blob: 33d5537881ed7b39e33199dfe978bbe5b912d706 [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
112void inetpeer_invalidate_family(int family)
113{
114 atomic_t *fp = inetpeer_seq_ptr(family);
115
116 atomic_inc(fp);
117}
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121/* Exported for sysctl_net_ipv4. */
Eric Dumazet243bbca2007-03-06 20:23:10 -0800122int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 * aggressively at this stage */
Eric Dumazet243bbca2007-03-06 20:23:10 -0800124int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */
125int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000127static void inetpeer_gc_worker(struct work_struct *work)
128{
Eric Dumazetda557372012-06-20 04:02:10 +0000129 struct inet_peer *p, *n, *c;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000130 LIST_HEAD(list);
131
132 spin_lock_bh(&gc_lock);
133 list_replace_init(&gc_list, &list);
134 spin_unlock_bh(&gc_lock);
135
136 if (list_empty(&list))
137 return;
138
139 list_for_each_entry_safe(p, n, &list, gc_list) {
140
Eric Dumazetda557372012-06-20 04:02:10 +0000141 if (need_resched())
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000142 cond_resched();
143
Eric Dumazetda557372012-06-20 04:02:10 +0000144 c = rcu_dereference_protected(p->avl_left, 1);
145 if (c != peer_avl_empty) {
146 list_add_tail(&c->gc_list, &list);
147 p->avl_left = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000148 }
149
Eric Dumazetda557372012-06-20 04:02:10 +0000150 c = rcu_dereference_protected(p->avl_right, 1);
151 if (c != peer_avl_empty) {
152 list_add_tail(&c->gc_list, &list);
153 p->avl_right = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000154 }
155
156 n = list_entry(p->gc_list.next, struct inet_peer, gc_list);
157
158 if (!atomic_read(&p->refcnt)) {
159 list_del(&p->gc_list);
160 kmem_cache_free(peer_cachep, p);
161 }
162 }
163
164 if (list_empty(&list))
165 return;
166
167 spin_lock_bh(&gc_lock);
168 list_splice(&list, &gc_list);
169 spin_unlock_bh(&gc_lock);
170
171 schedule_delayed_work(&gc_work, gc_delay);
172}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174/* Called from ip_output.c:ip_init */
175void __init inet_initpeers(void)
176{
177 struct sysinfo si;
178
179 /* Use the straight interface to information about memory. */
180 si_meminfo(&si);
181 /* The values below were suggested by Alexey Kuznetsov
182 * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values
183 * myself. --SAW
184 */
185 if (si.totalram <= (32768*1024)/PAGE_SIZE)
186 inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
187 if (si.totalram <= (16384*1024)/PAGE_SIZE)
188 inet_peer_threshold >>= 1; /* about 512KB */
189 if (si.totalram <= (8192*1024)/PAGE_SIZE)
190 inet_peer_threshold >>= 2; /* about 128KB */
191
192 peer_cachep = kmem_cache_create("inet_peer_cache",
193 sizeof(struct inet_peer),
Eric Dumazet317fe0e2010-06-16 04:52:13 +0000194 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +0900195 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Tejun Heo203b42f2012-08-21 13:18:23 -0700197 INIT_DEFERRABLE_WORK(&gc_work, inetpeer_gc_worker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198}
199
David S. Miller8790ca12010-12-01 17:28:18 -0800200static int addr_compare(const struct inetpeer_addr *a,
201 const struct inetpeer_addr *b)
David S. Miller02663042010-11-30 12:08:53 -0800202{
203 int i, n = (a->family == AF_INET ? 1 : 4);
204
205 for (i = 0; i < n; i++) {
David S. Miller7a71ed82011-02-09 14:30:26 -0800206 if (a->addr.a6[i] == b->addr.a6[i])
David S. Miller02663042010-11-30 12:08:53 -0800207 continue;
Eric Dumazet747465e2012-01-16 19:27:39 +0000208 if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i])
David S. Miller02663042010-11-30 12:08:53 -0800209 return -1;
210 return 1;
211 }
212
213 return 0;
214}
215
Eric Dumazet65e83542011-03-04 14:33:59 -0800216#define rcu_deref_locked(X, BASE) \
217 rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))
218
Eric Dumazet243bbca2007-03-06 20:23:10 -0800219/*
220 * Called with local BH disabled and the pool lock held.
