blob: d4d61b694fab9bc497b1cccb808a3a568ad30cc2 [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
35 * PMTU in size uses a constant ID and do not use this code (see
36 * ip_select_ident() in include/net/ip.h).
37 *
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. Miller021e9292010-11-30 12:12:23 -080085struct inet_peer_base {
Eric Dumazetb914c4e2010-10-25 23:55:38 +000086 struct inet_peer __rcu *root;
Eric Dumazet65e83542011-03-04 14:33:59 -080087 seqlock_t lock;
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000088 int total;
David S. Miller021e9292010-11-30 12:12:23 -080089};
90
91static struct inet_peer_base v4_peers = {
Eric Dumazetb914c4e2010-10-25 23:55:38 +000092 .root = peer_avl_empty_rcu,
Eric Dumazet65e83542011-03-04 14:33:59 -080093 .lock = __SEQLOCK_UNLOCKED(v4_peers.lock),
Eric Dumazetd6cc1d62010-06-14 19:35:21 +000094 .total = 0,
95};
David S. Miller021e9292010-11-30 12:12:23 -080096
97static struct inet_peer_base v6_peers = {
98 .root = peer_avl_empty_rcu,
Eric Dumazet65e83542011-03-04 14:33:59 -080099 .lock = __SEQLOCK_UNLOCKED(v6_peers.lock),
David S. Miller021e9292010-11-30 12:12:23 -0800100 .total = 0,
101};
102
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105/* Exported for sysctl_net_ipv4. */
Eric Dumazet243bbca2007-03-06 20:23:10 -0800106int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * aggressively at this stage */
Eric Dumazet243bbca2007-03-06 20:23:10 -0800108int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */
109int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000111static void inetpeer_gc_worker(struct work_struct *work)
112{
113 struct inet_peer *p, *n;
114 LIST_HEAD(list);
115
116 spin_lock_bh(&gc_lock);
117 list_replace_init(&gc_list, &list);
118 spin_unlock_bh(&gc_lock);
119
120 if (list_empty(&list))
121 return;
122
123 list_for_each_entry_safe(p, n, &list, gc_list) {
124
125 if(need_resched())
126 cond_resched();
127
128 if (p->avl_left != peer_avl_empty) {
129 list_add_tail(&p->avl_left->gc_list, &list);
130 p->avl_left = peer_avl_empty;
131 }
132
133 if (p->avl_right != peer_avl_empty) {
134 list_add_tail(&p->avl_right->gc_list, &list);
135 p->avl_right = peer_avl_empty;
136 }
137
138 n = list_entry(p->gc_list.next, struct inet_peer, gc_list);
139
140 if (!atomic_read(&p->refcnt)) {
141 list_del(&p->gc_list);
142 kmem_cache_free(peer_cachep, p);
143 }
144 }
145
146 if (list_empty(&list))
147 return;
148
149 spin_lock_bh(&gc_lock);
150 list_splice(&list, &gc_list);
151 spin_unlock_bh(&gc_lock);
152
153 schedule_delayed_work(&gc_work, gc_delay);
154}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
156/* Called from ip_output.c:ip_init */
157void __init inet_initpeers(void)
158{
159 struct sysinfo si;
160
161 /* Use the straight interface to information about memory. */
162 si_meminfo(&si);
163 /* The values below were suggested by Alexey Kuznetsov
164 * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values
165 * myself. --SAW
166 */
167 if (si.totalram <= (32768*1024)/PAGE_SIZE)
168 inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
169 if (si.totalram <= (16384*1024)/PAGE_SIZE)
170 inet_peer_threshold >>= 1; /* about 512KB */
171 if (si.totalram <= (8192*1024)/PAGE_SIZE)
172 inet_peer_threshold >>= 2; /* about 128KB */
173
174 peer_cachep = kmem_cache_create("inet_peer_cache",
175 sizeof(struct inet_peer),
Eric Dumazet317fe0e2010-06-16 04:52:13 +0000176 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +0900177 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000179 INIT_DELAYED_WORK_DEFERRABLE(&gc_work, inetpeer_gc_worker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180}
181
David S. Miller8790ca12010-12-01 17:28:18 -0800182static int addr_compare(const struct inetpeer_addr *a,
183 const struct inetpeer_addr *b)
David S. Miller02663042010-11-30 12:08:53 -0800184{
185 int i, n = (a->family == AF_INET ? 1 : 4);
186
187 for (i = 0; i < n; i++) {
David S. Miller7a71ed82011-02-09 14:30:26 -0800188 if (a->addr.a6[i] == b->addr.a6[i])
David S. Miller02663042010-11-30 12:08:53 -0800189 continue;
Eric Dumazet747465e2012-01-16 19:27:39 +0000190 if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i])
David S. Miller02663042010-11-30 12:08:53 -0800191 return -1;
192 return 1;
193 }
194
195 return 0;
196}
197
Eric Dumazet65e83542011-03-04 14:33:59 -0800198#define rcu_deref_locked(X, BASE) \
199 rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))
200
Eric Dumazet243bbca2007-03-06 20:23:10 -0800201/*
202 * Called with local BH disabled and the pool lock held.
