~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/netfilter/nf_conntrack_core.c

Version: ~ [ linux-5.11 ] ~ [ linux-5.10.17 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.99 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.176 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.221 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.257 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.257 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /* Connection state tracking for netfilter.  This is separated from,
  2    but required by, the NAT layer; it can also be used by an iptables
  3    extension. */
  4 
  5 /* (C) 1999-2001 Paul `Rusty' Russell
  6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
  7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
  8  * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
  9  *
 10  * This program is free software; you can redistribute it and/or modify
 11  * it under the terms of the GNU General Public License version 2 as
 12  * published by the Free Software Foundation.
 13  */
 14 
 15 #include <linux/types.h>
 16 #include <linux/netfilter.h>
 17 #include <linux/module.h>
 18 #include <linux/sched.h>
 19 #include <linux/skbuff.h>
 20 #include <linux/proc_fs.h>
 21 #include <linux/vmalloc.h>
 22 #include <linux/stddef.h>
 23 #include <linux/slab.h>
 24 #include <linux/random.h>
 25 #include <linux/jhash.h>
 26 #include <linux/err.h>
 27 #include <linux/percpu.h>
 28 #include <linux/moduleparam.h>
 29 #include <linux/notifier.h>
 30 #include <linux/kernel.h>
 31 #include <linux/netdevice.h>
 32 #include <linux/socket.h>
 33 #include <linux/mm.h>
 34 #include <linux/nsproxy.h>
 35 #include <linux/rculist_nulls.h>
 36 
 37 #include <net/netfilter/nf_conntrack.h>
 38 #include <net/netfilter/nf_conntrack_l3proto.h>
 39 #include <net/netfilter/nf_conntrack_l4proto.h>
 40 #include <net/netfilter/nf_conntrack_expect.h>
 41 #include <net/netfilter/nf_conntrack_helper.h>
 42 #include <net/netfilter/nf_conntrack_seqadj.h>
 43 #include <net/netfilter/nf_conntrack_core.h>
 44 #include <net/netfilter/nf_conntrack_extend.h>
 45 #include <net/netfilter/nf_conntrack_acct.h>
 46 #include <net/netfilter/nf_conntrack_ecache.h>
 47 #include <net/netfilter/nf_conntrack_zones.h>
 48 #include <net/netfilter/nf_conntrack_timestamp.h>
 49 #include <net/netfilter/nf_conntrack_timeout.h>
 50 #include <net/netfilter/nf_conntrack_labels.h>
 51 #include <net/netfilter/nf_conntrack_synproxy.h>
 52 #include <net/netfilter/nf_nat.h>
 53 #include <net/netfilter/nf_nat_core.h>
 54 #include <net/netfilter/nf_nat_helper.h>
 55 
 56 #define NF_CONNTRACK_VERSION    "0.5.0"
 57 
 58 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
 59                                       enum nf_nat_manip_type manip,
 60                                       const struct nlattr *attr) __read_mostly;
 61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
 62 
 63 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
 64                               struct nf_conn *ct,
 65                               enum ip_conntrack_info ctinfo,
 66                               unsigned int protoff);
 67 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
 68 
 69 DEFINE_SPINLOCK(nf_conntrack_lock);
 70 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 71 
 72 unsigned int nf_conntrack_htable_size __read_mostly;
 73 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 74 
 75 unsigned int nf_conntrack_max __read_mostly;
 76 EXPORT_SYMBOL_GPL(nf_conntrack_max);
 77 
 78 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
 79 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
 80 
 81 unsigned int nf_conntrack_hash_rnd __read_mostly;
 82 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
 83 
 84 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
 85 {
 86         unsigned int n;
 87 
 88         /* The direction must be ignored, so we hash everything up to the
 89          * destination ports (which is a multiple of 4) and treat the last
 90          * three bytes manually.
 91          */
 92         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
 93         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
 94                       (((__force __u16)tuple->dst.u.all << 16) |
 95                       tuple->dst.protonum));
 96 }
 97 
 98 static u32 __hash_bucket(u32 hash, unsigned int size)
 99 {
100         return ((u64)hash * size) >> 32;
101 }
102 
103 static u32 hash_bucket(u32 hash, const struct net *net)
104 {
105         return __hash_bucket(hash, net->ct.htable_size);
106 }
107 
108 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
109                                   u16 zone, unsigned int size)
110 {
111         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
112 }
113 
114 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
115                                        const struct nf_conntrack_tuple *tuple)
116 {
117         return __hash_conntrack(tuple, zone, net->ct.htable_size);
118 }
119 
120 bool
121 nf_ct_get_tuple(const struct sk_buff *skb,
122                 unsigned int nhoff,
123                 unsigned int dataoff,
124                 u_int16_t l3num,
125                 u_int8_t protonum,
126                 struct nf_conntrack_tuple *tuple,
127                 const struct nf_conntrack_l3proto *l3proto,
128                 const struct nf_conntrack_l4proto *l4proto)
129 {
130         memset(tuple, 0, sizeof(*tuple));
131 
132         tuple->src.l3num = l3num;
133         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
134                 return false;
135 
136         tuple->dst.protonum = protonum;
137         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
138 
139         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
140 }
141 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
142 
143 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
144                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
145 {
146         struct nf_conntrack_l3proto *l3proto;
147         struct nf_conntrack_l4proto *l4proto;
148         unsigned int protoff;
149         u_int8_t protonum;
150         int ret;
151 
152         rcu_read_lock();
153 
154         l3proto = __nf_ct_l3proto_find(l3num);
155         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
156         if (ret != NF_ACCEPT) {
157                 rcu_read_unlock();
158                 return false;
159         }
160 
161         l4proto = __nf_ct_l4proto_find(l3num, protonum);
162 
163         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
164                               l3proto, l4proto);
165 
166         rcu_read_unlock();
167         return ret;
168 }
169 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
170 
171 bool
172 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
173                    const struct nf_conntrack_tuple *orig,
174                    const struct nf_conntrack_l3proto *l3proto,
175                    const struct nf_conntrack_l4proto *l4proto)
176 {
177         memset(inverse, 0, sizeof(*inverse));
178 
179         inverse->src.l3num = orig->src.l3num;
180         if (l3proto->invert_tuple(inverse, orig) == 0)
181                 return false;
182 
183         inverse->dst.dir = !orig->dst.dir;
184 
185         inverse->dst.protonum = orig->dst.protonum;
186         return l4proto->invert_tuple(inverse, orig);
187 }
188 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
189 
190 static void
191 clean_from_lists(struct nf_conn *ct)
192 {
193         pr_debug("clean_from_lists(%p)\n", ct);
194         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
195         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
196 
197         /* Destroy all pending expectations */
198         nf_ct_remove_expectations(ct);
199 }
200 
201 static void
202 destroy_conntrack(struct nf_conntrack *nfct)
203 {
204         struct nf_conn *ct = (struct nf_conn *)nfct;
205         struct net *net = nf_ct_net(ct);
206         struct nf_conntrack_l4proto *l4proto;
207 
208         pr_debug("destroy_conntrack(%p)\n", ct);
209         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
210         NF_CT_ASSERT(!timer_pending(&ct->timeout));
211 
212         /* To make sure we don't get any weird locking issues here:
213          * destroy_conntrack() MUST NOT be called with a write lock
214          * to nf_conntrack_lock!!! -HW */
215         rcu_read_lock();
216         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
217         if (l4proto && l4proto->destroy)
218                 l4proto->destroy(ct);
219 
220         rcu_read_unlock();
221 
222         spin_lock_bh(&nf_conntrack_lock);
223         /* Expectations will have been removed in clean_from_lists,
224          * except TFTP can create an expectation on the first packet,
225          * before connection is in the list, so we need to clean here,
226          * too. */
227         nf_ct_remove_expectations(ct);
228 
229         /* We overload first tuple to link into unconfirmed or dying list.*/
230         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
231         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
232 
233         NF_CT_STAT_INC(net, delete);
234         spin_unlock_bh(&nf_conntrack_lock);
235 
236         if (ct->master)
237                 nf_ct_put(ct->master);
238 
239         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
240         nf_conntrack_free(ct);
241 }
242 
243 static void nf_ct_delete_from_lists(struct nf_conn *ct)
244 {
245         struct net *net = nf_ct_net(ct);
246 
247         nf_ct_helper_destroy(ct);
248         spin_lock_bh(&nf_conntrack_lock);
249         /* Inside lock so preempt is disabled on module removal path.