Eric Dumazet243bbca2007-03-06 20:23:10 -0800221 */
David S. Miller98158f52010-11-30 11:41:59 -0800222#define lookup(_daddr, _stack, _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; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000226 \
227 stackptr = _stack; \
David S. Miller98158f52010-11-30 11:41:59 -0800228 *stackptr++ = &_base->root; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800229 for (u = rcu_deref_locked(_base->root, _base); \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000230 u != peer_avl_empty; ) { \
David S. Miller02663042010-11-30 12:08:53 -0800231 int cmp = addr_compare(_daddr, &u->daddr); \
232 if (cmp == 0) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 break; \
David S. Miller02663042010-11-30 12:08:53 -0800234 if (cmp == -1) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 v = &u->avl_left; \
236 else \
237 v = &u->avl_right; \
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000238 *stackptr++ = v; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800239 u = rcu_deref_locked(*v, _base); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 } \
241 u; \
242})
243
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000244/*
David S. Miller7b46ac42011-03-08 14:59:28 -0800245 * Called with rcu_read_lock()
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000246 * Because we hold no lock against a writer, its quite possible we fall
247 * in an endless loop.
248 * But every pointer we follow is guaranteed to be valid thanks to RCU.
249 * We exit from this function if number of links exceeds PEER_MAXDEPTH
250 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800251static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000252 struct inet_peer_base *base)
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000253{
David S. Miller7b46ac42011-03-08 14:59:28 -0800254 struct inet_peer *u = rcu_dereference(base->root);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000255 int count = 0;
256
257 while (u != peer_avl_empty) {
David S. Miller02663042010-11-30 12:08:53 -0800258 int cmp = addr_compare(daddr, &u->daddr);
259 if (cmp == 0) {
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700260 /* Before taking a reference, check if this entry was
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000261 * deleted (refcnt=-1)
Eric Dumazet5f2f8922010-06-15 21:47:39 -0700262 */
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000263 if (!atomic_add_unless(&u->refcnt, 1, -1))
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000264 u = NULL;
265 return u;
266 }
David S. Miller02663042010-11-30 12:08:53 -0800267 if (cmp == -1)
David S. Miller7b46ac42011-03-08 14:59:28 -0800268 u = rcu_dereference(u->avl_left);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000269 else
David S. Miller7b46ac42011-03-08 14:59:28 -0800270 u = rcu_dereference(u->avl_right);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000271 if (unlikely(++count == PEER_MAXDEPTH))
272 break;
273 }
274 return NULL;
275}
276
277/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800278#define lookup_rightempty(start, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279({ \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000280 struct inet_peer *u; \
281 struct inet_peer __rcu **v; \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 *stackptr++ = &start->avl_left; \
283 v = &start->avl_left; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800284 for (u = rcu_deref_locked(*v, base); \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000285 u->avl_right != peer_avl_empty_rcu; ) { \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 v = &u->avl_right; \
287 *stackptr++ = v; \
Eric Dumazet65e83542011-03-04 14:33:59 -0800288 u = rcu_deref_locked(*v, base); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 } \
290 u; \
291})
292
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000293/* Called with local BH disabled and the pool lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * Variable names are the proof of operation correctness.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000295 * Look into mm/map_avl.c for more detail description of the ideas.
