blob: 48f4244651125fb4ef7bcfda155137f287b07c77 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INETPEER - A storage for permanent information about peers
3 *
4 * This source is covered by the GNU GPL, the same as all kernel sources.
5 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 * Authors: Andrey V. Savochkin <saw@msu.ru>
7 */
8
9#include <linux/module.h>
10#include <linux/types.h>
11#include <linux/slab.h>
12#include <linux/interrupt.h>
13#include <linux/spinlock.h>
14#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/timer.h>
16#include <linux/time.h>
17#include <linux/kernel.h>
18#include <linux/mm.h>
19#include <linux/net.h>
Steffen Klassert5faa5df2012-03-06 21:20:26 +000020#include <linux/workqueue.h>
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -030021#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <net/inetpeer.h>
David S. Miller6e5714e2011-08-03 20:50:44 -070023#include <net/secure_seq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
25/*
26 * Theory of operations.
27 * We keep one entry for each peer IP address. The nodes contains long-living
28 * information about the peer which doesn't depend on routes.
29 * At this moment this information consists only of ID field for the next
30 * outgoing IP packet. This field is incremented with each packet as encoded
31 * in inet_getid() function (include/net/inetpeer.h).
32 * At the moment of writing this notes identifier of IP packets is generated
33 * to be unpredictable using this code only for packets subjected
34 * (actually or potentially) to defragmentation. I.e. DF packets less than
Ansis Atteka703133d2013-09-18 15:29:53 -070035 * PMTU in size when local fragmentation is disabled use a constant ID and do
36 * not use this code (see ip_select_ident() in include/net/ip.h).
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 *
38 * Route cache entries hold references to our nodes.
39 * New cache entries get references via lookup by destination IP address in
40 * the avl tree. The reference is grabbed only when it's needed i.e. only
41 * when we try to output IP packet which needs an unpredictable ID (see
42 * __ip_select_ident() in net/ipv4/route.c).
43 * Nodes are removed only when reference counter goes to 0.
44 * When it's happened the node may be removed when a sufficient amount of
45 * time has been passed since its last use. The less-recently-used entry can
46 * also be removed if the pool is overloaded i.e. if the total amount of
47 * entries is greater-or-equal than the threshold.
48 *
49 * Node pool is organised as an AVL tree.
50 * Such an implementation has been chosen not just for fun. It's a way to
51 * prevent easy and efficient DoS attacks by creating hash collisions. A huge
52 * amount of long living nodes in a single hash slot would significantly delay
53 * lookups performed with disabled BHs.
54 *
55 * Serialisation issues.
Eric Dumazetaa1039e2010-06-15 08:23:14 +000056 * 1. Nodes may appear in the tree only with the pool lock held.
57 * 2. Nodes may disappear from the tree only with the pool lock held
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 * AND reference count being 0.
Eric Dumazet4b9d9be2011-06-08 13:35:34 +000059 * 3. Global variable peer_total is modified under the pool lock.
60 * 4. struct inet_peer fields modification:
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * avl_left, avl_right, avl_parent, avl_height: pool lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * refcnt: atomically against modifications on other CPU;
63 * usually under some other lock to prevent node disappearing
David S. Miller582a72d2010-11-30 11:53:55 -080064 * daddr: unchangeable
Eric Dumazet317fe0e2010-06-16 04:52:13 +000065 * ip_id_count: atomic value (no lock needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 */
67
Christoph Lametere18b8902006-12-06 20:33:20 -080068static struct kmem_cache *peer_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Steffen Klassert5faa5df2012-03-06 21:20:26 +000070static LIST_HEAD(gc_list);
71static const int gc_delay = 60 * HZ;
72static struct delayed_work gc_work;
73static DEFINE_SPINLOCK(gc_lock);
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#define node_height(x) x->avl_height
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000076
77#define peer_avl_empty ((struct inet_peer *)&peer_fake_node)
Eric Dumazetb914c4e2010-10-25 23:55:38 +000078#define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node)
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000079static const struct inet_peer peer_fake_node = {
Eric Dumazetb914c4e2010-10-25 23:55:38 +000080 .avl_left = peer_avl_empty_rcu,
81 .avl_right = peer_avl_empty_rcu,
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 .avl_height = 0
83};
