blob: 86fa4580954060b1798c7f3382d77cbec41e6ff7 [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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 * Nodes are removed only when reference counter goes to 0.
31 * When it's happened the node may be removed when a sufficient amount of
32 * time has been passed since its last use. The less-recently-used entry can
33 * also be removed if the pool is overloaded i.e. if the total amount of
34 * entries is greater-or-equal than the threshold.
35 *
36 * Node pool is organised as an AVL tree.
37 * Such an implementation has been chosen not just for fun. It's a way to
38 * prevent easy and efficient DoS attacks by creating hash collisions. A huge
39 * amount of long living nodes in a single hash slot would significantly delay
40 * lookups performed with disabled BHs.
41 *
42 * Serialisation issues.
Eric Dumazetaa1039e2010-06-15 08:23:14 +000043 * 1. Nodes may appear in the tree only with the pool lock held.
44 * 2. Nodes may disappear from the tree only with the pool lock held
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 * AND reference count being 0.
Eric Dumazet4b9d9be2011-06-08 13:35:34 +000046 * 3. Global variable peer_total is modified under the pool lock.
47 * 4. struct inet_peer fields modification:
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 * avl_left, avl_right, avl_parent, avl_height: pool lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 * refcnt: atomically against modifications on other CPU;
50 * usually under some other lock to prevent node disappearing
David S. Miller582a72d2010-11-30 11:53:55 -080051 * daddr: unchangeable
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 */
53
Christoph Lametere18b8902006-12-06 20:33:20 -080054static struct kmem_cache *peer_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Steffen Klassert5faa5df2012-03-06 21:20:26 +000056static LIST_HEAD(gc_list);
57static const int gc_delay = 60 * HZ;
58static struct delayed_work gc_work;
59static DEFINE_SPINLOCK(gc_lock);
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#define node_height(x) x->avl_height
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000062
63#define peer_avl_empty ((struct inet_peer *)&peer_fake_node)
Eric Dumazetb914c4e2010-10-25 23:55:38 +000064#define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node)
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000065static const struct inet_peer peer_fake_node = {
Eric Dumazetb914c4e2010-10-25 23:55:38 +000066 .avl_left = peer_avl_empty_rcu,
67 .avl_right = peer_avl_empty_rcu,
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 .avl_height = 0
69};
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000070
David S. Millerc3426b42012-06-09 16:27:05 -070071void inet_peer_base_init(struct inet_peer_base *bp)
72{
73 bp->root = peer_avl_empty_rcu;
74 seqlock_init(&bp->lock);
75 bp->total = 0;
76}
77EXPORT_SYMBOL_GPL(inet_peer_base_init);
David S. Miller021e9292010-11-30 12:12:23 -080078
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/* Exported for sysctl_net_ipv4. */
Eric Dumazet243bbca2007-03-06 20:23:10 -080082int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 * aggressively at this stage */
Eric Dumazet243bbca2007-03-06 20:23:10 -080084int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */
85int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Steffen Klassert5faa5df2012-03-06 21:20:26 +000087static void inetpeer_gc_worker(struct work_struct *work)
88{
Eric Dumazetda557372012-06-20 04:02:10 +000089 struct inet_peer *p, *n, *c;
xiao jin851bdd12014-04-25 08:53:29 +080090 struct list_head list;
Steffen Klassert5faa5df2012-03-06 21:20:26 +000091
92 spin_lock_bh(&gc_lock);
93 list_replace_init(&gc_list, &list);
94 spin_unlock_bh(&gc_lock);
95
96 if (list_empty(&list))
97 return;
98
99 list_for_each_entry_safe(p, n, &list, gc_list) {
100
Eric Dumazetda557372012-06-20 04:02:10 +0000101 if (need_resched())
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000102 cond_resched();
103
Eric Dumazetda557372012-06-20 04:02:10 +0000104 c = rcu_dereference_protected(p->avl_left, 1);
105 if (c != peer_avl_empty) {
106 list_add_tail(&c->gc_list, &list);
107 p->avl_left = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000108 }
109
Eric Dumazetda557372012-06-20 04:02:10 +0000110 c = rcu_dereference_protected(p->avl_right, 1);
111 if (c != peer_avl_empty) {
112 list_add_tail(&c->gc_list, &list);
113 p->avl_right = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000114 }
115
116 n = list_entry(p->gc_list.next, struct inet_peer, gc_list);
117
118 if (!atomic_read(&p->refcnt)) {
119 list_del(&p->gc_list);
120 kmem_cache_free(peer_cachep, p);
121 }
122 }
123
124 if (list_empty(&list))
125 return;
126
127 spin_lock_bh(&gc_lock);
128 list_splice(&list, &gc_list);
129 spin_unlock_bh(&gc_lock);
130
131 schedule_delayed_work(&gc_work, gc_delay);
132}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
134/* Called from ip_output.c:ip_init */
135void __init inet_initpeers(void)
136{
137 struct sysinfo si;
138
139 /* Use the straight interface to information about memory. */
140 si_meminfo(&si);
141 /* The values below were suggested by Alexey Kuznetsov
142 * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values