296 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000297static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
David S. Miller98158f52010-11-30 11:41:59 -0800298 struct inet_peer __rcu ***stackend,
299 struct inet_peer_base *base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000301 struct inet_peer __rcu **nodep;
302 struct inet_peer *node, *l, *r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 int lh, rh;
304
305 while (stackend > stack) {
306 nodep = *--stackend;
Eric Dumazet65e83542011-03-04 14:33:59 -0800307 node = rcu_deref_locked(*nodep, base);
308 l = rcu_deref_locked(node->avl_left, base);
309 r = rcu_deref_locked(node->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 lh = node_height(l);
311 rh = node_height(r);
312 if (lh > rh + 1) { /* l: RH+2 */
313 struct inet_peer *ll, *lr, *lrl, *lrr;
314 int lrh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800315 ll = rcu_deref_locked(l->avl_left, base);
316 lr = rcu_deref_locked(l->avl_right, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 lrh = node_height(lr);
318 if (lrh <= node_height(ll)) { /* ll: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000319 RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */
320 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 node->avl_height = lrh + 1; /* RH+1 or RH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000322 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */
323 RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 l->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000325 RCU_INIT_POINTER(*nodep, l);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 } else { /* ll: RH, lr: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800327 lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
328 lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000329 RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */
330 RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 node->avl_height = rh + 1; /* node: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000332 RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */
333 RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 l->avl_height = rh + 1; /* l: RH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000335 RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */
336 RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 lr->avl_height = rh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000338 RCU_INIT_POINTER(*nodep, lr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
340 } else if (rh > lh + 1) { /* r: LH+2 */
341 struct inet_peer *rr, *rl, *rlr, *rll;
342 int rlh;
Eric Dumazet65e83542011-03-04 14:33:59 -0800343 rr = rcu_deref_locked(r->avl_right, base);
344 rl = rcu_deref_locked(r->avl_left, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 rlh = node_height(rl);
346 if (rlh <= node_height(rr)) { /* rr: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000347 RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */
348 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 node->avl_height = rlh + 1; /* LH+1 or LH+2 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000350 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */
351 RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 r->avl_height = node->avl_height + 1;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000353 RCU_INIT_POINTER(*nodep, r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 } else { /* rr: RH, rl: RH+1 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800355 rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
356 rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000357 RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */
358 RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 node->avl_height = lh + 1; /* node: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000360 RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */
361 RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 r->avl_height = lh + 1; /* r: LH+1 */
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000363 RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */
364 RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 rl->avl_height = lh + 2;
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000366 RCU_INIT_POINTER(*nodep, rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
368 } else {
369 node->avl_height = (lh > rh ? lh : rh) + 1;
370 }
371 }
372}
373
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000374/* Called with local BH disabled and the pool lock held. */
David S. Miller98158f52010-11-30 11:41:59 -0800375#define link_to_pool(n, base) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376do { \
377 n->avl_height = 1; \
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000378 n->avl_left = peer_avl_empty_rcu; \
379 n->avl_right = peer_avl_empty_rcu; \
380 /* lockless readers can catch us now */ \
381 rcu_assign_pointer(**--stackptr, n); \
David S. Miller98158f52010-11-30 11:41:59 -0800382 peer_avl_rebalance(stack, stackptr, base); \
Eric Dumazetd6cc1d62010-06-14 19:35:21 +0000383} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000385static void inetpeer_free_rcu(struct rcu_head *head)
386{
387 kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
388}
389
Eric Dumazet66944e12011-04-11 22:39:40 +0000390static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
391 struct inet_peer __rcu **stack[PEER_MAXDEPTH])
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000393 struct inet_peer __rcu ***stackptr, ***delp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000395 if (lookup(&p->daddr, stack, base) != p)
396 BUG();
397 delp = stackptr - 1; /* *delp[0] == p */
398 if (p->avl_left == peer_avl_empty_rcu) {
399 *delp[0] = p->avl_right;
400 --stackptr;
401 } else {
402 /* look for a node to insert instead of p */
403 struct inet_peer *t;
404 t = lookup_rightempty(p, base);
405 BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
406 **--stackptr = t->avl_left;
407 /* t is removed, t->daddr > x->daddr for any
408 * x in p->avl_left subtree.