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000084
David S. Millerc3426b42012-06-09 16:27:05 -070085void inet_peer_base_init(struct inet_peer_base *bp)
86{
87 bp->root = peer_avl_empty_rcu;
88 seqlock_init(&bp->lock);
David S. Millerb48c80ece2012-06-10 00:24:21 -070089 bp->flush_seq = ~0U;
David S. Millerc3426b42012-06-09 16:27:05 -070090 bp->total = 0;
91}
92EXPORT_SYMBOL_GPL(inet_peer_base_init);
David S. Miller021e9292010-11-30 12:12:23 -080093
David S. Millerb48c80ece2012-06-10 00:24:21 -070094static atomic_t v4_seq = ATOMIC_INIT(0);
95static atomic_t v6_seq = ATOMIC_INIT(0);
96
97static atomic_t *inetpeer_seq_ptr(int family)
98{
99 return (family == AF_INET ? &v4_seq : &v6_seq);
100}
101
102static inline void flush_check(struct inet_peer_base *base, int family)
103{
104 atomic_t *fp = inetpeer_seq_ptr(family);
105
106 if (unlikely(base->flush_seq != atomic_read(fp))) {
107 inetpeer_invalidate_tree(base);
108 base->flush_seq = atomic_read(fp);
109 }
110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/* Exported for sysctl_net_ipv4. */
Eric Dumazet243bbca2007-03-06 20:23:10 -0800115int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 * aggressively at this stage */
Eric Dumazet243bbca2007-03-06 20:23:10 -0800117int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */
118int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000120static void inetpeer_gc_worker(struct work_struct *work)
121{
Eric Dumazetda557372012-06-20 04:02:10 +0000122 struct inet_peer *p, *n, *c;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000123 LIST_HEAD(list);
124
125 spin_lock_bh(&gc_lock);
126 list_replace_init(&gc_list, &list);
127 spin_unlock_bh(&gc_lock);
128
129 if (list_empty(&list))
130 return;
131
132 list_for_each_entry_safe(p, n, &list, gc_list) {
133
Eric Dumazetda557372012-06-20 04:02:10 +0000134 if (need_resched())
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000135 cond_resched();
136
Eric Dumazetda557372012-06-20 04:02:10 +0000137 c = rcu_dereference_protected(p->avl_left, 1);
138 if (c != peer_avl_empty) {
139 list_add_tail(&c->gc_list, &list);
140 p->avl_left = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000141 }
142
Eric Dumazetda557372012-06-20 04:02:10 +0000143 c = rcu_dereference_protected(p->avl_right, 1);
144 if (c != peer_avl_empty) {
145 list_add_tail(&c->gc_list, &list);
146 p->avl_right = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000147 }
148
149 n = list_entry(p->gc_list.next, struct inet_peer, gc_list);
150
151 if (!atomic_read(&p->refcnt)) {
152 list_del(&p->gc_list);
153 kmem_cache_free(peer_cachep, p);
154 }
155 }
156
157 if (list_empty(&list))
158 return;
159
160 spin_lock_bh(&gc_lock);
161 list_splice(&list, &gc_list);
162 spin_unlock_bh(&gc_lock);
163
164 schedule_delayed_work(&gc_work, gc_delay);
165}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167/* Called from ip_output.c:ip_init */
168void __init inet_initpeers(void)
169{
170 struct sysinfo si;
171
172 /* Use the straight interface to information about memory. */
173 si_meminfo(&si);
174 /* The values below were suggested by Alexey Kuznetsov
175 * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values
176 * myself. --SAW
177 */
178 if (si.totalram <= (32768*1024)/PAGE_SIZE)
179 inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
180 if (si.totalram <= (16384*1024)/PAGE_SIZE)
181 inet_peer_threshold >>= 1; /* about 512KB */
182 if (si.totalram <= (8192*1024)/PAGE_SIZE)
183 inet_peer_threshold >>= 2; /* about 128KB */
184
185 peer_cachep = kmem_cache_create("inet_peer_cache",
186 sizeof(struct inet_peer),
Eric Dumazet317fe0e2010-06-16 04:52:13 +0000187 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +0900188 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Tejun Heo203b42f2012-08-21 13:18:23 -0700190 INIT_DEFERRABLE_WORK(&gc_work, inetpeer_gc_worker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191}
192
David S. Miller8790ca12010-12-01 17:28:18 -0800193static int addr_compare(const struct inetpeer_addr *a,
194 const struct inetpeer_addr *b)
David S. Miller02663042010-11-30 12:08:53 -0800195{
196 int i, n = (a->family == AF_INET ? 1 : 4);
197
198 for (i = 0; i < n; i++) {
David S. Miller7a71ed82011-02-09 14:30:26 -0800199 if (a->addr.a6[i] == b->addr.a6[i])
David S. Miller02663042010-11-30 12:08:53 -0800200 continue;
Eric Dumazet747465e2012-01-16 19:27:39 +0000201 if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i])
David S. Miller02663042010-11-30 12:08:53 -0800202 return -1;
203 return 1;
204 }
205
206 return 0;
207}
208
Eric Dumazet65e83542011-03-04 14:33:59 -0800209#define rcu_deref_locked(X, BASE) \
210 rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))
211
Eric Dumazet243bbca2007-03-06 20:23:10 -0800212/*
213 * Called with local BH disabled and the pool lock held.