250          * Otherwise we can get spurious warnings. */
251         NF_CT_STAT_INC(net, delete_list);
252         clean_from_lists(ct);
253         /* add this conntrack to the dying list */
254         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
255                              &net->ct.dying);
256         spin_unlock_bh(&nf_conntrack_lock);
257 }
258 
259 static void death_by_event(unsigned long ul_conntrack)
260 {
261         struct nf_conn *ct = (void *)ul_conntrack;
262         struct net *net = nf_ct_net(ct);
263         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
264 
265         BUG_ON(ecache == NULL);
266 
267         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
268                 /* bad luck, let's retry again */
269                 ecache->timeout.expires = jiffies +
270                         (prandom_u32() % net->ct.sysctl_events_retry_timeout);
271                 add_timer(&ecache->timeout);
272                 return;
273         }
274         /* we've got the event delivered, now it's dying */
275         set_bit(IPS_DYING_BIT, &ct->status);
276         nf_ct_put(ct);
277 }
278 
279 static void nf_ct_dying_timeout(struct nf_conn *ct)
280 {
281         struct net *net = nf_ct_net(ct);
282         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
283 
284         BUG_ON(ecache == NULL);
285 
286         /* set a new timer to retry event delivery */
287         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
288         ecache->timeout.expires = jiffies +
289                 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
290         add_timer(&ecache->timeout);
291 }
292 
293 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
294 {
295         struct nf_conn_tstamp *tstamp;
296 
297         tstamp = nf_conn_tstamp_find(ct);
298         if (tstamp && tstamp->stop == 0)
299                 tstamp->stop = ktime_to_ns(ktime_get_real());
300 
301         if (!nf_ct_is_dying(ct) &&
302             unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
303             portid, report) < 0)) {
304                 /* destroy event was not delivered */
305                 nf_ct_delete_from_lists(ct);
306                 nf_ct_dying_timeout(ct);
307                 return false;
308         }
309         set_bit(IPS_DYING_BIT, &ct->status);
310         nf_ct_delete_from_lists(ct);
311         nf_ct_put(ct);
312         return true;
313 }
314 EXPORT_SYMBOL_GPL(nf_ct_delete);
315 
316 static void death_by_timeout(unsigned long ul_conntrack)
317 {
318         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
319 }
320 
321 /*
322  * Warning :
323  * - Caller must take a reference on returned object
324  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
325  * OR
326  * - Caller must lock nf_conntrack_lock before calling this function
327  */
328 static struct nf_conntrack_tuple_hash *
329 ____nf_conntrack_find(struct net *net, u16 zone,
330                       const struct nf_conntrack_tuple *tuple, u32 hash)
331 {
332         struct nf_conntrack_tuple_hash *h;
333         struct hlist_nulls_node *n;
334         unsigned int bucket = hash_bucket(hash, net);
335 
336         /* Disable BHs the entire time since we normally need to disable them
337          * at least once for the stats anyway.
338          */
339         local_bh_disable();
340 begin:
341         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
342                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
343                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
344                         NF_CT_STAT_INC(net, found);
345                         local_bh_enable();
346                         return h;
347                 }
348                 NF_CT_STAT_INC(net, searched);
349         }
350         /*
351          * if the nulls value we got at the end of this lookup is
352          * not the expected one, we must restart lookup.
353          * We probably met an item that was moved to another chain.