409 * Put t in the old place of p. */
410 RCU_INIT_POINTER(*delp[0], t);
411 t->avl_left = p->avl_left;
412 t->avl_right = p->avl_right;
413 t->avl_height = p->avl_height;
414 BUG_ON(delp[1] != &p->avl_left);
415 delp[1] = &t->avl_left; /* was &p->avl_left */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000417 peer_avl_rebalance(stack, stackptr, base);
418 base->total--;
419 call_rcu(&p->rcu, inetpeer_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000422/* perform garbage collect on all items stacked during a lookup */
423static int inet_peer_gc(struct inet_peer_base *base,
424 struct inet_peer __rcu **stack[PEER_MAXDEPTH],
425 struct inet_peer __rcu ***stackptr)
David S. Miller98158f52010-11-30 11:41:59 -0800426{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000427 struct inet_peer *p, *gchead = NULL;
428 __u32 delta, ttl;
429 int cnt = 0;
David S. Miller98158f52010-11-30 11:41:59 -0800430
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000431 if (base->total >= inet_peer_threshold)
432 ttl = 0; /* be aggressive */
433 else
434 ttl = inet_peer_maxttl
435 - (inet_peer_maxttl - inet_peer_minttl) / HZ *
436 base->total / inet_peer_threshold * HZ;
437 stackptr--; /* last stack slot is peer_avl_empty */
438 while (stackptr > stack) {
439 stackptr--;
440 p = rcu_deref_locked(**stackptr, base);
Eric Dumazet6d1a3e02011-07-11 02:49:52 +0000441 if (atomic_read(&p->refcnt) == 0) {
442 smp_rmb();
443 delta = (__u32)jiffies - p->dtime;
444 if (delta >= ttl &&
445 atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
446 p->gc_next = gchead;
447 gchead = p;
448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 }
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000451 while ((p = gchead) != NULL) {
452 gchead = p->gc_next;
453 cnt++;
454 unlink_from_pool(p, base, stack);
455 }
456 return cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457}
458
David S. Millerc0efc882012-06-09 19:12:36 -0700459struct inet_peer *inet_getpeer(struct inet_peer_base *base,
Gao fengc8a627e2012-06-08 01:20:41 +0000460 const struct inetpeer_addr *daddr,
461 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
Eric Dumazetb914c4e2010-10-25 23:55:38 +0000463 struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
David S. Miller98158f52010-11-30 11:41:59 -0800464 struct inet_peer *p;
Eric Dumazet65e83542011-03-04 14:33:59 -0800465 unsigned int sequence;
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000466 int invalidated, gccnt = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
David S. Millerb48c80ece2012-06-10 00:24:21 -0700468 flush_check(base, daddr->family);
469
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000470 /* Attempt a lockless lookup first.
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000471 * Because of a concurrent writer, we might not find an existing entry.
472 */
David S. Miller7b46ac42011-03-08 14:59:28 -0800473 rcu_read_lock();
Eric Dumazet65e83542011-03-04 14:33:59 -0800474 sequence = read_seqbegin(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000475 p = lookup_rcu(daddr, base);
Eric Dumazet65e83542011-03-04 14:33:59 -0800476 invalidated = read_seqretry(&base->lock, sequence);
David S. Miller7b46ac42011-03-08 14:59:28 -0800477 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000479 if (p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Eric Dumazet65e83542011-03-04 14:33:59 -0800482 /* If no writer did a change during our lookup, we can return early. */
483 if (!create && !invalidated)
484 return NULL;
485
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000486 /* retry an exact lookup, taking the lock before.
487 * At least, nodes should be hot in our cache.
488 */
Eric Dumazet65e83542011-03-04 14:33:59 -0800489 write_seqlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000490relookup:
David S. Miller02663042010-11-30 12:08:53 -0800491 p = lookup(daddr, stack, base);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000492 if (p != peer_avl_empty) {
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000493 atomic_inc(&p->refcnt);
Eric Dumazet65e83542011-03-04 14:33:59 -0800494 write_sequnlock_bh(&base->lock);
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000495 return p;
496 }
497 if (!gccnt) {
498 gccnt = inet_peer_gc(base, stack, stackptr);
499 if (gccnt && create)
500 goto relookup;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000501 }
502 p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
503 if (p) {
David S. Millerb534ecf2010-11-30 11:54:19 -0800504 p->daddr = *daddr;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000505 atomic_set(&p->refcnt, 1);
506 atomic_set(&p->rid, 0);
Eric Dumazet87c48fa2011-07-21 21:25:58 -0700507 atomic_set(&p->ip_id_count,
508 (daddr->family == AF_INET) ?
509 secure_ip_id(daddr->addr.a4) :
510 secure_ipv6_id(daddr->addr.a6));
David S. Miller144001b2011-01-27 13:52:16 -0800511 p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
David S. Miller92d86822011-02-04 15:55:25 -0800512 p->rate_tokens = 0;
Nicolas Dichtelbc9259a2012-09-27 04:11:00 +0000513 /* 60*HZ is arbitrary, but chosen enough high so that the first
514 * calculation of tokens is at its maximum.
