net/core/flow.c

   1 /* flow.c: Generic flow cache.
   2  *
   3  * Copyright (C) 2003 Alexey N. Kuznetsov (kuznet@ms2.inr.ac.ru)
   4  * Copyright (C) 2003 David S. Miller (davem@redhat.com)
   5  */
   6
   7 #include <linux/kernel.h>
   8 #include <linux/module.h>
   9 #include <linux/list.h>
  10 #include <linux/jhash.h>
  11 #include <linux/interrupt.h>
  12 #include <linux/mm.h>
  13 #include <linux/random.h>
  14 #include <linux/init.h>
  15 #include <linux/slab.h>
  16 #include <linux/smp.h>
  17 #include <linux/completion.h>
  18 #include <linux/percpu.h>
  19 #include <linux/bitops.h>
  20 #include <linux/notifier.h>
  21 #include <linux/cpu.h>
  22 #include <linux/cpumask.h>
  23 #include <linux/mutex.h>
  24 #include <net/flow.h>
  25 #include <linux/atomic.h>
  26 #include <linux/security.h>
  27 #include <net/net_namespace.h>
  28
  29 struct flow_cache_entry {
  30         union {
  31                 struct hlist_node       hlist;
  32                 struct list_head        gc_list;
  33         } u;
  34         struct net                      *net;
  35         u16                             family;
  36         u8                              dir;
  37         u32                             genid;
  38         struct flowi                    key;
  39         struct flow_cache_object        *object;
  40 };
  41
  42 struct flow_flush_info {
  43         struct flow_cache               *cache;
  44         atomic_t                        cpuleft;
  45         struct completion               completion;
  46 };
  47
  48 static struct kmem_cache *flow_cachep __read_mostly;
  49
  50 #define flow_cache_hash_size(cache)     (1U << (cache)->hash_shift)
  51 #define FLOW_HASH_RND_PERIOD            (10 * 60 * HZ)
  52
  53 static void flow_cache_new_hashrnd(unsigned long arg)
  54 {
  55         struct flow_cache *fc = (void *) arg;
  56         int i;
  57
  58         for_each_possible_cpu(i)
  59                 per_cpu_ptr(fc->percpu, i)->hash_rnd_recalc = 1;
  60
  61         fc->rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
  62         add_timer(&fc->rnd_timer);
  63 }
  64
  65 static int flow_entry_valid(struct flow_cache_entry *fle,
  66                                 struct netns_xfrm *xfrm)
  67 {
  68         if (atomic_read(&xfrm->flow_cache_genid) != fle->genid)
  69                 return 0;
  70         if (fle->object && !fle->object->ops->check(fle->object))
  71                 return 0;
  72         return 1;
  73 }
  74
  75 static void flow_entry_kill(struct flow_cache_entry *fle,
  76                                 struct netns_xfrm *xfrm)
  77 {
  78         if (fle->object)
  79                 fle->object->ops->delete(fle->object);
  80         kmem_cache_free(flow_cachep, fle);
  81 }
  82
  83 static void flow_cache_gc_task(struct work_struct *work)
  84 {
  85         struct list_head gc_list;
  86         struct flow_cache_entry *fce, *n;
  87         struct netns_xfrm *xfrm = container_of(work, struct netns_xfrm,
  88                                                 flow_cache_gc_work);
  89
  90         INIT_LIST_HEAD(&gc_list);
  91         spin_lock_bh(&xfrm->flow_cache_gc_lock);
  92         list_splice_tail_init(&xfrm->flow_cache_gc_list, &gc_list);
  93         spin_unlock_bh(&xfrm->flow_cache_gc_lock);
  94