143 * myself. --SAW
144 */
145 if (si.totalram <= (32768*1024)/PAGE_SIZE)
146 inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
147 if (si.totalram <= (16384*1024)/PAGE_SIZE)
148 inet_peer_threshold >>= 1; /* about 512KB */
149 if (si.totalram <= (8192*1024)/PAGE_SIZE)
150 inet_peer_threshold >>= 2; /* about 128KB */
151
152 peer_cachep = kmem_cache_create("inet_peer_cache",
153 sizeof(struct inet_peer),
Eric Dumazet317fe0e2010-06-16 04:52:13 +0000154 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +0900155 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Tejun Heo203b42f2012-08-21 13:18:23 -0700157 INIT_DEFERRABLE_WORK(&gc_work, inetpeer_gc_worker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
Eric Dumazet65e83542011-03-04 14:33:59 -0800160#define rcu_deref_locked(X, BASE) \
161 rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))
162
Eric Dumazet243bbca2007-03-06 20:23:10 -0800163/*
164 * Called with local BH disabled and the pool lock held.
Eric Dumazet243bbca2007-03-06 20:23:10 -0800165 */
David S. Miller98158f52010-11-30 11:41:59 -0800166#define lookup(_daddr, _stack, _base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167({ \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000168 struct inet_peer *u; \
169 struct inet_peer __rcu **v; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000170 \
171 stackptr = _stack; \
David S. Miller98158f52010-11-30 11:41:59 -0800172 *stackptr++ = &_base->root; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800173 for (u = rcu_deref_locked(_base->root, _base); \
Weilong Chen0c9a67d2013-12-23 14:37:27 +0800174 u != peer_avl_empty;) { \
David Ahernd39d14f2015-08-27 16:07:01 -0700175 int cmp = inetpeer_addr_cmp(_daddr, &u->daddr); \
David S. Miller02663042010-11-30 12:08:53 -0800176 if (cmp == 0) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 break; \
David S. Miller02663042010-11-30 12:08:53 -0800178 if (cmp == -1) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 v = &u->avl_left; \
180 else \
181 v = &u->avl_right; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000182 *stackptr++ = v; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800183 u = rcu_deref_locked(*v, _base); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 } \
185 u; \
186})
187
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000188/*
David S. Miller7b46ac42011-03-08 14:59:28 -0800189 * Called with rcu_read_lock()
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000190 * Because we hold no lock against a writer, its quite possible we fall
191 * in an endless loop.
192 * But every pointer we follow is guaranteed to be valid thanks to RCU.
193 * We exit from this function if number of links exceeds PEER_MAXDEPTH
194 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800195static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000196 struct inet_peer_base *base)
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000197{
David S. Miller7b46ac42011-03-08 14:59:28 -0800198 struct inet_peer *u = rcu_dereference(base->root);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000199 int count = 0;
200
201 while (u != peer_avl_empty) {
David Ahernd39d14f2015-08-27 16:07:01 -0700202 int cmp = inetpeer_addr_cmp(daddr, &u->daddr);
David S. Miller02663042010-11-30 12:08:53 -0800203 if (cmp == 0) {
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700204 /* Before taking a reference, check if this entry was
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000205 * deleted (refcnt=-1)
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700206 */
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000207 if (!atomic_add_unless(&u->refcnt, 1, -1))
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000208 u = NULL;
209 return u;
210 }
David S. Miller02663042010-11-30 12:08:53 -0800211 if (cmp == -1)
David S. Miller7b46ac42011-03-08 14:59:28 -0800212 u = rcu_dereference(u->avl_left);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000213 else
David S. Miller7b46ac42011-03-08 14:59:28 -0800214 u = rcu_dereference(u->avl_right);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000215 if (unlikely(++count == PEER_MAXDEPTH))
216 break;
217 }
218 return NULL;
219}
220
221/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800222#define lookup_rightempty(start, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223({ \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000224 struct inet_peer *u; \
225 struct inet_peer __rcu **v; \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 *stackptr++ = &start->avl_left; \
227 v = &start->avl_left; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800228 for (u = rcu_deref_locked(*v, base); \
Weilong Chen0c9a67d2013-12-23 14:37:27 +0800229 u->avl_right != peer_avl_empty_rcu;) { \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 v = &u->avl_right; \
231 *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/* Called with local BH disabled and the pool lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 * Variable names are the proof of operation correctness.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000239 * Look into mm/map_avl.c for more detail description of the ideas.
