dst_entry *dst) 149 { 150 struct net_device *dev = dst->dev; 151 152 dst->obsolete = DST_OBSOLETE_DEAD; 153 if (dst->ops->ifdown) 154 dst->ops->ifdown(dst, dev, true); 155 dst->input = dst_discard; 156 dst->output = dst_discard_out; 157 dst->dev = blackhole_netdev; 158 netdev_ref_replace(dev, blackhole_netdev, &dst->dev_tracker, 159 GFP_ATOMIC); 160 } 161 EXPORT_SYMBOL(dst_dev_put); 162 163 void dst_release(struct dst_entry *dst) 164 { 165 if (dst) { 166 int newrefcnt; 167 168 newrefcnt = atomic_dec_return(&dst->__refcnt); 169 if (WARN_ONCE(newrefcnt < 0, "dst_release underflow")) 170 net_warn_ratelimited("%s: dst:%p refcnt:%d\n", 171 __func__, dst, newrefcnt); 172 if (!newrefcnt) 173 call_rcu_hurry(&dst->rcu_head, dst_destroy_rcu); 174 } 175 } 176 EXPORT_SYMBOL(dst_release); 177 178 void dst_release_immediate(struct dst_entry *dst) 179 { 180 if (dst) { 181 int newrefcnt; 182 183 newrefcnt = atomic_dec_return(&dst->__refcnt); 184 if (WARN_ONCE(newrefcnt < 0, "dst_release_immediate underflow")) 185 net_warn_ratelimited("%s: dst:%p refcnt:%d\n", 186 __func__, dst, newrefcnt); 187 if (!newrefcnt) 188 dst_destroy(dst); 189 } 190 } 191 EXPORT_SYMBOL(dst_release_immediate); 192 193 u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old) 194 { 195 struct dst_metrics *p = kmalloc(sizeof(*p), GFP_ATOMIC); 196 197 if (p) { 198 struct dst_metrics *old_p = (struct dst_metrics *)__DST_METRICS_PTR(old); 199 unsigned long prev, new; 200 201 refcount_set(&p->refcnt, 1); 202 memcpy(p->metrics, old_p->metrics, sizeof(p->metrics)); 203 204 new = (unsigned long) p; 205 prev = cmpxchg(&dst->_metrics, old, new); 206 207 if (prev != old) { 208 kfree(p); 209 p = (struct dst_metrics *)__DST_METRICS_PTR(prev); 210 if (prev & DST_METRICS_READ_ONLY) 211 p = NULL; 212 } else if (prev & DST_METRICS_REFCOUNTED) { 213 if (refcount_dec_and_test(&old_p->refcnt)) 214 kfree(old_p); 215 } 216 } 217 BUILD_BUG_ON(offsetof(struct dst_metrics, metrics) != 0); 218 return (u32 *)p; 219 } 220 EXPORT_SYMBOL(dst_cow_metrics_generic); 221 222 /* Caller asserts that dst_metrics_read_only(dst) is false. */ 223 void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old) 224 { 225 unsigned long prev, new; 226 227 new = ((unsigned long) &dst_default_metrics) | DST_METRICS_READ_ONLY; 228 prev = cmpxchg(&dst->_metrics, old, new); 229 if (prev == old) 230 kfree(__DST_METRICS_PTR(old)); 231 } 232 EXPORT_SYMBOL(__dst_destroy_metrics_generic); 233 234 struct dst_entry *dst_blackhole_check(struct dst_entry *dst, u32 cookie) 235 { 236 return NULL; 237 } 238 239 u32 *dst_blackhole_cow_metrics(struct dst_entry *dst, unsigned long old) 240 { 241 return NULL; 242 } 243 244 struct neighbour *dst_blackhole_neigh_lookup(const struct dst_entry *dst, 245 struct sk_buff *skb, 246 const void *daddr) 247 { 248 return NULL; 249 } 250 251 void dst_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 252 struct sk_buff *skb, u32 