  95         list_for_each_entry_safe(fce, n, &gc_list, u.gc_list) {
  96                 flow_entry_kill(fce, xfrm);
  97                 atomic_dec(&xfrm->flow_cache_gc_count);
  98         }
  99 }
 100
 101 static void flow_cache_queue_garbage(struct flow_cache_percpu *fcp,
 102                                      int deleted, struct list_head *gc_list,
 103                                      struct netns_xfrm *xfrm)
 104 {
 105         if (deleted) {
 106                 atomic_add(deleted, &xfrm->flow_cache_gc_count);
 107                 fcp->hash_count -= deleted;
 108                 spin_lock_bh(&xfrm->flow_cache_gc_lock);
 109                 list_splice_tail(gc_list, &xfrm->flow_cache_gc_list);
 110                 spin_unlock_bh(&xfrm->flow_cache_gc_lock);
 111                 schedule_work(&xfrm->flow_cache_gc_work);
 112         }
 113 }
 114
 115 static void __flow_cache_shrink(struct flow_cache *fc,
 116                                 struct flow_cache_percpu *fcp,
 117                                 int shrink_to)
 118 {
 119         struct flow_cache_entry *fle;
 120         struct hlist_node *tmp;
 121         LIST_HEAD(gc_list);
 122         int deleted = 0;
 123         struct netns_xfrm *xfrm = container_of(fc, struct netns_xfrm,
 124                                                 flow_cache_global);
 125         unsigned int i;
 126
 127         for (i = 0; i < flow_cache_hash_size(fc); i++) {
 128                 int saved = 0;
 129
 130                 hlist_for_each_entry_safe(fle, tmp,
 131                                           &fcp->hash_table[i], u.hlist) {
 132                         if (saved < shrink_to &&
 133                             flow_entry_valid(fle, xfrm)) {
 134                                 saved++;
 135                         } else {
 136                                 deleted++;
 137                                 hlist_del(&fle->u.hlist);
 138                                 list_add_tail(&fle->u.gc_list, &gc_list);
 139                         }
 140                 }
 141         }
 142
 143         flow_cache_queue_garbage(fcp, deleted, &gc_list, xfrm);
 144 }
 145
 146 static void flow_cache_shrink(struct flow_cache *fc,
 147                               struct flow_cache_percpu *fcp)
 148 {
 149         int shrink_to = fc->low_watermark / flow_cache_hash_size(fc);
 150
 151         __flow_cache_shrink(fc, fcp, shrink_to);
 152 }
 153
 154 static void flow_new_hash_rnd(struct flow_cache *fc,
 155                               struct flow_cache_percpu *fcp)
 156 {
 157         get_random_bytes(&fcp->hash_rnd, sizeof(u32));
 158         fcp->hash_rnd_recalc = 0;
 159         __flow_cache_shrink(fc, fcp, 0);
 160 }
 161
 162 static u32 flow_hash_code(struct flow_cache *fc,
 163                           struct flow_cache_percpu *fcp,
 164                           const struct flowi *key,
 165                           unsigned int keysize)
 166 {
 167         const u32 *k = (const u32 *) key;
 168         const u32 length = keysize * sizeof(flow_compare_t) / sizeof(u32);
 169
 170         return jhash2(k, length, fcp->hash_rnd)
 171                 & (flow_cache_hash_size(fc) - 1);
 172 }
 173
 174 /* I hear what you're saying, use memcmp.  But memcmp cannot make
 175  * important assumptions that we can here, such as alignment.