Eric Dumazet243bbca2007-03-06 20:23:10 -0800203 */
David S. Miller98158f52010-11-30 11:41:59 -0800204#define lookup(_daddr, _stack, _base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205({ \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000206 struct inet_peer *u; \
207 struct inet_peer __rcu **v; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000208 \
209 stackptr = _stack; \
David S. Miller98158f52010-11-30 11:41:59 -0800210 *stackptr++ = &_base->root; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800211 for (u = rcu_deref_locked(_base->root, _base); \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000212 u != peer_avl_empty; ) { \
David S. Miller02663042010-11-30 12:08:53 -0800213 int cmp = addr_compare(_daddr, &u->daddr); \
214 if (cmp == 0) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 break; \
David S. Miller02663042010-11-30 12:08:53 -0800216 if (cmp == -1) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 v = &u->avl_left; \
218 else \
219 v = &u->avl_right; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000220 *stackptr++ = v; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800221 u = rcu_deref_locked(*v, _base); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 } \
223 u; \
224})
225
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000226/*
David S. Miller7b46ac42011-03-08 14:59:28 -0800227 * Called with rcu_read_lock()
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000228 * Because we hold no lock against a writer, its quite possible we fall
229 * in an endless loop.
230 * But every pointer we follow is guaranteed to be valid thanks to RCU.
231 * We exit from this function if number of links exceeds PEER_MAXDEPTH
232 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800233static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000234 struct inet_peer_base *base)
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000235{
David S. Miller7b46ac42011-03-08 14:59:28 -0800236 struct inet_peer *u = rcu_dereference(base->root);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000237 int count = 0;
238
239 while (u != peer_avl_empty) {
David S. Miller02663042010-11-30 12:08:53 -0800240 int cmp = addr_compare(daddr, &u->daddr);
241 if (cmp == 0) {
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700242 /* Before taking a reference, check if this entry was
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000243 * deleted (refcnt=-1)
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700244 */
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000245 if (!atomic_add_unless(&u->refcnt, 1, -1))
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000246 u = NULL;
247 return u;
248 }
David S. Miller02663042010-11-30 12:08:53 -0800249 if (cmp == -1)
David S. Miller7b46ac42011-03-08 14:59:28 -0800250 u = rcu_dereference(u->avl_left);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000251 else
David S. Miller7b46ac42011-03-08 14:59:28 -0800252 u = rcu_dereference(u->avl_right);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000253 if (unlikely(++count == PEER_MAXDEPTH))
254 break;
255 }
256 return NULL;
257}
258
259/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800260#define lookup_rightempty(start, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261({ \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000262 struct inet_peer *u; \
263 struct inet_peer __rcu **v; \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 *stackptr++ = &start->avl_left; \
265 v = &start->avl_left; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800266 for (u = rcu_deref_locked(*v, base); \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000267 u->avl_right != peer_avl_empty_rcu; ) { \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 v = &u->avl_right; \
269 *stackptr++ = v; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800270 u = rcu_deref_locked(*v, base); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 } \
272 u; \
273})
274
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000275/* Called with local BH disabled and the pool lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 * Variable names are the proof of operation correctness.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000277 * Look into mm/map_avl.c for more detail description of the ideas.