354          */
355         if (get_nulls_value(n) != bucket) {
356                 NF_CT_STAT_INC(net, search_restart);
357                 goto begin;
358         }
359         local_bh_enable();
360 
361         return NULL;
362 }
363 
364 struct nf_conntrack_tuple_hash *
365 __nf_conntrack_find(struct net *net, u16 zone,
366                     const struct nf_conntrack_tuple *tuple)
367 {
368         return ____nf_conntrack_find(net, zone, tuple,
369                                      hash_conntrack_raw(tuple, zone));
370 }
371 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
372 
373 /* Find a connection corresponding to a tuple. */
374 static struct nf_conntrack_tuple_hash *
375 __nf_conntrack_find_get(struct net *net, u16 zone,
376                         const struct nf_conntrack_tuple *tuple, u32 hash)
377 {
378         struct nf_conntrack_tuple_hash *h;
379         struct nf_conn *ct;
380 
381         rcu_read_lock();
382 begin:
383         h = ____nf_conntrack_find(net, zone, tuple, hash);
384         if (h) {
385                 ct = nf_ct_tuplehash_to_ctrack(h);
386                 if (unlikely(nf_ct_is_dying(ct) ||
387                              !atomic_inc_not_zero(&ct->ct_general.use)))
388                         h = NULL;
389                 else {
390                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
391                                      nf_ct_zone(ct) != zone)) {
392                                 nf_ct_put(ct);
393                                 goto begin;
394                         }
395                 }
396         }
397         rcu_read_unlock();
398 
399         return h;
400 }
401 
402 struct nf_conntrack_tuple_hash *
403 nf_conntrack_find_get(struct net *net, u16 zone,
404                       const struct nf_conntrack_tuple *tuple)
405 {
406         return __nf_conntrack_find_get(net, zone, tuple,
407                                        hash_conntrack_raw(tuple, zone));
408 }
409 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
410 
411 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
412                                        unsigned int hash,
413                                        unsigned int repl_hash)
414 {
415         struct net *net = nf_ct_net(ct);
416 
417         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
418                            &net->ct.hash[hash]);
419         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
420                            &net->ct.hash[repl_hash]);
421 }
422 
423 int
424 nf_conntrack_hash_check_insert(struct nf_conn *ct)
425 {
426         struct net *net = nf_ct_net(ct);
427         unsigned int hash, repl_hash;
428         struct nf_conntrack_tuple_hash *h;
429         struct hlist_nulls_node *n;
430         u16 zone;
431 
432         zone = nf_ct_zone(ct);
433         hash = hash_conntrack(net, zone,
434                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
435         repl_hash = hash_conntrack(net, zone,
436                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
437 
438         spin_lock_bh(&nf_conntrack_lock);
439 
440         /* See if there's one in the list already, including reverse */
441         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
442                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
443                                       &h->tuple) &&
444                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
445                         goto out;
446         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
447                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
448                                       &h->tuple) &&
449                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
450                         goto out;
451 
452         add_timer(&ct->timeout);
453         nf_conntrack_get(&ct->ct_general);
454         __nf_conntrack_hash_insert(ct, hash, repl_hash);
455         NF_CT_STAT_INC(net, insert);
456         spin_unlock_bh(&nf_conntrack_lock);
457 
458         return 0;
459 
460 out:
461         NF_CT_STAT_INC(net, insert_failed);
462         spin_unlock_bh(&nf_conntrack_lock);
463         return -EEXIST;
464 }
465 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
466 
467 /* Confirm a connection given skb; places it in hash table */
468 int
469 __nf_conntrack_confirm(struct sk_buff *skb)
470 {
471         unsigned int hash, repl_hash;
472         struct nf_conntrack_tuple_hash *h;
473         struct nf_conn *ct;
474         struct nf_conn_help *help;
475         struct nf_conn_tstamp *tstamp;
476         struct hlist_nulls_node *n;
477         enum ip_conntrack_info ctinfo;
478         struct net *net;
479         u16 zone;
480 
481         ct = nf_ct_get(skb, &ctinfo);
482         net = nf_ct_net(ct);
483 
484         /* ipt_REJECT uses nf_conntrack_attach to attach related
485            ICMP/TCP RST packets in other direction.  Actual packet
486            which created connection will be IP_CT_NEW or for an
487            expected connection, IP_CT_RELATED. */
488         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
489                 return NF_ACCEPT;
490 
491         zone = nf_ct_zone(ct);
492         /* reuse the hash saved before */
493         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
494         hash = hash_bucket(hash, net);
495         repl_hash = hash_conntrack(net, zone,
496                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
497 
498         /* We're not in hash table, and we refuse to set up related
499            connections for unconfirmed conns.  But packet copies and
500            REJECT will give spurious warnings here. */
501         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
502 
503         /* No external references means no one else could have
504            confirmed us. */
505         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
506         pr_debug("Confirming conntrack %p\n", ct);
507 
508         spin_lock_bh(&nf_conntrack_lock);
509 
510         /* We have to check the DYING flag inside the lock to prevent
511            a race against nf_ct_get_next_corpse() possibly called from
512            user context, else we insert an already 'dead' hash, blocking
513            further use of that particular connection -JM */
514 
515         if (unlikely(nf_ct_is_dying(ct))) {
516                 spin_unlock_bh(&nf_conntrack_lock);
517                 return NF_ACCEPT;
518         }
519 
520         /* See if there's one in the list already, including reverse:
521            NAT could have grabbed it without realizing, since we're
522            not in the hash.  If there is, we lost race. */
523         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
524                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
525                                       &h->tuple) &&
526                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
527                         goto out;
528         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
529                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
530                                       &h->tuple) &&
531                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
532                         goto out;
533 
534         /* Remove from unconfirmed list */
535         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
536 
537         /* Timer relative to confirmation time, not original
538            setting time, otherwise we'd get timer wrap in
539            weird delay cases. */
540         ct->timeout.expires += jiffies;
541         add_timer(&ct->timeout);
542         atomic_inc(&ct->ct_general.use);
543         ct->status |= IPS_CONFIRMED;
544 
545         /* set conntrack timestamp, if enabled. */
546         tstamp = nf_conn_tstamp_find(ct);
547         if (tstamp) {
548                 if (skb->tstamp.tv64 == 0)
549                         __net_timestamp(skb);
550 
551                 tstamp->start = ktime_to_ns(skb->tstamp);
552         }
553         /* Since the lookup is lockless, hash insertion must be done after
554          * starting the timer and setting the CONFIRMED bit. The RCU barriers
555          * guarantee that no other CPU can find the conntrack before the above
556          * stores are visible.
557          */
558         __nf_conntrack_hash_insert(ct, hash, repl_hash);
559         NF_CT_STAT_INC(net, insert);
560         spin_unlock_bh(&nf_conntrack_lock);
561 
562         help = nfct_help(ct);
563         if (help && help->helper)
564                 nf_conntrack_event_cache(IPCT_HELPER, ct);
565 
566         nf_conntrack_event_cache(master_ct(ct) ?
567                                  IPCT_RELATED : IPCT_NEW, ct);
568         return NF_ACCEPT;
569 
570 out:
571         NF_CT_STAT_INC(net, insert_failed);
572         spin_unlock_bh(&nf_conntrack_lock);
573         return NF_DROP;
574 }
575 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
576 
577 /* Returns true if a connection correspondings to the tuple (required
578    for NAT). */
579 int
580 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
581                          const struct nf_conn *ignored_conntrack)
582 {
583         struct net *net = nf_ct_net(ignored_conntrack);
584         struct nf_conntrack_tuple_hash *h;
585         struct hlist_nulls_node *n;
586         struct nf_conn *ct;
587         u16 zone = nf_ct_zone(ignored_conntrack);
588         unsigned int hash = hash_conntrack(net, zone, tuple);
589 
590         /* Disable BHs the entire time since we need to disable them at
591          * least once for the stats anyway.