Eric Dumazet243bbca2007-03-06 20:23:10 -0800214 */
David S. Miller98158f52010-11-30 11:41:59 -0800215#define lookup(_daddr, _stack, _base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216({ \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000217 struct inet_peer *u; \
218 struct inet_peer __rcu **v; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000219 \
220 stackptr = _stack; \
David S. Miller98158f52010-11-30 11:41:59 -0800221 *stackptr++ = &_base->root; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800222 for (u = rcu_deref_locked(_base->root, _base); \
Weilong Chen0c9a67d2013-12-23 14:37:27 +0800223 u != peer_avl_empty;) { \
David S. Miller02663042010-11-30 12:08:53 -0800224 int cmp = addr_compare(_daddr, &u->daddr); \
225 if (cmp == 0) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 break; \
David S. Miller02663042010-11-30 12:08:53 -0800227 if (cmp == -1) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 v = &u->avl_left; \
229 else \
230 v = &u->avl_right; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000231 *stackptr++ = v; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800232 u = rcu_deref_locked(*v, _base); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 } \
234 u; \
235})
236
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000237/*
David S. Miller7b46ac42011-03-08 14:59:28 -0800238 * Called with rcu_read_lock()
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000239 * Because we hold no lock against a writer, its quite possible we fall
240 * in an endless loop.
241 * But every pointer we follow is guaranteed to be valid thanks to RCU.
242 * We exit from this function if number of links exceeds PEER_MAXDEPTH
243 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800244static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000245 struct inet_peer_base *base)
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000246{
David S. Miller7b46ac42011-03-08 14:59:28 -0800247 struct inet_peer *u = rcu_dereference(base->root);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000248 int count = 0;
249
250 while (u != peer_avl_empty) {
David S. Miller02663042010-11-30 12:08:53 -0800251 int cmp = addr_compare(daddr, &u->daddr);
252 if (cmp == 0) {
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700253 /* Before taking a reference, check if this entry was
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000254 * deleted (refcnt=-1)
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700255 */
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000256 if (!atomic_add_unless(&u->refcnt, 1, -1))
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000257 u = NULL;
258 return u;
259 }
David S. Miller02663042010-11-30 12:08:53 -0800260 if (cmp == -1)
David S. Miller7b46ac42011-03-08 14:59:28 -0800261 u = rcu_dereference(u->avl_left);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000262 else
David S. Miller7b46ac42011-03-08 14:59:28 -0800263 u = rcu_dereference(u->avl_right);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000264 if (unlikely(++count == PEER_MAXDEPTH))
265 break;
266 }
267 return NULL;
268}
269
270/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800271#define lookup_rightempty(start, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272({ \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000273 struct inet_peer *u; \
274 struct inet_peer __rcu **v; \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 *stackptr++ = &start->avl_left; \
276 v = &start->avl_left; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800277 for (u = rcu_deref_locked(*v, base); \
Weilong Chen0c9a67d2013-12-23 14:37:27 +0800278 u->avl_right != peer_avl_empty_rcu;) { \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 v = &u->avl_right; \
280 *stackptr++ = v; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800281 u = rcu_deref_locked(*v, base); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 } \
283 u; \
284})
285
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000286/* Called with local BH disabled and the pool lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 * Variable names are the proof of operation correctness.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000288 * Look into mm/map_avl.c for more detail description of the ideas.