515 */
516 p->rate_last = jiffies - 60*HZ;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000517 INIT_LIST_HEAD(&p->gc_list);
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000518
519 /* Link the node. */
David S. Miller98158f52010-11-30 11:41:59 -0800520 link_to_pool(p, base);
521 base->total++;
Eric Dumazetaa1039e2010-06-15 08:23:14 +0000522 }
Eric Dumazet65e83542011-03-04 14:33:59 -0800523 write_sequnlock_bh(&base->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return p;
526}
David S. Millerb3419362010-11-30 12:27:11 -0800527EXPORT_SYMBOL_GPL(inet_getpeer);
David S. Miller98158f52010-11-30 11:41:59 -0800528
Eric Dumazet4663afe2006-10-12 21:21:06 -0700529void inet_putpeer(struct inet_peer *p)
530{
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000531 p->dtime = (__u32)jiffies;
Eric Dumazet6d1a3e02011-07-11 02:49:52 +0000532 smp_mb__before_atomic_dec();
Eric Dumazet4b9d9be2011-06-08 13:35:34 +0000533 atomic_dec(&p->refcnt);
Eric Dumazet4663afe2006-10-12 21:21:06 -0700534}
David S. Millerb3419362010-11-30 12:27:11 -0800535EXPORT_SYMBOL_GPL(inet_putpeer);
David S. Miller92d86822011-02-04 15:55:25 -0800536
537/*
538 * Check transmit rate limitation for given message.
539 * The rate information is held in the inet_peer entries now.
540 * This function is generic and could be used for other purposes
541 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
542 *
543 * Note that the same inet_peer fields are modified by functions in
544 * route.c too, but these work for packet destinations while xrlim_allow
545 * works for icmp destinations. This means the rate limiting information
546 * for one "ip object" is shared - and these ICMPs are twice limited:
547 * by source and by destination.
548 *
549 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
550 * SHOULD allow setting of rate limits
551 *
552 * Shared between ICMPv4 and ICMPv6.
553 */
554#define XRLIM_BURST_FACTOR 6
555bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
556{
557 unsigned long now, token;
558 bool rc = false;
559
560 if (!peer)
561 return true;
562
563 token = peer->rate_tokens;
564 now = jiffies;
565 token += now - peer->rate_last;
566 peer->rate_last = now;
567 if (token > XRLIM_BURST_FACTOR * timeout)
568 token = XRLIM_BURST_FACTOR * timeout;
569 if (token >= timeout) {
570 token -= timeout;
571 rc = true;
572 }
573 peer->rate_tokens = token;
574 return rc;
575}
576EXPORT_SYMBOL(inet_peer_xrlim_allow);
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000577
Eric Dumazet55432d22012-06-05 03:00:18 +0000578static void inetpeer_inval_rcu(struct rcu_head *head)
579{
580 struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu);
581
582 spin_lock_bh(&gc_lock);
583 list_add_tail(&p->gc_list, &gc_list);
584 spin_unlock_bh(&gc_lock);
585
586 schedule_delayed_work(&gc_work, gc_delay);
587}
588
David S. Miller56a6b242012-06-09 16:32:41 -0700589void inetpeer_invalidate_tree(struct inet_peer_base *base)
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000590{
Eric Dumazetda557372012-06-20 04:02:10 +0000591 struct inet_peer *root;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000592
593 write_seqlock_bh(&base->lock);
594
Eric Dumazetda557372012-06-20 04:02:10 +0000595 root = rcu_deref_locked(base->root, base);
596 if (root != peer_avl_empty) {
597 base->root = peer_avl_empty_rcu;
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000598 base->total = 0;
Eric Dumazetda557372012-06-20 04:02:10 +0000599 call_rcu(&root->gc_rcu, inetpeer_inval_rcu);
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000600 }
601
Steffen Klassert5faa5df2012-03-06 21:20:26 +0000602 write_sequnlock_bh(&base->lock);
603}
604EXPORT_SYMBOL(inetpeer_invalidate_tree);