592          */
593         rcu_read_lock_bh();
594         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
595                 ct = nf_ct_tuplehash_to_ctrack(h);
596                 if (ct != ignored_conntrack &&
597                     nf_ct_tuple_equal(tuple, &h->tuple) &&
598                     nf_ct_zone(ct) == zone) {
599                         NF_CT_STAT_INC(net, found);
600                         rcu_read_unlock_bh();
601                         return 1;
602                 }
603                 NF_CT_STAT_INC(net, searched);
604         }
605         rcu_read_unlock_bh();
606 
607         return 0;
608 }
609 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
610 
611 #define NF_CT_EVICTION_RANGE    8
612 
613 /* There's a small race here where we may free a just-assured
614    connection.  Too bad: we're in trouble anyway. */
615 static noinline int early_drop(struct net *net, unsigned int hash)
616 {
617         /* Use oldest entry, which is roughly LRU */
618         struct nf_conntrack_tuple_hash *h;
619         struct nf_conn *ct = NULL, *tmp;
620         struct hlist_nulls_node *n;
621         unsigned int i, cnt = 0;
622         int dropped = 0;
623 
624         rcu_read_lock();
625         for (i = 0; i < net->ct.htable_size; i++) {
626                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
627                                          hnnode) {
628                         tmp = nf_ct_tuplehash_to_ctrack(h);
629                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
630                                 ct = tmp;
631                         cnt++;
632                 }
633 
634                 if (ct != NULL) {
635                         if (likely(!nf_ct_is_dying(ct) &&
636                                    atomic_inc_not_zero(&ct->ct_general.use)))
637                                 break;
638                         else
639                                 ct = NULL;
640                 }
641 
642                 if (cnt >= NF_CT_EVICTION_RANGE)
643                         break;
644 
645                 hash = (hash + 1) % net->ct.htable_size;
646         }
647         rcu_read_unlock();
648 
649         if (!ct)
650                 return dropped;
651 
652         if (del_timer(&ct->timeout)) {
653                 if (nf_ct_delete(ct, 0, 0)) {
654                         dropped = 1;
655                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
656                 }
657         }
658         nf_ct_put(ct);
659         return dropped;
660 }
661 
662 void init_nf_conntrack_hash_rnd(void)
663 {
664         unsigned int rand;
665 
666         /*
667          * Why not initialize nf_conntrack_rnd in a "init()" function ?
668          * Because there isn't enough entropy when system initializing,
669          * and we initialize it as late as possible.
670          */
671         do {
672                 get_random_bytes(&rand, sizeof(rand));
673         } while (!rand);
674         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
675 }
676 
677 static struct nf_conn *
678 __nf_conntrack_alloc(struct net *net, u16 zone,
679                      const struct nf_conntrack_tuple *orig,
680                      const struct nf_conntrack_tuple *repl,
681                      gfp_t gfp, u32 hash)
682 {
683         struct nf_conn *ct;
684 
685         if (unlikely(!nf_conntrack_hash_rnd)) {
686                 init_nf_conntrack_hash_rnd();
687                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
688                 hash = hash_conntrack_raw(orig, zone);
689         }
690 
691         /* We don't want any race condition at early drop stage */
692         atomic_inc(&net->ct.count);
693 
694         if (nf_conntrack_max &&
695             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
696                 if (!early_drop(net, hash_bucket(hash, net))) {
697                         atomic_dec(&net->ct.count);
698                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
699                         return ERR_PTR(-ENOMEM);
700                 }
701         }
702 
703         /*
704          * Do not use kmem_cache_zalloc(), as this cache uses
705          * SLAB_DESTROY_BY_RCU.
706          */
707         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
708         if (ct == NULL) {
709                 atomic_dec(&net->ct.count);
710                 return ERR_PTR(-ENOMEM);
711         }
712         /*
713          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
714          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
715          */
716         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
717                offsetof(struct nf_conn, proto) -
718                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
719         spin_lock_init(&ct->lock);
720         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
721         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
722         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
723         /* save hash for reusing when confirming */
724         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
725         /* Don't set timer yet: wait for confirmation */
726         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
727         write_pnet(&ct->ct_net, net);
728 #ifdef CONFIG_NF_CONNTRACK_ZONES
729         if (zone) {
730                 struct nf_conntrack_zone *nf_ct_zone;
731 
732                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
733                 if (!nf_ct_zone)
734                         goto out_free;
735                 nf_ct_zone->id = zone;
736         }
737 #endif
738         /*
739          * changes to lookup keys must be done before setting refcnt to 1
740          */
741         smp_wmb();
742         atomic_set(&ct->ct_general.use, 1);
743         return ct;
744 
745 #ifdef CONFIG_NF_CONNTRACK_ZONES
746 out_free:
747         atomic_dec(&net->ct.count);
748         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
749         return ERR_PTR(-ENOMEM);
750 #endif
751 }
752 
753 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
754                                    const struct nf_conntrack_tuple *orig,
755                                    const struct nf_conntrack_tuple *repl,
756                                    gfp_t gfp)
757 {
758         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
759 }
760 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
761 
762 void nf_conntrack_free(struct nf_conn *ct)
763 {
764         struct net *net = nf_ct_net(ct);
765 
766         nf_ct_ext_destroy(ct);
767         nf_ct_ext_free(ct);
768         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
769         smp_mb__before_atomic_dec();
770         atomic_dec(&net->ct.count);
771 }
772 EXPORT_SYMBOL_GPL(nf_conntrack_free);
773 
774 
775 /* Allocate a new conntrack: we return -ENOMEM if classification
776    failed due to stress.  Otherwise it really is unclassifiable. */
777 static struct nf_conntrack_tuple_hash *
778 init_conntrack(struct net *net, struct nf_conn *tmpl,
779                const struct nf_conntrack_tuple *tuple,
780                struct nf_conntrack_l3proto *l3proto,
781                struct nf_conntrack_l4proto *l4proto,
782                struct sk_buff *skb,
783                unsigned int dataoff, u32 hash)
784 {
785         struct nf_conn *ct;
786         struct nf_conn_help *help;
787         struct nf_conntrack_tuple repl_tuple;
788         struct nf_conntrack_ecache *ecache;
789         struct nf_conntrack_expect *exp;
790         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