mtu, 253 bool confirm_neigh) 254 { 255 } 256 EXPORT_SYMBOL_GPL(dst_blackhole_update_pmtu); 257 258 void dst_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 259 struct sk_buff *skb) 260 { 261 } 262 EXPORT_SYMBOL_GPL(dst_blackhole_redirect); 263 264 unsigned int dst_blackhole_mtu(const struct dst_entry *dst) 265 { 266 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 267 268 return mtu ? : dst->dev->mtu; 269 } 270 EXPORT_SYMBOL_GPL(dst_blackhole_mtu); 271 272 static struct dst_ops dst_blackhole_ops = { 273 .family = AF_UNSPEC, 274 .neigh_lookup = dst_blackhole_neigh_lookup, 275 .check = dst_blackhole_check, 276 .cow_metrics = dst_blackhole_cow_metrics, 277 .update_pmtu = dst_blackhole_update_pmtu, 278 .redirect = dst_blackhole_redirect, 279 .mtu = dst_blackhole_mtu, 280 }; 281 282 static void __metadata_dst_init(struct metadata_dst *md_dst, 283 enum metadata_type type, u8 optslen) 284 { 285 struct dst_entry *dst; 286 287 dst = &md_dst->dst; 288 dst_init(dst, &dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 289 DST_METADATA | DST_NOCOUNT); 290 memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst)); 291 md_dst->type = type; 292 } 293 294 struct metadata_dst *metadata_dst_alloc(u8 optslen, enum metadata_type type, 295 gfp_t flags) 296 { 297 struct metadata_dst *md_dst; 298 299 md_dst = kmalloc(sizeof(*md_dst) + optslen, flags); 300 if (!md_dst) 301 return NULL; 302 303 __metadata_dst_init(md_dst, type, optslen); 304 305 return md_dst; 306 } 307 EXPORT_SYMBOL_GPL(metadata_dst_alloc); 308 309 void metadata_dst_free(struct metadata_dst *md_dst) 310 { 311 #ifdef CONFIG_DST_CACHE 312 if (md_dst->type == METADATA_IP_TUNNEL) 313 dst_cache_destroy(&md_dst->u.tun_info.dst_cache); 314 #endif 315 if (md_dst->type == METADATA_XFRM) 316 dst_release(md_dst->u.xfrm_info.dst_orig); 317 kfree(md_dst); 318 } 319 EXPORT_SYMBOL_GPL(metadata_dst_free); 320 321 struct metadata_dst __percpu * 322 metadata_dst_alloc_percpu(u8 optslen, enum metadata_type type, gfp_t flags) 323 { 324 int cpu; 325 struct metadata_dst __percpu *md_dst; 326 327 md_dst = __alloc_percpu_gfp(sizeof(struct metadata_dst) + optslen, 328 __alignof__(struct metadata_dst), flags); 329 if (!md_dst) 330 return NULL; 331 332 for_each_possible_cpu(cpu) 333 __metadata_dst_init(per_cpu_ptr(md_dst, cpu), type, optslen); 334 335 return md_dst; 336 } 337 EXPORT_SYMBOL_GPL(metadata_dst_alloc_percpu); 338 339 void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst) 340 { 341 int cpu; 342 343 for_each_possible_cpu(cpu) { 344 struct metadata_dst *one_md_dst = per_cpu_ptr(md_dst, cpu); 345 346 #ifdef CONFIG_DST_CACHE 347 if (one_md_dst->type == METADATA_IP_TUNNEL) 348 dst_cache_destroy(&one_md_dst->u.tun_info.dst_cache); 349 #endif 350 if (one_md_dst->type == METADATA_XFRM) 351 dst_release(one_md_dst->u.xfrm_info.dst_orig); 352 } 353 free_percpu(md_dst); 354 } 355 EXPORT_SYMBOL_GPL(metadata_dst_free_percpu); 356