240 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000241static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
David S. Miller98158f52010-11-30 11:41:59 -0800242 struct inet_peer __rcu ***stackend,
243 struct inet_peer_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000245 struct inet_peer __rcu **nodep;
246 struct inet_peer *node, *l, *r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 int lh, rh;
248
249 while (stackend > stack) {
250 nodep = *--stackend;
Eric Dumazet65e83542011-03-04 14:33:59 -0800251 node = rcu_deref_locked(*nodep, base);
252 l = rcu_deref_locked(node->avl_left, base);
253 r = rcu_deref_locked(node->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 lh = node_height(l);
255 rh = node_height(r);
256 if (lh > rh + 1) { /* l: RH+2 */
257 struct inet_peer *ll, *lr, *lrl, *lrr;
258 int lrh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800259 ll = rcu_deref_locked(l->avl_left, base);
260 lr = rcu_deref_locked(l->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 lrh = node_height(lr);
262 if (lrh <= node_height(ll)) { /* ll: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000263 RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */
264 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 node->avl_height = lrh + 1; /* RH+1 or RH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000266 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */
267 RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 l->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000269 RCU_INIT_POINTER(*nodep, l);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 } else { /* ll: RH, lr: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800271 lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
272 lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000273 RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */
274 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 node->avl_height = rh + 1; /* node: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000276 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */
277 RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 l->avl_height = rh + 1; /* l: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000279 RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */
280 RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 lr->avl_height = rh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000282 RCU_INIT_POINTER(*nodep, lr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 }
284 } else if (rh > lh + 1) { /* r: LH+2 */
285 struct inet_peer *rr, *rl, *rlr, *rll;
286 int rlh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800287 rr = rcu_deref_locked(r->avl_right, base);
288 rl = rcu_deref_locked(r->avl_left, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 rlh = node_height(rl);
290 if (rlh <= node_height(rr)) { /* rr: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000291 RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */
292 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 node->avl_height = rlh + 1; /* LH+1 or LH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000294 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */
295 RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 r->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000297 RCU_INIT_POINTER(*nodep, r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 } else { /* rr: RH, rl: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800299 rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
300 rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000301 RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */
302 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 node->avl_height = lh + 1; /* node: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000304 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */
305 RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 r->avl_height = lh + 1; /* r: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000307 RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */
308 RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 rl->avl_height = lh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000310 RCU_INIT_POINTER(*nodep, rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
312 } else {
313 node->avl_height = (lh > rh ? lh : rh) + 1;
314 }
315 }
316}
317
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000318/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800319#define link_to_pool(n, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320do { \
321 n->avl_height = 1; \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000322 n->avl_left = peer_avl_empty_rcu; \
323 n->avl_right = peer_avl_empty_rcu; \
324 /* lockless readers can catch us now */ \
325 rcu_assign_pointer(**--stackptr, n); \
David S. Miller98158f52010-11-30 11:41:59 -0800326 peer_avl_rebalance(stack, stackptr, base); \
Eric Dumazetd6cc1d62010-06-14 19:35:21 +0000327} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000329static void inetpeer_free_rcu(struct rcu_head *head)
330{
331 kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
332}
333
Eric Dumazet66944e12011-04-11 22:39:40 +0000334static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
335 struct inet_peer __rcu **stack[PEER_MAXDEPTH])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000337 struct inet_peer __rcu ***stackptr, ***delp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000339 if (lookup(&p->daddr, stack, base) != p)
340 BUG();
341 delp = stackptr - 1; /* *delp[0] == p */
342 if (p->avl_left == peer_avl_empty_rcu) {
343 *delp[0] = p->avl_right;
344 --stackptr;
345 } else {
346 /* look for a node to insert instead of p */
347 struct inet_peer *t;
348 t = lookup_rightempty(p, base);
349 BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
350 **--stackptr = t->avl_left;
351 /* t is removed, t->daddr > x->daddr for any
352 * x in p->avl_left subtree.