791         struct nf_conn_timeout *timeout_ext;
792         unsigned int *timeouts;
793 
794         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
795                 pr_debug("Can't invert tuple.\n");
796                 return NULL;
797         }
798 
799         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
800                                   hash);
801         if (IS_ERR(ct))
802                 return (struct nf_conntrack_tuple_hash *)ct;
803 
804         if (tmpl && nfct_synproxy(tmpl)) {
805                 nfct_seqadj_ext_add(ct);
806                 nfct_synproxy_ext_add(ct);
807         }
808 
809         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
810         if (timeout_ext)
811                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
812         else
813                 timeouts = l4proto->get_timeouts(net);
814 
815         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
816                 nf_conntrack_free(ct);
817                 pr_debug("init conntrack: can't track with proto module\n");
818                 return NULL;
819         }
820 
821         if (timeout_ext)
822                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
823 
824         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
825         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
826         nf_ct_labels_ext_add(ct);
827 
828         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
829         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
830                                  ecache ? ecache->expmask : 0,
831                              GFP_ATOMIC);
832 
833         spin_lock_bh(&nf_conntrack_lock);
834         exp = nf_ct_find_expectation(net, zone, tuple);
835         if (exp) {
836                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
837                          ct, exp);
838                 /* Welcome, Mr. Bond.  We've been expecting you... */
839                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
840                 ct->master = exp->master;
841                 if (exp->helper) {
842                         help = nf_ct_helper_ext_add(ct, exp->helper,
843                                                     GFP_ATOMIC);
844                         if (help)
845                                 rcu_assign_pointer(help->helper, exp->helper);
846                 }
847 
848 #ifdef CONFIG_NF_CONNTRACK_MARK
849                 ct->mark = exp->master->mark;
850 #endif
851 #ifdef CONFIG_NF_CONNTRACK_SECMARK
852                 ct->secmark = exp->master->secmark;
853 #endif
854                 nf_conntrack_get(&ct->master->ct_general);
855                 NF_CT_STAT_INC(net, expect_new);
856         } else {
857                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
858                 NF_CT_STAT_INC(net, new);
859         }
860 
861         /* Overload tuple linked list to put us in unconfirmed list. */
862         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
863                        &net->ct.unconfirmed);
864 
865         spin_unlock_bh(&nf_conntrack_lock);
866 
867         if (exp) {
868                 if (exp->expectfn)
869                         exp->expectfn(ct, exp);
870                 nf_ct_expect_put(exp);
871         }
872 
873         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
874 }
875 
876 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
877 static inline struct nf_conn *
878 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
879                   struct sk_buff *skb,
880                   unsigned int dataoff,
881                   u_int16_t l3num,
882                   u_int8_t protonum,
883                   struct nf_conntrack_l3proto *l3proto,
884                   struct nf_conntrack_l4proto *l4proto,
885                   int *set_reply,
886                   enum ip_conntrack_info *ctinfo)
887 {
888         struct nf_conntrack_tuple tuple;
889         struct nf_conntrack_tuple_hash *h;
890         struct nf_conn *ct;
891         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
892         u32 hash;
893 
894         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
895                              dataoff, l3num, protonum, &tuple, l3proto,
896                              l4proto)) {
897                 pr_debug("resolve_normal_ct: Can't get tuple\n");
898                 return NULL;
899         }
900 
901         /* look for tuple match */
902         hash = hash_conntrack_raw(&tuple, zone);
903         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
904         if (!h) {
905                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
906                                    skb, dataoff, hash);
907                 if (!h)
908                         return NULL;
909                 if (IS_ERR(h))
910                         return (void *)h;
911         }
912         ct = nf_ct_tuplehash_to_ctrack(h);
913 
914         /* It exists; we have (non-exclusive) reference. */
915         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
916                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
917                 /* Please set reply bit if this packet OK */
918                 *set_reply = 1;
919         } else {
920                 /* Once we've had two way comms, always ESTABLISHED. */
921                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
922                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
923                         *ctinfo = IP_CT_ESTABLISHED;
924                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
925                         pr_debug("nf_conntrack_in: related packet for %p\n",
926                                  ct);
927                         *ctinfo = IP_CT_RELATED;
928                 } else {
929                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
930                         *ctinfo = IP_CT_NEW;
931                 }
932                 *set_reply = 0;
933         }
934         skb->nfct = &ct->ct_general;
935         skb->nfctinfo = *ctinfo;
936         return ct;
937 }
938 
939 unsigned int
940 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
941                 struct sk_buff *skb)
942 {
943         struct nf_conn *ct, *tmpl = NULL;
944         enum ip_conntrack_info ctinfo;
945         struct nf_conntrack_l3proto *l3proto;
946         struct nf_conntrack_l4proto *l4proto;
947         unsigned int *timeouts;
948         unsigned int dataoff;
949         u_int8_t protonum;
950         int set_reply = 0;
951         int ret;
952 
953         if (skb->nfct) {
954                 /* Previously seen (loopback or untracked)?  Ignore. */
955                 tmpl = (struct nf_conn *)skb->nfct;
956                 if (!nf_ct_is_template(tmpl)) {
957                         NF_CT_STAT_INC_ATOMIC(net, ignore);
958                         return NF_ACCEPT;
959                 }
960                 skb->nfct = NULL;
961         }
962 
963         /* rcu_read_lock()ed by nf_hook_slow */
964         l3proto = __nf_ct_l3proto_find(pf);
965         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
966                                    &dataoff, &protonum);
967         if (ret <= 0) {
968                 pr_debug("not prepared to track yet or error occurred\n");
969                 NF_CT_STAT_INC_ATOMIC(net, error);
970                 NF_CT_STAT_INC_ATOMIC(net, invalid);
971                 ret = -ret;
972                 goto out;
973         }
974 
975         l4proto = __nf_ct_l4proto_find(pf, protonum);
976 
977         /* It may be an special packet, error, unclean...
