inetpeer.c 14.2 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
/*
 *		INETPEER - A storage for permanent information about peers
 *
 *  This source is covered by the GNU GPL, the same as all kernel sources.
 *
 *  Authors:	Andrey V. Savochkin <saw@msu.ru>
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/interrupt.h>
#include <linux/spinlock.h>
#include <linux/random.h>
#include <linux/timer.h>
#include <linux/time.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/net.h>
20
#include <net/ip.h>
L
Linus Torvalds 已提交
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61
#include <net/inetpeer.h>

/*
 *  Theory of operations.
 *  We keep one entry for each peer IP address.  The nodes contains long-living
 *  information about the peer which doesn't depend on routes.
 *  At this moment this information consists only of ID field for the next
 *  outgoing IP packet.  This field is incremented with each packet as encoded
 *  in inet_getid() function (include/net/inetpeer.h).
 *  At the moment of writing this notes identifier of IP packets is generated
 *  to be unpredictable using this code only for packets subjected
 *  (actually or potentially) to defragmentation.  I.e. DF packets less than
 *  PMTU in size uses a constant ID and do not use this code (see
 *  ip_select_ident() in include/net/ip.h).
 *
 *  Route cache entries hold references to our nodes.
 *  New cache entries get references via lookup by destination IP address in
 *  the avl tree.  The reference is grabbed only when it's needed i.e. only
 *  when we try to output IP packet which needs an unpredictable ID (see
 *  __ip_select_ident() in net/ipv4/route.c).
 *  Nodes are removed only when reference counter goes to 0.
 *  When it's happened the node may be removed when a sufficient amount of
 *  time has been passed since its last use.  The less-recently-used entry can
 *  also be removed if the pool is overloaded i.e. if the total amount of
 *  entries is greater-or-equal than the threshold.
 *
 *  Node pool is organised as an AVL tree.
 *  Such an implementation has been chosen not just for fun.  It's a way to
 *  prevent easy and efficient DoS attacks by creating hash collisions.  A huge
 *  amount of long living nodes in a single hash slot would significantly delay
 *  lookups performed with disabled BHs.
 *
 *  Serialisation issues.
 *  1.  Nodes may appear in the tree only with the pool write lock held.
 *  2.  Nodes may disappear from the tree only with the pool write lock held
 *      AND reference count being 0.
 *  3.  Nodes appears and disappears from unused node list only under
 *      "inet_peer_unused_lock".
 *  4.  Global variable peer_total is modified under the pool lock.
 *  5.  struct inet_peer fields modification:
 *		avl_left, avl_right, avl_parent, avl_height: pool lock
62
 *		unused: unused node list lock
L
Linus Torvalds 已提交
63 64 65 66 67 68 69
 *		refcnt: atomically against modifications on other CPU;
 *		   usually under some other lock to prevent node disappearing
 *		dtime: unused node list lock
 *		v4daddr: unchangeable
 *		ip_id_count: idlock
 */

70
static struct kmem_cache *peer_cachep __read_mostly;
L
Linus Torvalds 已提交
71 72 73 74 75 76 77 78 79 80 81 82

#define node_height(x) x->avl_height
static struct inet_peer peer_fake_node = {
	.avl_left	= &peer_fake_node,
	.avl_right	= &peer_fake_node,
	.avl_height	= 0
};
#define peer_avl_empty (&peer_fake_node)
static struct inet_peer *peer_root = peer_avl_empty;
static DEFINE_RWLOCK(peer_pool_lock);
#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */

83
static int peer_total;
L
Linus Torvalds 已提交
84
/* Exported for sysctl_net_ipv4.  */
E
Eric Dumazet 已提交
85
int inet_peer_threshold __read_mostly = 65536 + 128;	/* start to throw entries more
L
Linus Torvalds 已提交
86
					 * aggressively at this stage */
E
Eric Dumazet 已提交
87 88 89 90
int inet_peer_minttl __read_mostly = 120 * HZ;	/* TTL under high load: 120 sec */
int inet_peer_maxttl __read_mostly = 10 * 60 * HZ;	/* usual time to live: 10 min */
int inet_peer_gc_mintime __read_mostly = 10 * HZ;
int inet_peer_gc_maxtime __read_mostly = 120 * HZ;
L
Linus Torvalds 已提交
91

92
static LIST_HEAD(unused_peers);
93
static DEFINE_SPINLOCK(inet_peer_unused_lock);
L
Linus Torvalds 已提交
94 95

static void peer_check_expire(unsigned long dummy);
96
static DEFINE_TIMER(peer_periodic_timer, peer_check_expire, 0, 0);
L
Linus Torvalds 已提交
97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118