289 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000290static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
David S. Miller98158f52010-11-30 11:41:59 -0800291 struct inet_peer __rcu ***stackend,
292 struct inet_peer_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000294 struct inet_peer __rcu **nodep;
295 struct inet_peer *node, *l, *r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 int lh, rh;
297
298 while (stackend > stack) {
299 nodep = *--stackend;
Eric Dumazet65e83542011-03-04 14:33:59 -0800300 node = rcu_deref_locked(*nodep, base);
301 l = rcu_deref_locked(node->avl_left, base);
302 r = rcu_deref_locked(node->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 lh = node_height(l);
304 rh = node_height(r);
305 if (lh > rh + 1) { /* l: RH+2 */
306 struct inet_peer *ll, *lr, *lrl, *lrr;
307 int lrh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800308 ll = rcu_deref_locked(l->avl_left, base);
309 lr = rcu_deref_locked(l->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 lrh = node_height(lr);
311 if (lrh <= node_height(ll)) { /* ll: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000312 RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */
313 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 node->avl_height = lrh + 1; /* RH+1 or RH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000315 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */
316 RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 l->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000318 RCU_INIT_POINTER(*nodep, l);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 } else { /* ll: RH, lr: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800320 lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
321 lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000322 RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */
323 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 node->avl_height = rh + 1; /* node: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000325 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */
326 RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 l->avl_height = rh + 1; /* l: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000328 RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */
329 RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 lr->avl_height = rh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000331 RCU_INIT_POINTER(*nodep, lr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 }
333 } else if (rh > lh + 1) { /* r: LH+2 */
334 struct inet_peer *rr, *rl, *rlr, *rll;
335 int rlh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800336 rr = rcu_deref_locked(r->avl_right, base);
337 rl = rcu_deref_locked(r->avl_left, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 rlh = node_height(rl);
339 if (rlh <= node_height(rr)) { /* rr: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000340 RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */
341 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 node->avl_height = rlh + 1; /* LH+1 or LH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000343 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */
344 RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 r->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000346 RCU_INIT_POINTER(*nodep, r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 } else { /* rr: RH, rl: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800348 rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
349 rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000350 RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */
351 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 node->avl_height = lh + 1; /* node: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000353 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */
354 RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 r->avl_height = lh + 1; /* r: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000356 RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */
357 RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 rl->avl_height = lh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000359 RCU_INIT_POINTER(*nodep, rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 }
361 } else {
362 node->avl_height = (lh > rh ? lh : rh) + 1;
363 }
364 }
365}
366
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000367/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800368#define link_to_pool(n, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369do { \
370 n->avl_height = 1; \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000371 n->avl_left = peer_avl_empty_rcu; \
372 n->avl_right = peer_avl_empty_rcu; \
373 /* lockless readers can catch us now */ \
374 rcu_assign_pointer(**--stackptr, n); \
David S. Miller98158f52010-11-30 11:41:59 -0800375 peer_avl_rebalance(stack, stackptr, base); \
Eric Dumazetd6cc1d62010-06-14 19:35:21 +0000376} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000378static void inetpeer_free_rcu(struct rcu_head *head)
379{
380 kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
381}
382
Eric Dumazet66944e12011-04-11 22:39:40 +0000383static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
384 struct inet_peer __rcu **stack[PEER_MAXDEPTH])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000386 struct inet_peer __rcu ***stackptr, ***delp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000388 if (lookup(&p->daddr, stack, base) != p)
389 BUG();
390 delp = stackptr - 1; /* *delp[0] == p */
391 if (p->avl_left == peer_avl_empty_rcu) {
392 *delp[0] = p->avl_right;
393 --stackptr;
394 } else {
395 /* look for a node to insert instead of p */
396 struct inet_peer *t;
397 t = lookup_rightempty(p, base);
398 BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
399 **--stackptr = t->avl_left;
400 /* t is removed, t->daddr > x->daddr for any
401 * x in p->avl_left subtree.