978          * inverse of the return code tells to the netfilter
979          * core what to do with the packet. */
980         if (l4proto->error != NULL) {
981                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
982                                      pf, hooknum);
983                 if (ret <= 0) {
984                         NF_CT_STAT_INC_ATOMIC(net, error);
985                         NF_CT_STAT_INC_ATOMIC(net, invalid);
986                         ret = -ret;
987                         goto out;
988                 }
989                 /* ICMP[v6] protocol trackers may assign one conntrack. */
990                 if (skb->nfct)
991                         goto out;
992         }
993 
994         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
995                                l3proto, l4proto, &set_reply, &ctinfo);
996         if (!ct) {
997                 /* Not valid part of a connection */
998                 NF_CT_STAT_INC_ATOMIC(net, invalid);
999                 ret = NF_ACCEPT;
1000                 goto out;
1001         }
1002 
1003         if (IS_ERR(ct)) {
1004                 /* Too stressed to deal. */
1005                 NF_CT_STAT_INC_ATOMIC(net, drop);
1006                 ret = NF_DROP;
1007                 goto out;
1008         }
1009 
1010         NF_CT_ASSERT(skb->nfct);
1011 
1012         /* Decide what timeout policy we want to apply to this flow. */
1013         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1014 
1015         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1016         if (ret <= 0) {
1017                 /* Invalid: inverse of the return code tells
1018                  * the netfilter core what to do */
1019                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1020                 nf_conntrack_put(skb->nfct);
1021                 skb->nfct = NULL;
1022                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1023                 if (ret == -NF_DROP)
1024                         NF_CT_STAT_INC_ATOMIC(net, drop);
1025                 ret = -ret;
1026                 goto out;
1027         }
1028 
1029         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1030                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1031 out:
1032         if (tmpl) {
1033                 /* Special case: we have to repeat this hook, assign the
1034                  * template again to this packet. We assume that this packet
1035                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1036                 if (ret == NF_REPEAT)
1037                         skb->nfct = (struct nf_conntrack *)tmpl;
1038                 else
1039                         nf_ct_put(tmpl);
1040         }
1041 
1042         return ret;
1043 }
1044 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1045 
1046 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1047                           const struct nf_conntrack_tuple *orig)
1048 {
1049         bool ret;
1050 
1051         rcu_read_lock();
1052         ret = nf_ct_invert_tuple(inverse, orig,
1053                                  __nf_ct_l3proto_find(orig->src.l3num),
1054                                  __nf_ct_l4proto_find(orig->src.l3num,
1055                                                       orig->dst.protonum));
1056         rcu_read_unlock();
1057         return ret;
1058 }
1059 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1060 
1061 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1062    implicitly racy: see __nf_conntrack_confirm */
1063 void nf_conntrack_alter_reply(struct nf_conn *ct,
1064                               const struct nf_conntrack_tuple *newreply)
1065 {
1066         struct nf_conn_help *help = nfct_help(ct);
1067 
1068         /* Should be unconfirmed, so not in hash table yet */
1069         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1070 
1071         pr_debug("Altering reply tuple of %p to ", ct);
1072         nf_ct_dump_tuple(newreply);
1073 
1074         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1075         if (ct->master || (help && !hlist_empty(&help->expectations)))
1076                 return;
1077 
1078         rcu_read_lock();
1079         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1080         rcu_read_unlock();
1081 }
1082 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1083 
1084 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1085 void __nf_ct_refresh_acct(struct nf_conn *ct,
1086                           enum ip_conntrack_info ctinfo,
1087                           const struct sk_buff *skb,
1088                           unsigned long extra_jiffies,
1089                           int do_acct)
1090 {
1091         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1092         NF_CT_ASSERT(skb);
1093 
1094         /* Only update if this is not a fixed timeout */
1095         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1096                 goto acct;
1097 
1098         /* If not in hash table, timer will not be active yet */
1099         if (!nf_ct_is_confirmed(ct)) {
1100                 ct->timeout.expires = extra_jiffies;
1101         } else {
1102                 unsigned long newtime = jiffies + extra_jiffies;
1103 
1104                 /* Only update the timeout if the new timeout is at least
1105                    HZ jiffies from the old timeout. Need del_timer for race
1106                    avoidance (may already be dying). */
1107                 if (newtime - ct->timeout.expires >= HZ)
1108                         mod_timer_pending(&ct->timeout, newtime);
1109         }
1110 
1111 acct:
1112         if (do_acct) {
1113                 struct nf_conn_acct *acct;
1114 
1115                 acct = nf_conn_acct_find(ct);
1116                 if (acct) {
1117                         struct nf_conn_counter *counter = acct->counter;
1118 
1119                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1120                         atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1121                 }
1122         }
1123 }
1124 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1125 
1126 bool __nf_ct_kill_acct(struct nf_conn *ct,
1127                        enum ip_conntrack_info ctinfo,
1128                        const struct sk_buff *skb,
1129                        int do_acct)
1130 {
1131         if (do_acct) {
1132                 struct nf_conn_acct *acct;
1133 
1134                 acct = nf_conn_acct_find(ct);
1135                 if (acct) {
1136                         struct nf_conn_counter *counter = acct->counter;
1137 
1138                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1139                         atomic64_add(skb->len - skb_network_offset(skb),
1140                                      &counter[CTINFO2DIR(ctinfo)].bytes);
1141                 }
1142         }
1143 
1144         if (del_timer(&ct->timeout)) {
1145                 ct->timeout.function((unsigned long)ct);
1146                 return true;
1147         }
1148         return false;
1149 }
1150 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1151 
1152 #ifdef CONFIG_NF_CONNTRACK_ZONES
1153 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1154         .len    = sizeof(struct nf_conntrack_zone),
1155         .align  = __alignof__(struct nf_conntrack_zone),
1156         .id     = NF_CT_EXT_ZONE,
1157 };
1158 #endif
1159 
1160 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1161 
1162 #include <linux/netfilter/nfnetlink.h>
1163 #include <linux/netfilter/nfnetlink_conntrack.h>
1164 #include <linux/mutex.h>
1165 
1166 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1167  * in ip_conntrack_core, since we don't want the protocols to autoload
1168  * or depend on ctnetlink */
1169 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1170                                const struct nf_conntrack_tuple *tuple)
1171 {
1172         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1173             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1174                 goto nla_put_failure;
1175         return 0;
1176 
1177 nla_put_failure:
1178         return -1;
1179 }
1180 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1181 
1182 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1183         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1184         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1185 };
1186 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1187 
1188 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1189                                struct nf_conntrack_tuple *t)
1190 {
1191         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1192                 return -EINVAL;
1193 
1194         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1195         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1196 
1197         return 0;
1198 }
1199 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1200 
1201 int nf_ct_port_nlattr_tuple_size(void)
1202 {