 176  */
 177 static int flow_key_compare(const struct flowi *key1, const struct flowi *key2,
 178                             unsigned int keysize)
 179 {
 180         const flow_compare_t *k1, *k1_lim, *k2;
 181
 182         k1 = (const flow_compare_t *) key1;
 183         k1_lim = k1 + keysize;
 184
 185         k2 = (const flow_compare_t *) key2;
 186
 187         do {
 188                 if (*k1++ != *k2++)
 189                         return 1;
 190         } while (k1 < k1_lim);
 191
 192         return 0;
 193 }
 194
 195 struct flow_cache_object *
 196 flow_cache_lookup(struct net *net, const struct flowi *key, u16 family, u8 dir,
 197                   flow_resolve_t resolver, void *ctx)
 198 {
 199         struct flow_cache *fc = &net->xfrm.flow_cache_global;
 200         struct flow_cache_percpu *fcp;
 201         struct flow_cache_entry *fle, *tfle;
 202         struct flow_cache_object *flo;
 203         unsigned int keysize;
 204         unsigned int hash;
 205
 206         local_bh_disable();
 207         fcp = this_cpu_ptr(fc->percpu);
 208
 209         fle = NULL;
 210         flo = NULL;
 211
 212         keysize = flow_key_size(family);
 213         if (!keysize)
 214                 goto nocache;
 215
 216         /* Packet really early in init?  Making flow_cache_init a
 217          * pre-smp initcall would solve this.  --RR */
 218         if (!fcp->hash_table)
 219                 goto nocache;
 220
 221         if (fcp->hash_rnd_recalc)
 222                 flow_new_hash_rnd(fc, fcp);
 223
 224         hash = flow_hash_code(fc, fcp, key, keysize);
 225         hlist_for_each_entry(tfle, &fcp->hash_table[hash], u.hlist) {
 226                 if (tfle->net == net &&
 227                     tfle->family == family &&
 228                     tfle->dir == dir &&
 229                     flow_key_compare(key, &tfle->key, keysize) == 0) {
 230                         fle = tfle;
 231                         break;
 232                 }
 233         }
 234
 235         if (unlikely(!fle)) {
 236                 if (fcp->hash_count > fc->high_watermark)
 237                         flow_cache_shrink(fc, fcp);
 238
 239                 if (atomic_read(&net->xfrm.flow_cache_gc_count) >
 240                     2 * num_online_cpus() * fc->high_watermark) {
 241                         flo = ERR_PTR(-ENOBUFS);
 242                         goto ret_object;
 243                 }
 244
 245                 fle = kmem_cache_alloc(flow_cachep, GFP_ATOMIC);
 246                 if (fle) {
 247                         fle->net = net;
 248                         fle->family = family;
 249                         fle->dir = dir;
 250                         memcpy(&fle->key, key, keysize * sizeof(flow_compare_t));
 251                         fle->object = NULL;
 252                         hlist_add_head(&fle->u.hlist, &fcp->hash_table[hash]);
 253                         fcp->hash_count++;
 254                 }
 255         } else if (likely(fle->genid == atomic_read(&net->xfrm.flow_cache_genid))) {
 256                 flo = fle->object;
 257                 if (!flo)
 258                         goto ret_object;
 259                 flo = flo->ops->get(flo);
 260                 if (flo)
 261                         goto ret_object;
 262         } else if (fle->object) {
 263                 flo = fle->object;
 264                 flo->ops->delete(flo);
 265                 fle->object = NULL;
 266         }
 267
 268 nocache:
 269         flo = NULL;
 270         if (fle) {
 271                 flo = fle->object;
 272                 fle->object = NULL;
 273         }
 274         flo = resolver(net, key, family, dir, flo, ctx);
 275         if (fle) {
 276                 fle->genid = atomic_read(&net->xfrm.flow_cache_genid);
 277                 if (!IS_ERR(flo))
 278                         fle->object = flo;
 279                 else
 280                         fle->genid--;
 281         } else {
 282                 if (!IS_ERR_OR_NULL(flo))
 283                         flo->ops->delete(flo);
 284         }
 285 ret_object:
 286         local_bh_enable();
 287         return flo;
 288 }
 289 EXPORT_SYMBOL(flow_cache_lookup);
 290
 291 static void flow_cache_flush_tasklet(unsigned long data)
 292 {
 293         struct flow_flush_info *info = (void *)data;
 294         struct flow_cache *fc = info->cache;
 295         struct flow_cache_percpu *fcp;
 296         struct flow_cache_entry *fle;
 297         struct hlist_node *tmp;
 298         LIST_HEAD(gc_list);
 299         int deleted = 0;
 300         struct netns_xfrm *xfrm = container_of(fc, struct netns_xfrm,
 301                                                 flow_cache_global);
 302         unsigned int i;
 303
 304         fcp = this_cpu_ptr(fc->percpu);
 305         for (i = 0; i < flow_cache_hash_size(fc); i++) {
 306                 hlist_for_each_entry_safe(fle, tmp,
 307                                           &fcp->hash_table[i], u.hlist) {
 308                         if (flow_entry_valid(fle, xfrm))
 309                                 continue;
 310
 311                         deleted++;
 312                         hlist_del(&fle->u.hlist);
 313                         list_add_tail(&fle->u.gc_list, &gc_list);
 314                 }
 315         }
 316
 317         flow_cache_queue_garbage(fcp, deleted, &gc_list, xfrm);
 318
 319         if (atomic_dec_and_test(&info->cpuleft))
 320                 complete(&info->completion);
 321 }
 322
 323 /*
 324  * Return whether a cpu needs flushing.  Conservatively, we assume
 325  * the presence of any entries means the core may require flushing,
 326  * since the flow_cache_ops.check() function may assume it's running
 327  * on the same core as the per-cpu cache component.