402 * Put t in the old place of p. */
403 RCU_INIT_POINTER(*delp[0], t);
404 t->avl_left = p->avl_left;
405 t->avl_right = p->avl_right;
406 t->avl_height = p->avl_height;
407 BUG_ON(delp[1] != &p->avl_left);
408 delp[1] = &t->avl_left; /* was &p->avl_left */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000410 peer_avl_rebalance(stack, stackptr, base);
411 base->total--;
412 call_rcu(&p->rcu, inetpeer_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413}
414
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000415/* perform garbage collect on all items stacked during a lookup */
416static int inet_peer_gc(struct inet_peer_base *base,
417 struct inet_peer __rcu **stack[PEER_MAXDEPTH],
418 struct inet_peer __rcu ***stackptr)
David S. Miller98158f52010-11-30 11:41:59 -0800419{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000420 struct inet_peer *p, *gchead = NULL;
421 __u32 delta, ttl;
422 int cnt = 0;
David S. Miller98158f52010-11-30 11:41:59 -0800423
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000424 if (base->total >= inet_peer_threshold)
425 ttl = 0; /* be aggressive */
426 else
427 ttl = inet_peer_maxttl
428 - (inet_peer_maxttl - inet_peer_minttl) / HZ *
429 base->total / inet_peer_threshold * HZ;
430 stackptr--; /* last stack slot is peer_avl_empty */
431 while (stackptr > stack) {
432 stackptr--;
433 p = rcu_deref_locked(**stackptr, base);
Eric Dumazet6d1a3e02011-07-11 02:49:52 +0000434 if (atomic_read(&p->refcnt) == 0) {
435 smp_rmb();
436 delta = (__u32)jiffies - p->dtime;
437 if (delta >= ttl &&
438 atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
439 p->gc_next = gchead;
440 gchead = p;
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000444 while ((p = gchead) != NULL) {
445 gchead = p->gc_next;
446 cnt++;
447 unlink_from_pool(p, base, stack);
448 }
449 return cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450}
451
David S. Millerc0efc882012-06-09 19:12:36 -0700452struct inet_peer *inet_getpeer(struct inet_peer_base *base,
Gao fengc8a627e2012-06-08 01:20:41 +0000453 const struct inetpeer_addr *daddr,
454 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000456 struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
David S. Miller98158f52010-11-30 11:41:59 -0800457 struct inet_peer *p;
Eric Dumazet65e83542011-03-04 14:33:59 -0800458 unsigned int sequence;
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000459 int invalidated, gccnt = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
David S. Millerb48c80ece2012-06-10 00:24:21 -0700461 flush_check(base, daddr->family);
462
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000463 /* Attempt a lockless lookup first.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000464 * Because of a concurrent writer, we might not find an existing entry.
465 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800466 rcu_read_lock();
Eric Dumazet65e83542011-03-04 14:33:59 -0800467 sequence = read_seqbegin(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000468 p = lookup_rcu(daddr, base);
Eric Dumazet65e83542011-03-04 14:33:59 -0800469 invalidated = read_seqretry(&base->lock, sequence);
David S. Miller7b46ac42011-03-08 14:59:28 -0800470 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000472 if (p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
Eric Dumazet65e83542011-03-04 14:33:59 -0800475 /* If no writer did a change during our lookup, we can return early. */
476 if (!create && !invalidated)
477 return NULL;
478
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000479 /* retry an exact lookup, taking the lock before.
480 * At least, nodes should be hot in our cache.
481 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800482 write_seqlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000483relookup:
David S. Miller02663042010-11-30 12:08:53 -0800484 p = lookup(daddr, stack, base);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000485 if (p != peer_avl_empty) {
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000486 atomic_inc(&p->refcnt);
Eric Dumazet65e83542011-03-04 14:33:59 -0800487 write_sequnlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000488 return p;
489 }
490 if (!gccnt) {
491 gccnt = inet_peer_gc(base, stack, stackptr);
492 if (gccnt && create)
493 goto relookup;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000494 }
495 p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
496 if (p) {
David S. Millerb534ecf2010-11-30 11:54:19 -0800497 p->daddr = *daddr;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000498 atomic_set(&p->refcnt, 1);
499 atomic_set(&p->rid, 0);
Eric Dumazet87c48fa2011-07-21 21:25:58 -0700500 atomic_set(&p->ip_id_count,
501 (daddr->family == AF_INET) ?