1203         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1204 }
1205 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1206 #endif
1207 
1208 /* Used by ipt_REJECT and ip6t_REJECT. */
1209 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1210 {
1211         struct nf_conn *ct;
1212         enum ip_conntrack_info ctinfo;
1213 
1214         /* This ICMP is in reverse direction to the packet which caused it */
1215         ct = nf_ct_get(skb, &ctinfo);
1216         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1217                 ctinfo = IP_CT_RELATED_REPLY;
1218         else
1219                 ctinfo = IP_CT_RELATED;
1220 
1221         /* Attach to new skbuff, and increment count */
1222         nskb->nfct = &ct->ct_general;
1223         nskb->nfctinfo = ctinfo;
1224         nf_conntrack_get(nskb->nfct);
1225 }
1226 
1227 /* Bring out ya dead! */
1228 static struct nf_conn *
1229 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1230                 void *data, unsigned int *bucket)
1231 {
1232         struct nf_conntrack_tuple_hash *h;
1233         struct nf_conn *ct;
1234         struct hlist_nulls_node *n;
1235 
1236         spin_lock_bh(&nf_conntrack_lock);
1237         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1238                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1239                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1240                                 continue;
1241                         ct = nf_ct_tuplehash_to_ctrack(h);
1242                         if (iter(ct, data))
1243                                 goto found;
1244                 }
1245         }
1246         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1247                 ct = nf_ct_tuplehash_to_ctrack(h);
1248                 if (iter(ct, data))
1249                         set_bit(IPS_DYING_BIT, &ct->status);
1250         }
1251         spin_unlock_bh(&nf_conntrack_lock);
1252         return NULL;
1253 found:
1254         atomic_inc(&ct->ct_general.use);
1255         spin_unlock_bh(&nf_conntrack_lock);
1256         return ct;
1257 }
1258 
1259 void nf_ct_iterate_cleanup(struct net *net,
1260                            int (*iter)(struct nf_conn *i, void *data),
1261                            void *data, u32 portid, int report)
1262 {
1263         struct nf_conn *ct;
1264         unsigned int bucket = 0;
1265 
1266         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1267                 /* Time to push up daises... */
1268                 if (del_timer(&ct->timeout))
1269                         nf_ct_delete(ct, portid, report);
1270 
1271                 /* ... else the timer will get him soon. */
1272 
1273                 nf_ct_put(ct);
1274         }
1275 }
1276 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1277 
1278 static int kill_all(struct nf_conn *i, void *data)
1279 {
1280         return 1;
1281 }
1282 
1283 void nf_ct_free_hashtable(void *hash, unsigned int size)
1284 {
1285         if (is_vmalloc_addr(hash))
1286                 vfree(hash);
1287         else
1288                 free_pages((unsigned long)hash,
1289                            get_order(sizeof(struct hlist_head) * size));
1290 }
1291 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1292 
1293 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1294 {
1295         nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1296 }
1297 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1298 
1299 static void nf_ct_release_dying_list(struct net *net)
1300 {
1301         struct nf_conntrack_tuple_hash *h;
1302         struct nf_conn *ct;
1303         struct hlist_nulls_node *n;
1304 
1305         spin_lock_bh(&nf_conntrack_lock);
1306         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1307                 ct = nf_ct_tuplehash_to_ctrack(h);
1308                 /* never fails to remove them, no listeners at this point */
1309                 nf_ct_kill(ct);
1310         }
1311         spin_unlock_bh(&nf_conntrack_lock);
1312 }
1313 
1314 static int untrack_refs(void)
1315 {
1316         int cnt = 0, cpu;
1317 
1318         for_each_possible_cpu(cpu) {
1319                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1320 
1321                 cnt += atomic_read(&ct->ct_general.use) - 1;
1322         }
1323         return cnt;
1324 }
1325 
1326 void nf_conntrack_cleanup_start(void)
1327 {
1328         RCU_INIT_POINTER(ip_ct_attach, NULL);
1329 }
1330 
1331 void nf_conntrack_cleanup_end(void)
1332 {
1333         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1334         while (untrack_refs() > 0)
1335                 schedule();
1336 
1337 #ifdef CONFIG_NF_CONNTRACK_ZONES
1338         nf_ct_extend_unregister(&nf_ct_zone_extend);
1339 #endif
1340         nf_conntrack_proto_fini();
1341         nf_conntrack_seqadj_fini();
1342         nf_conntrack_labels_fini();
1343         nf_conntrack_helper_fini();
1344         nf_conntrack_timeout_fini();
1345         nf_conntrack_ecache_fini();
1346         nf_conntrack_tstamp_fini();
1347         nf_conntrack_acct_fini();
1348         nf_conntrack_expect_fini();
1349 }
1350 
1351 /*
1352  * Mishearing the voices in his head, our hero wonders how he's
1353  * supposed to kill the mall.
1354  */
1355 void nf_conntrack_cleanup_net(struct net *net)
1356 {
1357         LIST_HEAD(single);
1358 
1359         list_add(&net->exit_list, &single);
1360         nf_conntrack_cleanup_net_list(&single);
1361 }
1362 
1363 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1364 {
1365         int busy;
1366         struct net *net;
1367 
1368         /*
1369          * This makes sure all current packets have passed through
1370          *  netfilter framework.  Roll on, two-stage module
1371          *  delete...
1372          */
1373         synchronize_net();
1374 i_see_dead_people:
1375         busy = 0;
1376         list_for_each_entry(net, net_exit_list, exit_list) {
1377                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1378                 nf_ct_release_dying_list(net);
1379                 if (atomic_read(&net->ct.count) != 0)
1380                         busy = 1;
1381         }
1382         if (busy) {
1383                 schedule();
1384                 goto i_see_dead_people;
1385         }
1386 
1387         list_for_each_entry(net, net_exit_list, exit_list) {
1388                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1389                 nf_conntrack_proto_pernet_fini(net);
1390                 nf_conntrack_helper_pernet_fini(net);
1391                 nf_conntrack_ecache_pernet_fini(net);
1392                 nf_conntrack_tstamp_pernet_fini(net);
1393                 nf_conntrack_acct_pernet_fini(net);
1394                 nf_conntrack_expect_pernet_fini(net);
1395                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1396                 kfree(net->ct.slabname);
1397                 free_percpu(net->ct.stat);
1398         }
1399 }
1400 
1401 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1402 {
1403         struct hlist_nulls_head *hash;
1404         unsigned int nr_slots, i;
1405         size_t sz;
1406 
1407         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1408         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1409         sz = nr_slots * sizeof(struct hlist_nulls_head);
1410         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1411                                         get_order(sz));
1412         if (!hash) {
1413                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1414                 hash = vzalloc(sz);
1415         }
1416 
1417         if (hash && nulls)
1418                 for (i = 0; i < nr_slots; i++)
1419                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1420 
1421         return hash;
1422 }
1423 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1424 
1425 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1426 {
1427         int i, bucket, rc;
1428         unsigned int hashsize, old_size;
1429         struct hlist_nulls_head *hash, *old_hash;
1430         struct nf_conntrack_tuple_hash *h;
1431         struct nf_conn *ct;
1432 
1433         if (current->nsproxy->net_ns != &init_net)
1434                 return -EOPNOTSUPP;
1435 
1436         /* On boot, we can set this without any fancy locking. */
1437         if (!nf_conntrack_htable_size)
1438                 return param_set_uint(val, kp);
1439 
1440         rc = kstrtouint(val, 0, &hashsize);
1441         if (rc)
1442                 return rc;
1443         if (!hashsize)
1444                 return -EINVAL;
1445 
1446         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1447         if (!hash)
1448                 return -ENOMEM;
1449 
1450         /* Lookups in the old hash might happen in parallel, which means we
1451          * might get false negatives during connection lookup. New connections
1452          * created because of a false negative won't make it into the hash
1453          * though since that required taking the lock.