 328  */
 329 static int flow_cache_percpu_empty(struct flow_cache *fc, int cpu)
 330 {
 331         struct flow_cache_percpu *fcp;
 332         unsigned int i;
 333
 334         fcp = per_cpu_ptr(fc->percpu, cpu);
 335         for (i = 0; i < flow_cache_hash_size(fc); i++)
 336                 if (!hlist_empty(&fcp->hash_table[i]))
 337                         return 0;
 338         return 1;
 339 }
 340
 341 static void flow_cache_flush_per_cpu(void *data)
 342 {
 343         struct flow_flush_info *info = data;
 344         struct tasklet_struct *tasklet;
 345
 346         tasklet = &this_cpu_ptr(info->cache->percpu)->flush_tasklet;
 347         tasklet->data = (unsigned long)info;
 348         tasklet_schedule(tasklet);
 349 }
 350
 351 void flow_cache_flush(struct net *net)
 352 {
 353         struct flow_flush_info info;
 354         cpumask_var_t mask;
 355         int i, self;
 356
 357         /* Track which cpus need flushing to avoid disturbing all cores. */
 358         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 359                 return;
 360         cpumask_clear(mask);
 361
 362         /* Don't want cpus going down or up during this. */
 363         get_online_cpus();
 364         mutex_lock(&net->xfrm.flow_flush_sem);
 365         info.cache = &net->xfrm.flow_cache_global;
 366         for_each_online_cpu(i)
 367                 if (!flow_cache_percpu_empty(info.cache, i))
 368                         cpumask_set_cpu(i, mask);
 369         atomic_set(&info.cpuleft, cpumask_weight(mask));
 370         if (atomic_read(&info.cpuleft) == 0)
 371                 goto done;
 372
 373         init_completion(&info.completion);
 374
 375         local_bh_disable();
 376         self = cpumask_test_and_clear_cpu(smp_processor_id(), mask);
 377         on_each_cpu_mask(mask, flow_cache_flush_per_cpu, &info, 0);
 378         if (self)
 379                 flow_cache_flush_tasklet((unsigned long)&info);
 380         local_bh_enable();
 381
 382         wait_for_completion(&info.completion);
 383
 384 done:
 385         mutex_unlock(&net->xfrm.flow_flush_sem);
 386         put_online_cpus();
 387         free_cpumask_var(mask);
 388 }
 389
 390 static void flow_cache_flush_task(struct work_struct *work)
 391 {
 392         struct netns_xfrm *xfrm = container_of(work, struct netns_xfrm,
 393                                                 flow_cache_flush_work);
 394         struct net *net = container_of(xfrm, struct net, xfrm);
 395
 396         flow_cache_flush(net);
 397 }
 398
 399 void flow_cache_flush_deferred(struct net *net)
 400 {
 401         schedule_work(&net->xfrm.flow_cache_flush_work);
 402 }
 403
 404 static int flow_cache_cpu_prepare(struct flow_cache *fc, int cpu)
 405 {
 406         struct flow_cache_percpu *fcp = per_cpu_ptr(fc->percpu, cpu);
 407         unsigned int sz = sizeof(struct hlist_head) * flow_cache_hash_size(fc);
 408
 409         if (!fcp->hash_table) {
 410                 fcp->hash_table = kzalloc_node(sz, GFP_KERNEL, cpu_to_node(cpu));
 411                 if (!fcp->hash_table) {
 412                         pr_err("NET: failed to allocate flow cache sz %u\n", sz);
 413                         return -ENOMEM;
 414                 }
 415                 fcp->hash_rnd_recalc = 1;
 416                 fcp->hash_count = 0;
 417                 tasklet_init(&fcp->flush_tasklet, flow_cache_flush_tasklet, 0);
 418         }
 419         return 0;