502 secure_ip_id(daddr->addr.a4) :
503 secure_ipv6_id(daddr->addr.a6));
David S. Miller144001b2011-01-27 13:52:16 -0800504 p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
David S. Miller92d86822011-02-04 15:55:25 -0800505 p->rate_tokens = 0;
Nicolas Dichtelbc9259a2012-09-27 04:11:00 +0000506 /* 60*HZ is arbitrary, but chosen enough high so that the first
507 * calculation of tokens is at its maximum.
508 */
509 p->rate_last = jiffies - 60*HZ;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000510 INIT_LIST_HEAD(&p->gc_list);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000511
512 /* Link the node. */
David S. Miller98158f52010-11-30 11:41:59 -0800513 link_to_pool(p, base);
514 base->total++;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000515 }
Eric Dumazet65e83542011-03-04 14:33:59 -0800516 write_sequnlock_bh(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return p;
519}
David S. Millerb3419362010-11-30 12:27:11 -0800520EXPORT_SYMBOL_GPL(inet_getpeer);
David S. Miller98158f52010-11-30 11:41:59 -0800521
Eric Dumazet4663afe2006-10-12 21:21:06 -0700522void inet_putpeer(struct inet_peer *p)
523{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000524 p->dtime = (__u32)jiffies;
Eric Dumazet6d1a3e02011-07-11 02:49:52 +0000525 smp_mb__before_atomic_dec();
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000526 atomic_dec(&p->refcnt);
Eric Dumazet4663afe2006-10-12 21:21:06 -0700527}
David S. Millerb3419362010-11-30 12:27:11 -0800528EXPORT_SYMBOL_GPL(inet_putpeer);
David S. Miller92d86822011-02-04 15:55:25 -0800529
530/*
531 * Check transmit rate limitation for given message.
532 * The rate information is held in the inet_peer entries now.
533 * This function is generic and could be used for other purposes
534 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
535 *
536 * Note that the same inet_peer fields are modified by functions in
537 * route.c too, but these work for packet destinations while xrlim_allow
538 * works for icmp destinations. This means the rate limiting information
539 * for one "ip object" is shared - and these ICMPs are twice limited:
540 * by source and by destination.
541 *
542 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
543 * SHOULD allow setting of rate limits
544 *
545 * Shared between ICMPv4 and ICMPv6.
546 */
547#define XRLIM_BURST_FACTOR 6
548bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
549{
550 unsigned long now, token;
551 bool rc = false;
552
553 if (!peer)
554 return true;
555
556 token = peer->rate_tokens;
557 now = jiffies;
558 token += now - peer->rate_last;
559 peer->rate_last = now;
560 if (token > XRLIM_BURST_FACTOR * timeout)
561 token = XRLIM_BURST_FACTOR * timeout;
562 if (token >= timeout) {
563 token -= timeout;
564 rc = true;
565 }
566 peer->rate_tokens = token;
567 return rc;
568}
569EXPORT_SYMBOL(inet_peer_xrlim_allow);
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000570
Eric Dumazet55432d22012-06-05 03:00:18 +0000571static void inetpeer_inval_rcu(struct rcu_head *head)
572{
573 struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu);
574
575 spin_lock_bh(&gc_lock);
576 list_add_tail(&p->gc_list, &gc_list);
577 spin_unlock_bh(&gc_lock);
578
579 schedule_delayed_work(&gc_work, gc_delay);
580}
581
David S. Miller56a6b242012-06-09 16:32:41 -0700582void inetpeer_invalidate_tree(struct inet_peer_base *base)
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000583{
Eric Dumazetda557372012-06-20 04:02:10 +0000584 struct inet_peer *root;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000585
586 write_seqlock_bh(&base->lock);
587
Eric Dumazetda557372012-06-20 04:02:10 +0000588 root = rcu_deref_locked(base->root, base);
589 if (root != peer_avl_empty) {
590 base->root = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000591 base->total = 0;
Eric Dumazetda557372012-06-20 04:02:10 +0000592 call_rcu(&root->gc_rcu, inetpeer_inval_rcu);
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000593 }
594
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000595 write_sequnlock_bh(&base->lock);
596}
597EXPORT_SYMBOL(inetpeer_invalidate_tree);