353 * Put t in the old place of p. */
354 RCU_INIT_POINTER(*delp[0], t);
355 t->avl_left = p->avl_left;
356 t->avl_right = p->avl_right;
357 t->avl_height = p->avl_height;
358 BUG_ON(delp[1] != &p->avl_left);
359 delp[1] = &t->avl_left; /* was &p->avl_left */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000361 peer_avl_rebalance(stack, stackptr, base);
362 base->total--;
363 call_rcu(&p->rcu, inetpeer_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
365
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000366/* perform garbage collect on all items stacked during a lookup */
367static int inet_peer_gc(struct inet_peer_base *base,
368 struct inet_peer __rcu **stack[PEER_MAXDEPTH],
369 struct inet_peer __rcu ***stackptr)
David S. Miller98158f52010-11-30 11:41:59 -0800370{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000371 struct inet_peer *p, *gchead = NULL;
372 __u32 delta, ttl;
373 int cnt = 0;
David S. Miller98158f52010-11-30 11:41:59 -0800374
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000375 if (base->total >= inet_peer_threshold)
376 ttl = 0; /* be aggressive */
377 else
378 ttl = inet_peer_maxttl
379 - (inet_peer_maxttl - inet_peer_minttl) / HZ *
380 base->total / inet_peer_threshold * HZ;
381 stackptr--; /* last stack slot is peer_avl_empty */
382 while (stackptr > stack) {
383 stackptr--;
384 p = rcu_deref_locked(**stackptr, base);
Eric Dumazet6d1a3e02011-07-11 02:49:52 +0000385 if (atomic_read(&p->refcnt) == 0) {
386 smp_rmb();
387 delta = (__u32)jiffies - p->dtime;
388 if (delta >= ttl &&
389 atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
390 p->gc_next = gchead;
391 gchead = p;
392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000395 while ((p = gchead) != NULL) {
396 gchead = p->gc_next;
397 cnt++;
398 unlink_from_pool(p, base, stack);
399 }
400 return cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
David S. Millerc0efc882012-06-09 19:12:36 -0700403struct inet_peer *inet_getpeer(struct inet_peer_base *base,
Gao fengc8a627e2012-06-08 01:20:41 +0000404 const struct inetpeer_addr *daddr,
405 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000407 struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
David S. Miller98158f52010-11-30 11:41:59 -0800408 struct inet_peer *p;
Eric Dumazet65e83542011-03-04 14:33:59 -0800409 unsigned int sequence;
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000410 int invalidated, gccnt = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000412 /* Attempt a lockless lookup first.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000413 * Because of a concurrent writer, we might not find an existing entry.
414 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800415 rcu_read_lock();
Eric Dumazet65e83542011-03-04 14:33:59 -0800416 sequence = read_seqbegin(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000417 p = lookup_rcu(daddr, base);
Eric Dumazet65e83542011-03-04 14:33:59 -0800418 invalidated = read_seqretry(&base->lock, sequence);
David S. Miller7b46ac42011-03-08 14:59:28 -0800419 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000421 if (p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
Eric Dumazet65e83542011-03-04 14:33:59 -0800424 /* If no writer did a change during our lookup, we can return early. */
425 if (!create && !invalidated)
426 return NULL;
427
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000428 /* retry an exact lookup, taking the lock before.
429 * At least, nodes should be hot in our cache.
430 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800431 write_seqlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000432relookup:
David S. Miller02663042010-11-30 12:08:53 -0800433 p = lookup(daddr, stack, base);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000434 if (p != peer_avl_empty) {
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000435 atomic_inc(&p->refcnt);
Eric Dumazet65e83542011-03-04 14:33:59 -0800436 write_sequnlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000437 return p;
438 }
439 if (!gccnt) {
440 gccnt = inet_peer_gc(base, stack, stackptr);
441 if (gccnt && create)
442 goto relookup;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000443 }
444 p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
445 if (p) {
David S. Millerb534ecf2010-11-30 11:54:19 -0800446 p->daddr = *daddr;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000447 atomic_set(&p->refcnt, 1);
448 atomic_set(&p->rid, 0);
David S. Miller144001b2011-01-27 13:52:16 -0800449 p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
David S. Miller92d86822011-02-04 15:55:25 -0800450 p->rate_tokens = 0;
Nicolas Dichtelbc9259a2012-09-27 04:11:00 +0000451 /* 60*HZ is arbitrary, but chosen enough high so that the first
452 * calculation of tokens is at its maximum.
453 */
454 p->rate_last = jiffies - 60*HZ;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000455 INIT_LIST_HEAD(&p->gc_list);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000456
457 /* Link the node. */
David S. Miller98158f52010-11-30 11:41:59 -0800458 link_to_pool(p, base);
459 base->total++;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000460 }
Eric Dumazet65e83542011-03-04 14:33:59 -0800461 write_sequnlock_bh(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return p;
464}
David S. Millerb3419362010-11-30 12:27:11 -0800465EXPORT_SYMBOL_GPL(inet_getpeer);
David S. Miller98158f52010-11-30 11:41:59 -0800466
Eric Dumazet4663afe2006-10-12 21:21:06 -0700467void inet_putpeer(struct inet_peer *p)
468{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000469 p->dtime = (__u32)jiffies;
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100470 smp_mb__before_atomic();
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000471 atomic_dec(&p->refcnt);
Eric Dumazet4663afe2006-10-12 21:21:06 -0700472}
David S. Millerb3419362010-11-30 12:27:11 -0800473EXPORT_SYMBOL_GPL(inet_putpeer);
David S. Miller92d86822011-02-04 15:55:25 -0800474
475/*
476 * Check transmit rate limitation for given message.
477 * The rate information is held in the inet_peer entries now.
478 * This function is generic and could be used for other purposes
479 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
480 *
481 * Note that the same inet_peer fields are modified by functions in
482 * route.c too, but these work for packet destinations while xrlim_allow
483 * works for icmp destinations. This means the rate limiting information
484 * for one "ip object" is shared - and these ICMPs are twice limited:
485 * by source and by destination.
486 *
487 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
488 * SHOULD allow setting of rate limits
489 *
490 * Shared between ICMPv4 and ICMPv6.
491 */
492#define XRLIM_BURST_FACTOR 6
493bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
494{
495 unsigned long now, token;
496 bool rc = false;
497
498 if (!peer)
499 return true;
500
501 token = peer->rate_tokens;
502 now = jiffies;
503 token += now - peer->rate_last;
504 peer->rate_last = now;
505 if (token > XRLIM_BURST_FACTOR * timeout)
506 token = XRLIM_BURST_FACTOR * timeout;
507 if (token >= timeout) {
508 token -= timeout;
509 rc = true;
510 }
511 peer->rate_tokens = token;
512 return rc;
513}
514EXPORT_SYMBOL(inet_peer_xrlim_allow);
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000515
Eric Dumazet55432d22012-06-05 03:00:18 +0000516static void inetpeer_inval_rcu(struct rcu_head *head)
517{
518 struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu);
519
520 spin_lock_bh(&gc_lock);
521 list_add_tail(&p->gc_list, &gc_list);
522 spin_unlock_bh(&gc_lock);
523
524 schedule_delayed_work(&gc_work, gc_delay);
525}
526
David S. Miller56a6b242012-06-09 16:32:41 -0700527void inetpeer_invalidate_tree(struct inet_peer_base *base)
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000528{
Eric Dumazetda557372012-06-20 04:02:10 +0000529 struct inet_peer *root;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000530
531 write_seqlock_bh(&base->lock);
532
Eric Dumazetda557372012-06-20 04:02:10 +0000533 root = rcu_deref_locked(base->root, base);
534 if (root != peer_avl_empty) {
535 base->root = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000536 base->total = 0;
Eric Dumazetda557372012-06-20 04:02:10 +0000537 call_rcu(&root->gc_rcu, inetpeer_inval_rcu);
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000538 }
539
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000540 write_sequnlock_bh(&base->lock);
541}
542EXPORT_SYMBOL(inetpeer_invalidate_tree);