 420 }
 421
 422 static int flow_cache_cpu_up_prep(unsigned int cpu, struct hlist_node *node)
 423 {
 424         struct flow_cache *fc = hlist_entry_safe(node, struct flow_cache, node);
 425
 426         return flow_cache_cpu_prepare(fc, cpu);
 427 }
 428
 429 static int flow_cache_cpu_dead(unsigned int cpu, struct hlist_node *node)
 430 {
 431         struct flow_cache *fc = hlist_entry_safe(node, struct flow_cache, node);
 432         struct flow_cache_percpu *fcp = per_cpu_ptr(fc->percpu, cpu);
 433
 434         __flow_cache_shrink(fc, fcp, 0);
 435         return 0;
 436 }
 437
 438 int flow_cache_init(struct net *net)
 439 {
 440         int i;
 441         struct flow_cache *fc = &net->xfrm.flow_cache_global;
 442
 443         if (!flow_cachep)
 444                 flow_cachep = kmem_cache_create("flow_cache",
 445                                                 sizeof(struct flow_cache_entry),
 446                                                 0, SLAB_PANIC, NULL);
 447         spin_lock_init(&net->xfrm.flow_cache_gc_lock);
 448         INIT_LIST_HEAD(&net->xfrm.flow_cache_gc_list);
 449         INIT_WORK(&net->xfrm.flow_cache_gc_work, flow_cache_gc_task);
 450         INIT_WORK(&net->xfrm.flow_cache_flush_work, flow_cache_flush_task);
 451         mutex_init(&net->xfrm.flow_flush_sem);
 452         atomic_set(&net->xfrm.flow_cache_gc_count, 0);
 453
 454         fc->hash_shift = 10;
 455         fc->low_watermark = 2 * flow_cache_hash_size(fc);
 456         fc->high_watermark = 4 * flow_cache_hash_size(fc);
 457
 458         fc->percpu = alloc_percpu(struct flow_cache_percpu);
 459         if (!fc->percpu)
 460                 return -ENOMEM;
 461
 462         if (cpuhp_state_add_instance(CPUHP_NET_FLOW_PREPARE, &fc->node))
 463                 goto err;
 464
 465         setup_timer(&fc->rnd_timer, flow_cache_new_hashrnd,
 466                     (unsigned long) fc);
 467         fc->rnd_timer.expires = jiffies + FLOW_HASH_RND_PERIOD;
 468         add_timer(&fc->rnd_timer);
 469
 470         return 0;
 471
 472 err:
 473         for_each_possible_cpu(i) {
 474                 struct flow_cache_percpu *fcp = per_cpu_ptr(fc->percpu, i);
 475                 kfree(fcp->hash_table);
 476                 fcp->hash_table = NULL;
 477         }
 478
 479         free_percpu(fc->percpu);
 480         fc->percpu = NULL;
 481
 482         return -ENOMEM;
 483 }
 484 EXPORT_SYMBOL(flow_cache_init);
 485
 486 void flow_cache_fini(struct net *net)
 487 {
 488         int i;
 489         struct flow_cache *fc = &net->xfrm.flow_cache_global;
 490
 491         del_timer_sync(&fc->rnd_timer);
 492
 493         cpuhp_state_remove_instance_nocalls(CPUHP_NET_FLOW_PREPARE, &fc->node);
 494
 495         for_each_possible_cpu(i) {
 496                 struct flow_cache_percpu *fcp = per_cpu_ptr(fc->percpu, i);
 497                 kfree(fcp->hash_table);
 498                 fcp->hash_table = NULL;
 499         }
 500
 501         free_percpu(fc->percpu);
 502         fc->percpu = NULL;
 503 }
 504 EXPORT_SYMBOL(flow_cache_fini);
 505
 506 void __init flow_cache_hp_init(void)
 507 {
 508         int ret;
 509
 510         ret = cpuhp_setup_state_multi(CPUHP_NET_FLOW_PREPARE,
 511                                       "net/flow:prepare",
 512                                       flow_cache_cpu_up_prep,
 513                                       flow_cache_cpu_dead);
 514         WARN_ON(ret < 0);
 515 }