278 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000279static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
David S. Miller98158f52010-11-30 11:41:59 -0800280 struct inet_peer __rcu ***stackend,
281 struct inet_peer_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000283 struct inet_peer __rcu **nodep;
284 struct inet_peer *node, *l, *r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 int lh, rh;
286
287 while (stackend > stack) {
288 nodep = *--stackend;
Eric Dumazet65e83542011-03-04 14:33:59 -0800289 node = rcu_deref_locked(*nodep, base);
290 l = rcu_deref_locked(node->avl_left, base);
291 r = rcu_deref_locked(node->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 lh = node_height(l);
293 rh = node_height(r);
294 if (lh > rh + 1) { /* l: RH+2 */
295 struct inet_peer *ll, *lr, *lrl, *lrr;
296 int lrh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800297 ll = rcu_deref_locked(l->avl_left, base);
298 lr = rcu_deref_locked(l->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 lrh = node_height(lr);
300 if (lrh <= node_height(ll)) { /* ll: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000301 RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */
302 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 node->avl_height = lrh + 1; /* RH+1 or RH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000304 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */
305 RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 l->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000307 RCU_INIT_POINTER(*nodep, l);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 } else { /* ll: RH, lr: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800309 lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
310 lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000311 RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */
312 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 node->avl_height = rh + 1; /* node: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000314 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */
315 RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 l->avl_height = rh + 1; /* l: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000317 RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */
318 RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 lr->avl_height = rh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000320 RCU_INIT_POINTER(*nodep, lr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 }
322 } else if (rh > lh + 1) { /* r: LH+2 */
323 struct inet_peer *rr, *rl, *rlr, *rll;
324 int rlh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800325 rr = rcu_deref_locked(r->avl_right, base);
326 rl = rcu_deref_locked(r->avl_left, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 rlh = node_height(rl);
328 if (rlh <= node_height(rr)) { /* rr: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000329 RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */
330 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 node->avl_height = rlh + 1; /* LH+1 or LH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000332 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */
333 RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 r->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000335 RCU_INIT_POINTER(*nodep, r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 } else { /* rr: RH, rl: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800337 rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
338 rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000339 RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */
340 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 node->avl_height = lh + 1; /* node: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000342 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */
343 RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 r->avl_height = lh + 1; /* r: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000345 RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */
346 RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 rl->avl_height = lh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000348 RCU_INIT_POINTER(*nodep, rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 }
350 } else {
351 node->avl_height = (lh > rh ? lh : rh) + 1;
352 }
353 }
354}
355
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000356/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800357#define link_to_pool(n, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358do { \
359 n->avl_height = 1; \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000360 n->avl_left = peer_avl_empty_rcu; \
361 n->avl_right = peer_avl_empty_rcu; \
362 /* lockless readers can catch us now */ \
363 rcu_assign_pointer(**--stackptr, n); \
David S. Miller98158f52010-11-30 11:41:59 -0800364 peer_avl_rebalance(stack, stackptr, base); \
Eric Dumazetd6cc1d62010-06-14 19:35:21 +0000365} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000367static void inetpeer_free_rcu(struct rcu_head *head)
368{
369 kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
370}
371
Eric Dumazet66944e12011-04-11 22:39:40 +0000372static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
373 struct inet_peer __rcu **stack[PEER_MAXDEPTH])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000375 struct inet_peer __rcu ***stackptr, ***delp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000377 if (lookup(&p->daddr, stack, base) != p)
378 BUG();
379 delp = stackptr - 1; /* *delp[0] == p */
380 if (p->avl_left == peer_avl_empty_rcu) {
381 *delp[0] = p->avl_right;
382 --stackptr;
383 } else {
384 /* look for a node to insert instead of p */
385 struct inet_peer *t;
386 t = lookup_rightempty(p, base);
387 BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
388 **--stackptr = t->avl_left;
389 /* t is removed, t->daddr > x->daddr for any
390 * x in p->avl_left subtree.
391 * Put t in the old place of p. */
392 RCU_INIT_POINTER(*delp[0], t);
393 t->avl_left = p->avl_left;
394 t->avl_right = p->avl_right;
395 t->avl_height = p->avl_height;
396 BUG_ON(delp[1] != &p->avl_left);
397 delp[1] = &t->avl_left; /* was &p->avl_left */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000399 peer_avl_rebalance(stack, stackptr, base);
400 base->total--;
401 call_rcu(&p->rcu, inetpeer_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
David S. Miller021e9292010-11-30 12:12:23 -0800404static struct inet_peer_base *family_to_base(int family)
405{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000406 return family == AF_INET ? &v4_peers : &v6_peers;
David S. Miller021e9292010-11-30 12:12:23 -0800407}
408
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000409/* perform garbage collect on all items stacked during a lookup */
410static int inet_peer_gc(struct inet_peer_base *base,
411 struct inet_peer __rcu **stack[PEER_MAXDEPTH],
412 struct inet_peer __rcu ***stackptr)
David S. Miller98158f52010-11-30 11:41:59 -0800413{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000414 struct inet_peer *p, *gchead = NULL;
415 __u32 delta, ttl;
416 int cnt = 0;
David S. Miller98158f52010-11-30 11:41:59 -0800417
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000418 if (base->total >= inet_peer_threshold)
419 ttl = 0; /* be aggressive */
420 else
421 ttl = inet_peer_maxttl
422 - (inet_peer_maxttl - inet_peer_minttl) / HZ *
423 base->total / inet_peer_threshold * HZ;
424 stackptr--; /* last stack slot is peer_avl_empty */
425 while (stackptr > stack) {
426 stackptr--;
427 p = rcu_deref_locked(**stackptr, base);
Eric Dumazet6d1a3e02011-07-11 02:49:52 +0000428 if (atomic_read(&p->refcnt) == 0) {
429 smp_rmb();
430 delta = (__u32)jiffies - p->dtime;
431 if (delta >= ttl &&
432 atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
433 p->gc_next = gchead;
434 gchead = p;
435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000438 while ((p = gchead) != NULL) {
439 gchead = p->gc_next;
440 cnt++;
441 unlink_from_pool(p, base, stack);
442 }
443 return cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444}
445
Eric Dumazet87c48fa2011-07-21 21:25:58 -0700446struct inet_peer *inet_getpeer(const struct inetpeer_addr *daddr, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000448 struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
David S. Miller34084042011-01-24 14:37:46 -0800449 struct inet_peer_base *base = family_to_base(daddr->family);
David S. Miller98158f52010-11-30 11:41:59 -0800450 struct inet_peer *p;
Eric Dumazet65e83542011-03-04 14:33:59 -0800451 unsigned int sequence;
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000452 int invalidated, gccnt = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000454 /* Attempt a lockless lookup first.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000455 * Because of a concurrent writer, we might not find an existing entry.
456 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800457 rcu_read_lock();
Eric Dumazet65e83542011-03-04 14:33:59 -0800458 sequence = read_seqbegin(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000459 p = lookup_rcu(daddr, base);
Eric Dumazet65e83542011-03-04 14:33:59 -0800460 invalidated = read_seqretry(&base->lock, sequence);
David S. Miller7b46ac42011-03-08 14:59:28 -0800461 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000463 if (p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Eric Dumazet65e83542011-03-04 14:33:59 -0800466 /* If no writer did a change during our lookup, we can return early. */
467 if (!create && !invalidated)
468 return NULL;
469
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000470 /* retry an exact lookup, taking the lock before.
471 * At least, nodes should be hot in our cache.
472 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800473 write_seqlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000474relookup:
David S. Miller02663042010-11-30 12:08:53 -0800475 p = lookup(daddr, stack, base);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000476 if (p != peer_avl_empty) {
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000477 atomic_inc(&p->refcnt);
Eric Dumazet65e83542011-03-04 14:33:59 -0800478 write_sequnlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000479 return p;
480 }
481 if (!gccnt) {
482 gccnt = inet_peer_gc(base, stack, stackptr);
483 if (gccnt && create)
484 goto relookup;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000485 }
486 p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
487 if (p) {
David S. Millerb534ecf2010-11-30 11:54:19 -0800488 p->daddr = *daddr;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000489 atomic_set(&p->refcnt, 1);
490 atomic_set(&p->rid, 0);
Eric Dumazet87c48fa2011-07-21 21:25:58 -0700491 atomic_set(&p->ip_id_count,
492 (daddr->family == AF_INET) ?
493 secure_ip_id(daddr->addr.a4) :
494 secure_ipv6_id(daddr->addr.a6));
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000495 p->tcp_ts_stamp = 0;
David S. Miller144001b2011-01-27 13:52:16 -0800496 p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
David S. Miller92d86822011-02-04 15:55:25 -0800497 p->rate_tokens = 0;
498 p->rate_last = 0;
David S. Millerddd4aa42011-02-09 15:36:47 -0800499 p->pmtu_expires = 0;
Hiroaki SHIMODA46af3182011-03-09 20:09:58 +0000500 p->pmtu_orig = 0;
David S. Millerddd4aa42011-02-09 15:36:47 -0800501 memset(&p->redirect_learned, 0, sizeof(p->redirect_learned));
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000502 INIT_LIST_HEAD(&p->gc_list);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000503
504 /* Link the node. */
David S. Miller98158f52010-11-30 11:41:59 -0800505 link_to_pool(p, base);
506 base->total++;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000507 }
Eric Dumazet65e83542011-03-04 14:33:59 -0800508 write_sequnlock_bh(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return p;
511}
David S. Millerb3419362010-11-30 12:27:11 -0800512EXPORT_SYMBOL_GPL(inet_getpeer);
David S. Miller98158f52010-11-30 11:41:59 -0800513
Eric Dumazet4663afe2006-10-12 21:21:06 -0700514void inet_putpeer(struct inet_peer *p)
515{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000516 p->dtime = (__u32)jiffies;
Eric Dumazet6d1a3e02011-07-11 02:49:52 +0000517 smp_mb__before_atomic_dec();
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000518 atomic_dec(&p->refcnt);
Eric Dumazet4663afe2006-10-12 21:21:06 -0700519}
David S. Millerb3419362010-11-30 12:27:11 -0800520EXPORT_SYMBOL_GPL(inet_putpeer);
David S. Miller92d86822011-02-04 15:55:25 -0800521
522/*
523 * Check transmit rate limitation for given message.
524 * The rate information is held in the inet_peer entries now.
525 * This function is generic and could be used for other purposes
526 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
527 *
528 * Note that the same inet_peer fields are modified by functions in
529 * route.c too, but these work for packet destinations while xrlim_allow
530 * works for icmp destinations. This means the rate limiting information
531 * for one "ip object" is shared - and these ICMPs are twice limited:
532 * by source and by destination.
533 *
534 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
535 * SHOULD allow setting of rate limits
536 *
537 * Shared between ICMPv4 and ICMPv6.
538 */
539#define XRLIM_BURST_FACTOR 6
540bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
541{
542 unsigned long now, token;
543 bool rc = false;
544
545 if (!peer)
546 return true;
547
548 token = peer->rate_tokens;
549 now = jiffies;
550 token += now - peer->rate_last;
551 peer->rate_last = now;
552 if (token > XRLIM_BURST_FACTOR * timeout)
553 token = XRLIM_BURST_FACTOR * timeout;
554 if (token >= timeout) {
555 token -= timeout;
556 rc = true;
557 }
558 peer->rate_tokens = token;
559 return rc;
560}
561EXPORT_SYMBOL(inet_peer_xrlim_allow);
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000562
563void inetpeer_invalidate_tree(int family)
564{
565 struct inet_peer *old, *new, *prev;
566 struct inet_peer_base *base = family_to_base(family);
567
568 write_seqlock_bh(&base->lock);
569
570 old = base->root;
571 if (old == peer_avl_empty_rcu)
572 goto out;
573
574 new = peer_avl_empty_rcu;
575
576 prev = cmpxchg(&base->root, old, new);
577 if (prev == old) {
578 base->total = 0;
579 spin_lock(&gc_lock);
580 list_add_tail(&prev->gc_list, &gc_list);
581 spin_unlock(&gc_lock);
582 schedule_delayed_work(&gc_work, gc_delay);
583 }
584
585out:
586 write_sequnlock_bh(&base->lock);
587}
588EXPORT_SYMBOL(inetpeer_invalidate_tree);