1454          */
1455         spin_lock_bh(&nf_conntrack_lock);
1456         for (i = 0; i < init_net.ct.htable_size; i++) {
1457                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1458                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1459                                         struct nf_conntrack_tuple_hash, hnnode);
1460                         ct = nf_ct_tuplehash_to_ctrack(h);
1461                         hlist_nulls_del_rcu(&h->hnnode);
1462                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1463                                                   hashsize);
1464                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1465                 }
1466         }
1467         old_size = init_net.ct.htable_size;
1468         old_hash = init_net.ct.hash;
1469 
1470         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1471         init_net.ct.hash = hash;
1472         spin_unlock_bh(&nf_conntrack_lock);
1473 
1474         nf_ct_free_hashtable(old_hash, old_size);
1475         return 0;
1476 }
1477 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1478 
1479 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1480                   &nf_conntrack_htable_size, 0600);
1481 
1482 void nf_ct_untracked_status_or(unsigned long bits)
1483 {
1484         int cpu;
1485 
1486         for_each_possible_cpu(cpu)
1487                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1488 }
1489 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1490 
1491 int nf_conntrack_init_start(void)
1492 {
1493         int max_factor = 8;
1494         int ret, cpu;
1495 
1496         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1497          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1498         if (!nf_conntrack_htable_size) {
1499                 nf_conntrack_htable_size
1500                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1501                            / sizeof(struct hlist_head));
1502                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1503                         nf_conntrack_htable_size = 16384;
1504                 if (nf_conntrack_htable_size < 32)
1505                         nf_conntrack_htable_size = 32;
1506 
1507                 /* Use a max. factor of four by default to get the same max as
1508                  * with the old struct list_heads. When a table size is given
1509                  * we use the old value of 8 to avoid reducing the max.
1510                  * entries. */
1511                 max_factor = 4;
1512         }
1513         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1514 
1515         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1516                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1517                nf_conntrack_max);
1518 
1519         ret = nf_conntrack_expect_init();
1520         if (ret < 0)
1521                 goto err_expect;
1522 
1523         ret = nf_conntrack_acct_init();
1524         if (ret < 0)
1525                 goto err_acct;
1526 
1527         ret = nf_conntrack_tstamp_init();
1528         if (ret < 0)
1529                 goto err_tstamp;
1530 
1531         ret = nf_conntrack_ecache_init();
1532         if (ret < 0)
1533                 goto err_ecache;
1534 
1535         ret = nf_conntrack_timeout_init();
1536         if (ret < 0)
1537                 goto err_timeout;
1538 
1539         ret = nf_conntrack_helper_init();
1540         if (ret < 0)
1541                 goto err_helper;
1542 
1543         ret = nf_conntrack_labels_init();
1544         if (ret < 0)
1545                 goto err_labels;
1546 
1547         ret = nf_conntrack_seqadj_init();
1548         if (ret < 0)
1549                 goto err_seqadj;
1550 
1551 #ifdef CONFIG_NF_CONNTRACK_ZONES
1552         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1553         if (ret < 0)
1554                 goto err_extend;
1555 #endif
1556         ret = nf_conntrack_proto_init();
1557         if (ret < 0)
1558                 goto err_proto;
1559 
1560         /* Set up fake conntrack: to never be deleted, not in any hashes */
1561         for_each_possible_cpu(cpu) {
1562                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1563                 write_pnet(&ct->ct_net, &init_net);
1564                 atomic_set(&ct->ct_general.use, 1);
1565         }
1566         /*  - and look it like as a confirmed connection */
1567         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1568         return 0;
1569 
1570 err_proto:
1571 #ifdef CONFIG_NF_CONNTRACK_ZONES
1572         nf_ct_extend_unregister(&nf_ct_zone_extend);
1573 err_extend:
1574 #endif
1575         nf_conntrack_seqadj_fini();
1576 err_seqadj:
1577         nf_conntrack_labels_fini();
1578 err_labels:
1579         nf_conntrack_helper_fini();
1580 err_helper:
1581         nf_conntrack_timeout_fini();
1582 err_timeout:
1583         nf_conntrack_ecache_fini();
1584 err_ecache:
1585         nf_conntrack_tstamp_fini();
1586 err_tstamp:
1587         nf_conntrack_acct_fini();
1588 err_acct:
1589         nf_conntrack_expect_fini();
1590 err_expect:
1591         return ret;
1592 }
1593 
1594 void nf_conntrack_init_end(void)
1595 {
1596         /* For use by REJECT target */
1597         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1598         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1599 }
1600 
1601 /*
1602  * We need to use special "null" values, not used in hash table
1603  */
1604 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1605 #define DYING_NULLS_VAL         ((1<<30)+1)
1606 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1607 
1608 int nf_conntrack_init_net(struct net *net)
1609 {
1610         int ret;
1611 
1612         atomic_set(&net->ct.count, 0);
1613         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1614         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1615         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1616         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1617         if (!net->ct.stat) {
1618                 ret = -ENOMEM;
1619                 goto err_stat;
1620         }
1621 
1622         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1623         if (!net->ct.slabname) {
1624                 ret = -ENOMEM;
1625                 goto err_slabname;
1626         }
1627 
1628         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1629                                                         sizeof(struct nf_conn), 0,
1630                                                         SLAB_DESTROY_BY_RCU, NULL);
1631         if (!net->ct.nf_conntrack_cachep) {
1632                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1633                 ret = -ENOMEM;
1634                 goto err_cache;
1635         }
1636 
1637         net->ct.htable_size = nf_conntrack_htable_size;
1638         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1639         if (!net->ct.hash) {
1640                 ret = -ENOMEM;
1641                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1642                 goto err_hash;
1643         }
1644         ret = nf_conntrack_expect_pernet_init(net);
1645         if (ret < 0)
1646                 goto err_expect;
1647         ret = nf_conntrack_acct_pernet_init(net);
1648         if (ret < 0)
1649                 goto err_acct;
1650         ret = nf_conntrack_tstamp_pernet_init(net);
1651         if (ret < 0)
1652                 goto err_tstamp;
1653         ret = nf_conntrack_ecache_pernet_init(net);
1654         if (ret < 0)
1655                 goto err_ecache;
1656         ret = nf_conntrack_helper_pernet_init(net);
1657         if (ret < 0)
1658                 goto err_helper;
1659         ret = nf_conntrack_proto_pernet_init(net);
1660         if (ret < 0)
1661                 goto err_proto;
1662         return 0;
1663 
1664 err_proto:
1665         nf_conntrack_helper_pernet_fini(net);
1666 err_helper:
1667         nf_conntrack_ecache_pernet_fini(net);
1668 err_ecache:
1669         nf_conntrack_tstamp_pernet_fini(net);
1670 err_tstamp:
1671         nf_conntrack_acct_pernet_fini(net);
1672 err_acct:
1673         nf_conntrack_expect_pernet_fini(net);
1674 err_expect:
1675         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1676 err_hash:
1677         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1678 err_cache:
1679         kfree(net->ct.slabname);
1680 err_slabname:
1681         free_percpu(net->ct.stat);
1682 err_stat:
1683         return ret;
1684 }
1685 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp