1 /*
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  *
6  *  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>
15 #include <linux/timer.h>
16 #include <linux/time.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/net.h>
20 #include <net/ip.h>
21 #include <net/inetpeer.h>
22 
23 /*
24  *  Theory of operations.
25  *  We keep one entry for each peer IP address.  The nodes contains long-living
26  *  information about the peer which doesn't depend on routes.
27  *  At this moment this information consists only of ID field for the next
28  *  outgoing IP packet.  This field is incremented with each packet as encoded
29  *  in inet_getid() function (include/net/inetpeer.h).
30  *  At the moment of writing this notes identifier of IP packets is generated
31  *  to be unpredictable using this code only for packets subjected
32  *  (actually or potentially) to defragmentation.  I.e. DF packets less than
33  *  PMTU in size uses a constant ID and do not use this code (see
34  *  ip_select_ident() in include/net/ip.h).
35  *
36  *  Route cache entries hold references to our nodes.
37  *  New cache entries get references via lookup by destination IP address in
38  *  the avl tree.  The reference is grabbed only when it's needed i.e. only
39  *  when we try to output IP packet which needs an unpredictable ID (see
40  *  __ip_select_ident() in net/ipv4/route.c).
41  *  Nodes are removed only when reference counter goes to 0.
42  *  When it's happened the node may be removed when a sufficient amount of
43  *  time has been passed since its last use.  The less-recently-used entry can
44  *  also be removed if the pool is overloaded i.e. if the total amount of
45  *  entries is greater-or-equal than the threshold.
46  *
47  *  Node pool is organised as an AVL tree.
48  *  Such an implementation has been chosen not just for fun.  It's a way to
49  *  prevent easy and efficient DoS attacks by creating hash collisions.  A huge
50  *  amount of long living nodes in a single hash slot would significantly delay
51  *  lookups performed with disabled BHs.
52  *
53  *  Serialisation issues.
54  *  1.  Nodes may appear in the tree only with the pool lock held.
55  *  2.  Nodes may disappear from the tree only with the pool lock held
56  *      AND reference count being 0.
57  *  3.  Nodes appears and disappears from unused node list only under
58  *      "inet_peer_unused_lock".
59  *  4.  Global variable peer_total is modified under the pool lock.
60  *  5.  struct inet_peer fields modification:
61  *		avl_left, avl_right, avl_parent, avl_height: pool lock
62  *		unused: unused node list lock
63  *		refcnt: atomically against modifications on other CPU;
64  *		   usually under some other lock to prevent node disappearing
65  *		dtime: unused node list lock
66  *		daddr: unchangeable
67  *		ip_id_count: atomic value (no lock needed)
68  */
69 
70 static struct kmem_cache *peer_cachep __read_mostly;
71 
72 #define node_height(x) x->avl_height
73 
74 #define peer_avl_empty ((struct inet_peer *)&peer_fake_node)
75 #define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node)
76 static const struct inet_peer peer_fake_node = {
77 	.avl_left	= peer_avl_empty_rcu,
78 	.avl_right	= peer_avl_empty_rcu,
79 	.avl_height	= 0
80 };
81 
82 struct inet_peer_base {
83 	struct inet_peer __rcu *root;
84 	seqlock_t	lock;
85 	int		total;
86 };
87 
88 static struct inet_peer_base v4_peers = {
89 	.root		= peer_avl_empty_rcu,
90 	.lock		= __SEQLOCK_UNLOCKED(v4_peers.lock),
91 	.total		= 0,
92 };
93 
94 static struct inet_peer_base v6_peers = {
95 	.root		= peer_avl_empty_rcu,
96 	.lock		= __SEQLOCK_UNLOCKED(v6_peers.lock),
97 	.total		= 0,
98 };
99 
100 #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
101 
102 /* Exported for sysctl_net_ipv4.  */
103 int inet_peer_threshold __read_mostly = 65536 + 128;	/* start to throw entries more
104 					 * aggressively at this stage */
105 int inet_peer_minttl __read_mostly = 120 * HZ;	/* TTL under high load: 120 sec */
106 int inet_peer_maxttl __read_mostly = 10 * 60 * HZ;	/* usual time to live: 10 min */
107 int inet_peer_gc_mintime __read_mostly = 10 * HZ;
108 int inet_peer_gc_maxtime __read_mostly = 120 * HZ;
109 
110 static struct {
111 	struct list_head	list;
112 	spinlock_t		lock;
113 } unused_peers = {
114 	.list			= LIST_HEAD_INIT(unused_peers.list),
115 	.lock			= __SPIN_LOCK_UNLOCKED(unused_peers.lock),
116 };
117 
118 static void peer_check_expire(unsigned long dummy);
119 static DEFINE_TIMER(peer_periodic_timer, peer_check_expire, 0, 0);
120 
121 
122 /* Called from ip_output.c:ip_init  */
inet_initpeers(void)123 void __init inet_initpeers(void)
124 {
125 	struct sysinfo si;
126 
127 	/* Use the straight interface to information about memory. */
128 	si_meminfo(&si);
129 	/* The values below were suggested by Alexey Kuznetsov
130 	 * <kuznet@ms2.inr.ac.ru>.  I don't have any opinion about the values
131 	 * myself.  --SAW
132 	 */
133 	if (si.totalram <= (32768*1024)/PAGE_SIZE)
134 		inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
135 	if (si.totalram <= (16384*1024)/PAGE_SIZE)
136 		inet_peer_threshold >>= 1; /* about 512KB */
137 	if (si.totalram <= (8192*1024)/PAGE_SIZE)
138 		inet_peer_threshold >>= 2; /* about 128KB */
139 
140 	peer_cachep = kmem_cache_create("inet_peer_cache",
141 			sizeof(struct inet_peer),
142 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
143 			NULL);
144 
145 	/* All the timers, started at system startup tend
146 	   to synchronize. Perturb it a bit.
147 	 */
148 	peer_periodic_timer.expires = jiffies
149 		+ net_random() % inet_peer_gc_maxtime
150 		+ inet_peer_gc_maxtime;
151 	add_timer(&peer_periodic_timer);
152 }
153 
154 /* Called with or without local BH being disabled. */
unlink_from_unused(struct inet_peer * p)155 static void unlink_from_unused(struct inet_peer *p)
156 {
157 	if (!list_empty(&p->unused)) {
158 		spin_lock_bh(&unused_peers.lock);
159 		list_del_init(&p->unused);
160 		spin_unlock_bh(&unused_peers.lock);
161 	}
162 }
163 
addr_compare(const struct inetpeer_addr * a,const struct inetpeer_addr * b)164 static int addr_compare(const struct inetpeer_addr *a,
165 			const struct inetpeer_addr *b)
166 {
167 	int i, n = (a->family == AF_INET ? 1 : 4);
168 
169 	for (i = 0; i < n; i++) {
170 		if (a->addr.a6[i] == b->addr.a6[i])
171 			continue;
172 		if (a->addr.a6[i] < b->addr.a6[i])
173 			return -1;
174 		return 1;
175 	}
176 
177 	return 0;
178 }
179 
180 #define rcu_deref_locked(X, BASE)				\
181 	rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))
182 
183 /*
184  * Called with local BH disabled and the pool lock held.
185  */
186 #define lookup(_daddr, _stack, _base)				\
187 ({								\
188 	struct inet_peer *u;					\
189 	struct inet_peer __rcu **v;				\
190 								\
191 	stackptr = _stack;					\
192 	*stackptr++ = &_base->root;				\
193 	for (u = rcu_deref_locked(_base->root, _base);		\
194 	     u != peer_avl_empty; ) {				\
195 		int cmp = addr_compare(_daddr, &u->daddr);	\
196 		if (cmp == 0)					\
197 			break;					\
198 		if (cmp == -1)					\
199 			v = &u->avl_left;			\
200 		else						\
201 			v = &u->avl_right;			\
202 		*stackptr++ = v;				\
203 		u = rcu_deref_locked(*v, _base);		\
204 	}							\
205 	u;							\
206 })
207 
208 /*
209  * Called with rcu_read_lock()
210  * Because we hold no lock against a writer, its quite possible we fall
211  * in an endless loop.
212  * But every pointer we follow is guaranteed to be valid thanks to RCU.
213  * We exit from this function if number of links exceeds PEER_MAXDEPTH
214  */
lookup_rcu(const struct inetpeer_addr * daddr,struct inet_peer_base * base)215 static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
216 				    struct inet_peer_base *base)
217 {
218 	struct inet_peer *u = rcu_dereference(base->root);
219 	int count = 0;
220 
221 	while (u != peer_avl_empty) {
222 		int cmp = addr_compare(daddr, &u->daddr);
223 		if (cmp == 0) {
224 			/* Before taking a reference, check if this entry was
225 			 * deleted, unlink_from_pool() sets refcnt=-1 to make
226 			 * distinction between an unused entry (refcnt=0) and
227 			 * a freed one.
228 			 */
229 			if (unlikely(!atomic_add_unless(&u->refcnt, 1, -1)))
230 				u = NULL;
231 			return u;
232 		}
233 		if (cmp == -1)
234 			u = rcu_dereference(u->avl_left);
235 		else
236 			u = rcu_dereference(u->avl_right);
237 		if (unlikely(++count == PEER_MAXDEPTH))
238 			break;
239 	}
240 	return NULL;
241 }
242 
243 /* Called with local BH disabled and the pool lock held. */
244 #define lookup_rightempty(start, base)				\
245 ({								\
246 	struct inet_peer *u;					\
247 	struct inet_peer __rcu **v;				\
248 	*stackptr++ = &start->avl_left;				\
249 	v = &start->avl_left;					\
250 	for (u = rcu_deref_locked(*v, base);			\
251 	     u->avl_right != peer_avl_empty_rcu; ) {		\
252 		v = &u->avl_right;				\
253 		*stackptr++ = v;				\
254 		u = rcu_deref_locked(*v, base);			\
255 	}							\
256 	u;							\
257 })
258 
259 /* Called with local BH disabled and the pool lock held.
260  * Variable names are the proof of operation correctness.
261  * Look into mm/map_avl.c for more detail description of the ideas.
262  */
peer_avl_rebalance(struct inet_peer __rcu ** stack[],struct inet_peer __rcu *** stackend,struct inet_peer_base * base)263 static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
264 			       struct inet_peer __rcu ***stackend,
265 			       struct inet_peer_base *base)
266 {
267 	struct inet_peer __rcu **nodep;
268 	struct inet_peer *node, *l, *r;
269 	int lh, rh;
270 
271 	while (stackend > stack) {
272 		nodep = *--stackend;
273 		node = rcu_deref_locked(*nodep, base);
274 		l = rcu_deref_locked(node->avl_left, base);
275 		r = rcu_deref_locked(node->avl_right, base);
276 		lh = node_height(l);
277 		rh = node_height(r);
278 		if (lh > rh + 1) { /* l: RH+2 */
279 			struct inet_peer *ll, *lr, *lrl, *lrr;
280 			int lrh;
281 			ll = rcu_deref_locked(l->avl_left, base);
282 			lr = rcu_deref_locked(l->avl_right, base);
283 			lrh = node_height(lr);
284 			if (lrh <= node_height(ll)) {	/* ll: RH+1 */
285 				RCU_INIT_POINTER(node->avl_left, lr);	/* lr: RH or RH+1 */
286 				RCU_INIT_POINTER(node->avl_right, r);	/* r: RH */
287 				node->avl_height = lrh + 1; /* RH+1 or RH+2 */
288 				RCU_INIT_POINTER(l->avl_left, ll);       /* ll: RH+1 */
289 				RCU_INIT_POINTER(l->avl_right, node);	/* node: RH+1 or RH+2 */
290 				l->avl_height = node->avl_height + 1;
291 				RCU_INIT_POINTER(*nodep, l);
292 			} else { /* ll: RH, lr: RH+1 */
293 				lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
294 				lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
295 				RCU_INIT_POINTER(node->avl_left, lrr);	/* lrr: RH or RH-1 */
296 				RCU_INIT_POINTER(node->avl_right, r);	/* r: RH */
297 				node->avl_height = rh + 1; /* node: RH+1 */
298 				RCU_INIT_POINTER(l->avl_left, ll);	/* ll: RH */
299 				RCU_INIT_POINTER(l->avl_right, lrl);	/* lrl: RH or RH-1 */
300 				l->avl_height = rh + 1;	/* l: RH+1 */
301 				RCU_INIT_POINTER(lr->avl_left, l);	/* l: RH+1 */
302 				RCU_INIT_POINTER(lr->avl_right, node);	/* node: RH+1 */
303 				lr->avl_height = rh + 2;
304 				RCU_INIT_POINTER(*nodep, lr);
305 			}
306 		} else if (rh > lh + 1) { /* r: LH+2 */
307 			struct inet_peer *rr, *rl, *rlr, *rll;
308 			int rlh;
309 			rr = rcu_deref_locked(r->avl_right, base);
310 			rl = rcu_deref_locked(r->avl_left, base);
311 			rlh = node_height(rl);
312 			if (rlh <= node_height(rr)) {	/* rr: LH+1 */
313 				RCU_INIT_POINTER(node->avl_right, rl);	/* rl: LH or LH+1 */
314 				RCU_INIT_POINTER(node->avl_left, l);	/* l: LH */
315 				node->avl_height = rlh + 1; /* LH+1 or LH+2 */
316 				RCU_INIT_POINTER(r->avl_right, rr);	/* rr: LH+1 */
317 				RCU_INIT_POINTER(r->avl_left, node);	/* node: LH+1 or LH+2 */
318 				r->avl_height = node->avl_height + 1;
319 				RCU_INIT_POINTER(*nodep, r);
320 			} else { /* rr: RH, rl: RH+1 */
321 				rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
322 				rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
323 				RCU_INIT_POINTER(node->avl_right, rll);	/* rll: LH or LH-1 */
324 				RCU_INIT_POINTER(node->avl_left, l);	/* l: LH */
325 				node->avl_height = lh + 1; /* node: LH+1 */
326 				RCU_INIT_POINTER(r->avl_right, rr);	/* rr: LH */
327 				RCU_INIT_POINTER(r->avl_left, rlr);	/* rlr: LH or LH-1 */
328 				r->avl_height = lh + 1;	/* r: LH+1 */
329 				RCU_INIT_POINTER(rl->avl_right, r);	/* r: LH+1 */
330 				RCU_INIT_POINTER(rl->avl_left, node);	/* node: LH+1 */
331 				rl->avl_height = lh + 2;
332 				RCU_INIT_POINTER(*nodep, rl);
333 			}
334 		} else {
335 			node->avl_height = (lh > rh ? lh : rh) + 1;
336 		}
337 	}
338 }
339 
340 /* Called with local BH disabled and the pool lock held. */
341 #define link_to_pool(n, base)					\
342 do {								\
343 	n->avl_height = 1;					\
344 	n->avl_left = peer_avl_empty_rcu;			\
345 	n->avl_right = peer_avl_empty_rcu;			\
346 	/* lockless readers can catch us now */			\
347 	rcu_assign_pointer(**--stackptr, n);			\
348 	peer_avl_rebalance(stack, stackptr, base);		\
349 } while (0)
350 
inetpeer_free_rcu(struct rcu_head * head)351 static void inetpeer_free_rcu(struct rcu_head *head)
352 {
353 	kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
354 }
355 
356 /* May be called with local BH enabled. */
unlink_from_pool(struct inet_peer * p,struct inet_peer_base * base,struct inet_peer __rcu ** stack[PEER_MAXDEPTH])357 static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
358 			     struct inet_peer __rcu **stack[PEER_MAXDEPTH])
359 {
360 	int do_free;
361 
362 	do_free = 0;
363 
364 	write_seqlock_bh(&base->lock);
365 	/* Check the reference counter.  It was artificially incremented by 1
366 	 * in cleanup() function to prevent sudden disappearing.  If we can
367 	 * atomically (because of lockless readers) take this last reference,
368 	 * it's safe to remove the node and free it later.
369 	 * We use refcnt=-1 to alert lockless readers this entry is deleted.
370 	 */
371 	if (atomic_cmpxchg(&p->refcnt, 1, -1) == 1) {
372 		struct inet_peer __rcu ***stackptr, ***delp;
373 		if (lookup(&p->daddr, stack, base) != p)
374 			BUG();
375 		delp = stackptr - 1; /* *delp[0] == p */
376 		if (p->avl_left == peer_avl_empty_rcu) {
377 			*delp[0] = p->avl_right;
378 			--stackptr;
379 		} else {
380 			/* look for a node to insert instead of p */
381 			struct inet_peer *t;
382 			t = lookup_rightempty(p, base);
383 			BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
384 			**--stackptr = t->avl_left;
385 			/* t is removed, t->daddr > x->daddr for any
386 			 * x in p->avl_left subtree.
387 			 * Put t in the old place of p. */
388 			RCU_INIT_POINTER(*delp[0], t);
389 			t->avl_left = p->avl_left;
390 			t->avl_right = p->avl_right;
391 			t->avl_height = p->avl_height;
392 			BUG_ON(delp[1] != &p->avl_left);
393 			delp[1] = &t->avl_left; /* was &p->avl_left */
394 		}
395 		peer_avl_rebalance(stack, stackptr, base);
396 		base->total--;
397 		do_free = 1;
398 	}
399 	write_sequnlock_bh(&base->lock);
400 
401 	if (do_free)
402 		call_rcu(&p->rcu, inetpeer_free_rcu);
403 	else
404 		/* The node is used again.  Decrease the reference counter
405 		 * back.  The loop "cleanup -> unlink_from_unused
406 		 *   -> unlink_from_pool -> putpeer -> link_to_unused
407 		 *   -> cleanup (for the same node)"
408 		 * doesn't really exist because the entry will have a
409 		 * recent deletion time and will not be cleaned again soon.
410 		 */
411 		inet_putpeer(p);
412 }
413 
family_to_base(int family)414 static struct inet_peer_base *family_to_base(int family)
415 {
416 	return (family == AF_INET ? &v4_peers : &v6_peers);
417 }
418 
peer_to_base(struct inet_peer * p)419 static struct inet_peer_base *peer_to_base(struct inet_peer *p)
420 {
421 	return family_to_base(p->daddr.family);
422 }
423 
424 /* May be called with local BH enabled. */
cleanup_once(unsigned long ttl,struct inet_peer __rcu ** stack[PEER_MAXDEPTH])425 static int cleanup_once(unsigned long ttl, struct inet_peer __rcu **stack[PEER_MAXDEPTH])
426 {
427 	struct inet_peer *p = NULL;
428 
429 	/* Remove the first entry from the list of unused nodes. */
430 	spin_lock_bh(&unused_peers.lock);
431 	if (!list_empty(&unused_peers.list)) {
432 		__u32 delta;
433 
434 		p = list_first_entry(&unused_peers.list, struct inet_peer, unused);
435 		delta = (__u32)jiffies - p->dtime;
436 
437 		if (delta < ttl) {
438 			/* Do not prune fresh entries. */
439 			spin_unlock_bh(&unused_peers.lock);
440 			return -1;
441 		}
442 
443 		list_del_init(&p->unused);
444 
445 		/* Grab an extra reference to prevent node disappearing
446 		 * before unlink_from_pool() call. */
447 		atomic_inc(&p->refcnt);
448 	}
449 	spin_unlock_bh(&unused_peers.lock);
450 
451 	if (p == NULL)
452 		/* It means that the total number of USED entries has
453 		 * grown over inet_peer_threshold.  It shouldn't really
454 		 * happen because of entry limits in route cache. */
455 		return -1;
456 
457 	unlink_from_pool(p, peer_to_base(p), stack);
458 	return 0;
459 }
460 
461 /* Called with or without local BH being disabled. */
inet_getpeer(struct inetpeer_addr * daddr,int create)462 struct inet_peer *inet_getpeer(struct inetpeer_addr *daddr, int create)
463 {
464 	struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
465 	struct inet_peer_base *base = family_to_base(daddr->family);
466 	struct inet_peer *p;
467 	unsigned int sequence;
468 	int invalidated;
469 
470 	/* Look up for the address quickly, lockless.
471 	 * Because of a concurrent writer, we might not find an existing entry.
472 	 */
473 	rcu_read_lock();
474 	sequence = read_seqbegin(&base->lock);
475 	p = lookup_rcu(daddr, base);
476 	invalidated = read_seqretry(&base->lock, sequence);
477 	rcu_read_unlock();
478 
479 	if (p) {
480 		/* The existing node has been found.
481 		 * Remove the entry from unused list if it was there.
482 		 */
483 		unlink_from_unused(p);
484 		return p;
485 	}
486 
487 	/* If no writer did a change during our lookup, we can return early. */
488 	if (!create && !invalidated)
489 		return NULL;
490 
491 	/* retry an exact lookup, taking the lock before.
492 	 * At least, nodes should be hot in our cache.
493 	 */
494 	write_seqlock_bh(&base->lock);
495 	p = lookup(daddr, stack, base);
496 	if (p != peer_avl_empty) {
497 		atomic_inc(&p->refcnt);
498 		write_sequnlock_bh(&base->lock);
499 		/* Remove the entry from unused list if it was there. */
500 		unlink_from_unused(p);
501 		return p;
502 	}
503 	p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
504 	if (p) {
505 		p->daddr = *daddr;
506 		atomic_set(&p->refcnt, 1);
507 		atomic_set(&p->rid, 0);
508 		atomic_set(&p->ip_id_count, secure_ip_id(daddr->addr.a4));
509 		p->tcp_ts_stamp = 0;
510 		p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
511 		p->rate_tokens = 0;
512 		p->rate_last = 0;
513 		p->pmtu_expires = 0;
514 		p->pmtu_orig = 0;
515 		memset(&p->redirect_learned, 0, sizeof(p->redirect_learned));
516 		INIT_LIST_HEAD(&p->unused);
517 
518 
519 		/* Link the node. */
520 		link_to_pool(p, base);
521 		base->total++;
522 	}
523 	write_sequnlock_bh(&base->lock);
524 
525 	if (base->total >= inet_peer_threshold)
526 		/* Remove one less-recently-used entry. */
527 		cleanup_once(0, stack);
528 
529 	return p;
530 }
531 
compute_total(void)532 static int compute_total(void)
533 {
534 	return v4_peers.total + v6_peers.total;
535 }
536 EXPORT_SYMBOL_GPL(inet_getpeer);
537 
538 /* Called with local BH disabled. */
peer_check_expire(unsigned long dummy)539 static void peer_check_expire(unsigned long dummy)
540 {
541 	unsigned long now = jiffies;
542 	int ttl, total;
543 	struct inet_peer __rcu **stack[PEER_MAXDEPTH];
544 
545 	total = compute_total();
546 	if (total >= inet_peer_threshold)
547 		ttl = inet_peer_minttl;
548 	else
549 		ttl = inet_peer_maxttl
550 				- (inet_peer_maxttl - inet_peer_minttl) / HZ *
551 					total / inet_peer_threshold * HZ;
552 	while (!cleanup_once(ttl, stack)) {
553 		if (jiffies != now)
554 			break;
555 	}
556 
557 	/* Trigger the timer after inet_peer_gc_mintime .. inet_peer_gc_maxtime
558 	 * interval depending on the total number of entries (more entries,
559 	 * less interval). */
560 	total = compute_total();
561 	if (total >= inet_peer_threshold)
562 		peer_periodic_timer.expires = jiffies + inet_peer_gc_mintime;
563 	else
564 		peer_periodic_timer.expires = jiffies
565 			+ inet_peer_gc_maxtime
566 			- (inet_peer_gc_maxtime - inet_peer_gc_mintime) / HZ *
567 				total / inet_peer_threshold * HZ;
568 	add_timer(&peer_periodic_timer);
569 }
570 
inet_putpeer(struct inet_peer * p)571 void inet_putpeer(struct inet_peer *p)
572 {
573 	local_bh_disable();
574 
575 	if (atomic_dec_and_lock(&p->refcnt, &unused_peers.lock)) {
576 		list_add_tail(&p->unused, &unused_peers.list);
577 		p->dtime = (__u32)jiffies;
578 		spin_unlock(&unused_peers.lock);
579 	}
580 
581 	local_bh_enable();
582 }
583 EXPORT_SYMBOL_GPL(inet_putpeer);
584 
585 /*
586  *	Check transmit rate limitation for given message.
587  *	The rate information is held in the inet_peer entries now.
588  *	This function is generic and could be used for other purposes
589  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
590  *
591  *	Note that the same inet_peer fields are modified by functions in
592  *	route.c too, but these work for packet destinations while xrlim_allow
593  *	works for icmp destinations. This means the rate limiting information
594  *	for one "ip object" is shared - and these ICMPs are twice limited:
595  *	by source and by destination.
596  *
597  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
598  *			  SHOULD allow setting of rate limits
599  *
600  * 	Shared between ICMPv4 and ICMPv6.
601  */
602 #define XRLIM_BURST_FACTOR 6
inet_peer_xrlim_allow(struct inet_peer * peer,int timeout)603 bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
604 {
605 	unsigned long now, token;
606 	bool rc = false;
607 
608 	if (!peer)
609 		return true;
610 
611 	token = peer->rate_tokens;
612 	now = jiffies;
613 	token += now - peer->rate_last;
614 	peer->rate_last = now;
615 	if (token > XRLIM_BURST_FACTOR * timeout)
616 		token = XRLIM_BURST_FACTOR * timeout;
617 	if (token >= timeout) {
618 		token -= timeout;
619 		rc = true;
620 	}
621 	peer->rate_tokens = token;
622 	return rc;
623 }
624 EXPORT_SYMBOL(inet_peer_xrlim_allow);
625