/* Called from ip_output.c:ip_init  */
void __init inet_initpeers(void)
{
	struct sysinfo si;

	/* Use the straight interface to information about memory. */
	si_meminfo(&si);
	/* The values below were suggested by Alexey Kuznetsov
	 * <kuznet@ms2.inr.ac.ru>.  I don't have any opinion about the values
	 * myself.  --SAW
	 */
	if (si.totalram <= (32768*1024)/PAGE_SIZE)
		inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
	if (si.totalram <= (16384*1024)/PAGE_SIZE)
		inet_peer_threshold >>= 1; /* about 512KB */
	if (si.totalram <= (8192*1024)/PAGE_SIZE)
		inet_peer_threshold >>= 2; /* about 128KB */

	peer_cachep = kmem_cache_create("inet_peer_cache",
			sizeof(struct inet_peer),
A
Alexey Dobriyan 已提交
119
			0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
120
			NULL);
L
Linus Torvalds 已提交
121 122 123 124 125 126 127 128 129 130 131 132 133 134

	/* All the timers, started at system startup tend
	   to synchronize. Perturb it a bit.
	 */
	peer_periodic_timer.expires = jiffies
		+ net_random() % inet_peer_gc_maxtime
		+ inet_peer_gc_maxtime;
	add_timer(&peer_periodic_timer);
}

/* Called with or without local BH being disabled. */
static void unlink_from_unused(struct inet_peer *p)
{
	spin_lock_bh(&inet_peer_unused_lock);
135
	list_del_init(&p->unused);
L
Linus Torvalds 已提交
136 137 138
	spin_unlock_bh(&inet_peer_unused_lock);
}

E
Eric Dumazet 已提交
139 140 141 142 143
/*
 * Called with local BH disabled and the pool lock held.
 * _stack is known to be NULL or not at compile time,
 * so compiler will optimize the if (_stack) tests.
 */
144
#define lookup(_daddr, _stack) 					\
L
Linus Torvalds 已提交
145 146
({								\
	struct inet_peer *u, **v;				\
147
	if (_stack != NULL) {					\
E
Eric Dumazet 已提交
148 149 150
		stackptr = _stack;				\
		*stackptr++ = &peer_root;			\
	}							\
L
Linus Torvalds 已提交
151
	for (u = peer_root; u != peer_avl_empty; ) {		\
E
Eric Dumazet 已提交
152
		if (_daddr == u->v4daddr)			\
L
Linus Torvalds 已提交
153
			break;					\
E
Eric Dumazet 已提交
154
		if ((__force __u32)_daddr < (__force __u32)u->v4daddr)	\
L
Linus Torvalds 已提交
155 156 157
			v = &u->avl_left;			\
		else						\
			v = &u->avl_right;			\
158
		if (_stack != NULL)				\
E
Eric Dumazet 已提交
159
			*stackptr++ = v;			\
L
Linus Torvalds 已提交
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
		u = *v;						\
	}							\
	u;							\
})

/* Called with local BH disabled and the pool write lock held. */
#define lookup_rightempty(start)				\
({								\
	struct inet_peer *u, **v;				\
	*stackptr++ = &start->avl_left;				\
	v = &start->avl_left;					\
	for (u = *v; u->avl_right != peer_avl_empty; ) {	\
		v = &u->avl_right;				\
		*stackptr++ = v;				\
		u = *v;						\
	}							\
	u;							\
})

/* Called with local BH disabled and the pool write lock held.
 * Variable names are the proof of operation correctness.
 * Look into mm/map_avl.c for more detail description of the ideas.  */
static void peer_avl_rebalance(struct inet_peer **stack[],
		struct inet_peer ***stackend)
{
	struct inet_peer **nodep, *node, *l, *r;
	int lh, rh;

	while (stackend > stack) {
		nodep = *--stackend;
		node = *nodep;
		l = node->avl_left;
		r = node->avl_right;
		lh = node_height(l);
		rh = node_height(r);
		if (lh > rh + 1) { /* l: RH+2 */
			struct inet_peer *ll, *lr, *lrl, *lrr;
			int lrh;
			ll = l->avl_left;
			lr = l->avl_right;
			lrh = node_height(lr);
			if (lrh <= node_height(ll)) {	/* ll: RH+1 */
				node->avl_left = lr;	/* lr: RH or RH+1 */
				node->avl_right = r;	/* r: RH */
				node->avl_height = lrh + 1; /* RH+1 or RH+2 */
				l->avl_left = ll;	/* ll: RH+1 */
				l->avl_right = node;	/* node: RH+1 or RH+2 */
				l->avl_height = node->avl_height + 1;
				*nodep = l;
			} else { /* ll: RH, lr: RH+1 */
				lrl = lr->avl_left;	/* lrl: RH or RH-1 */
				lrr = lr->avl_right;	/* lrr: RH or RH-1 */
				node->avl_left = lrr;	/* lrr: RH or RH-1 */
				node->avl_right = r;	/* r: RH */
				node->avl_height = rh + 1; /* node: RH+1 */
				l->avl_left = ll;	/* ll: RH */
				l->avl_right = lrl;	/* lrl: RH or RH-1 */
				l->avl_height = rh + 1;	/* l: RH+1 */
				lr->avl_left = l;	/* l: RH+1 */
				lr->avl_right = node;	/* node: RH+1 */
				lr->avl_height = rh + 2;
				*nodep = lr;
			}
		} else if (rh > lh + 1) { /* r: LH+2 */
			struct inet_peer *rr, *rl, *rlr, *rll;
			int rlh;
			rr = r->avl_right;
			rl = r->avl_left;
			rlh = node_height(rl);
			if (rlh <= node_height(rr)) {	/* rr: LH+1 */
				node->avl_right = rl;	/* rl: LH or LH+1 */
				node->avl_left = l;	/* l: LH */
				node->avl_height = rlh + 1; /* LH+1 or LH+2 */
				r->avl_right = rr;	/* rr: LH+1 */
				r->avl_left = node;	/* node: LH+1 or LH+2 */
				r->avl_height = node->avl_height + 1;
				*nodep = r;
			} else { /* rr: RH, rl: RH+1 */
				rlr = rl->avl_right;	/* rlr: LH or LH-1 */
				rll = rl->avl_left;	/* rll: LH or LH-1 */
				node->avl_right = rll;	/* rll: LH or LH-1 */
				node->avl_left = l;	/* l: LH */
				node->avl_height = lh + 1; /* node: LH+1 */
				r->avl_right = rr;	/* rr: LH */
				r->avl_left = rlr;	/* rlr: LH or LH-1 */
				r->avl_height = lh + 1;	/* r: LH+1 */
				rl->avl_right = r;	/* r: LH+1 */
				rl->avl_left = node;	/* node: LH+1 */
				rl->avl_height = lh + 2;
				*nodep = rl;
			}
		} else {
			node->avl_height = (lh > rh ? lh : rh) + 1;
		}
	}
}

/* Called with local BH disabled and the pool write lock held. */
#define link_to_pool(n)						\
do {								\
	n->avl_height = 1;					\
	n->avl_left = peer_avl_empty;				\
	n->avl_right = peer_avl_empty;				\
	**--stackptr = n;					\
	peer_avl_rebalance(stack, stackptr);			\
} while(0)

/* May be called with local BH enabled. */
static void unlink_from_pool(struct inet_peer *p)
{
	int do_free;

	do_free = 0;

	write_lock_bh(&peer_pool_lock);
	/* Check the reference counter.  It was artificially incremented by 1
	 * in cleanup() function to prevent sudden disappearing.  If the
	 * reference count is still 1 then the node is referenced only as `p'
	 * here and from the pool.  So under the exclusive pool lock it's safe
	 * to remove the node and free it later. */
	if (atomic_read(&p->refcnt) == 1) {
		struct inet_peer **stack[PEER_MAXDEPTH];
		struct inet_peer ***stackptr, ***delp;
E
Eric Dumazet 已提交
283
		if (lookup(p->v4daddr, stack) != p)
L
Linus Torvalds 已提交
284 285 286 287 288 289 290 291 292
			BUG();
		delp = stackptr - 1; /* *delp[0] == p */
		if (p->avl_left == peer_avl_empty) {
			*delp[0] = p->avl_right;
			--stackptr;
		} else {
			/* look for a node to insert instead of p */
			struct inet_peer *t;
			t = lookup_rightempty(p);
293
			BUG_ON(*stackptr[-1] != t);
L
Linus Torvalds 已提交
294 295 296 297 298 299 300 301
			**--stackptr = t->avl_left;
			/* t is removed, t->v4daddr > x->v4daddr for any
			 * x in p->avl_left subtree.
			 * Put t in the old place of p. */
			*delp[0] = t;
			t->avl_left = p->avl_left;
			t->avl_right = p->avl_right;
			t->avl_height = p->avl_height;
302
			BUG_ON(delp[1] != &p->avl_left);
L
Linus Torvalds 已提交
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
			delp[1] = &t->avl_left; /* was &p->avl_left */
		}
		peer_avl_rebalance(stack, stackptr);
		peer_total--;
		do_free = 1;
	}
	write_unlock_bh(&peer_pool_lock);

	if (do_free)
		kmem_cache_free(peer_cachep, p);
	else
		/* The node is used again.  Decrease the reference counter
		 * back.  The loop "cleanup -> unlink_from_unused
		 *   -> unlink_from_pool -> putpeer -> link_to_unused
		 *   -> cleanup (for the same node)"
		 * doesn't really exist because the entry will have a
		 * recent deletion time and will not be cleaned again soon. */
		inet_putpeer(p);
}

/* May be called with local BH enabled. */
static int cleanup_once(unsigned long ttl)
{
326
	struct inet_peer *p = NULL;
L
Linus Torvalds 已提交
327 328 329

	/* Remove the first entry from the list of unused nodes. */
	spin_lock_bh(&inet_peer_unused_lock);
330 331 332 333 334 335
	if (!list_empty(&unused_peers)) {
		__u32 delta;

		p = list_first_entry(&unused_peers, struct inet_peer, unused);
		delta = (__u32)jiffies - p->dtime;

336
		if (delta < ttl) {
L
Linus Torvalds 已提交
337 338 339 340
			/* Do not prune fresh entries. */
			spin_unlock_bh(&inet_peer_unused_lock);
			return -1;
		}
341 342 343

		list_del_init(&p->unused);

L
Linus Torvalds 已提交
344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
		/* Grab an extra reference to prevent node disappearing
		 * before unlink_from_pool() call. */
		atomic_inc(&p->refcnt);
	}
	spin_unlock_bh(&inet_peer_unused_lock);

	if (p == NULL)
		/* It means that the total number of USED entries has
		 * grown over inet_peer_threshold.  It shouldn't really
		 * happen because of entry limits in route cache. */
		return -1;

	unlink_from_pool(p);
	return 0;
}

/* Called with or without local BH being disabled. */
A
Al Viro 已提交
361
struct inet_peer *inet_getpeer(__be32 daddr, int create)
L
Linus Torvalds 已提交
362 363 364 365 366 367
{
	struct inet_peer *p, *n;
	struct inet_peer **stack[PEER_MAXDEPTH], ***stackptr;

	/* Look up for the address quickly. */
	read_lock_bh(&peer_pool_lock);
E
Eric Dumazet 已提交
368
	p = lookup(daddr, NULL);
L
Linus Torvalds 已提交
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
	if (p != peer_avl_empty)
		atomic_inc(&p->refcnt);
	read_unlock_bh(&peer_pool_lock);

	if (p != peer_avl_empty) {
		/* The existing node has been found. */
		/* Remove the entry from unused list if it was there. */
		unlink_from_unused(p);
		return p;
	}

	if (!create)
		return NULL;

	/* Allocate the space outside the locked region. */
	n = kmem_cache_alloc(peer_cachep, GFP_ATOMIC);
	if (n == NULL)
		return NULL;
	n->v4daddr = daddr;
	atomic_set(&n->refcnt, 1);
H
Herbert Xu 已提交
389
	atomic_set(&n->rid, 0);
E
Eric Dumazet 已提交
390
	atomic_set(&n->ip_id_count, secure_ip_id(daddr));
L
Linus Torvalds 已提交
391 392 393 394
	n->tcp_ts_stamp = 0;

	write_lock_bh(&peer_pool_lock);
	/* Check if an entry has suddenly appeared. */
E
Eric Dumazet 已提交
395
	p = lookup(daddr, stack);
L
Linus Torvalds 已提交
396 397 398 399 400
	if (p != peer_avl_empty)
		goto out_free;

	/* Link the node. */
	link_to_pool(n);
401
	INIT_LIST_HEAD(&n->unused);
L
Linus Torvalds 已提交
402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424
	peer_total++;
	write_unlock_bh(&peer_pool_lock);

	if (peer_total >= inet_peer_threshold)
		/* Remove one less-recently-used entry. */
		cleanup_once(0);

	return n;

out_free:
	/* The appropriate node is already in the pool. */
	atomic_inc(&p->refcnt);
	write_unlock_bh(&peer_pool_lock);
	/* Remove the entry from unused list if it was there. */
	unlink_from_unused(p);
	/* Free preallocated the preallocated node. */
	kmem_cache_free(peer_cachep, n);
	return p;
}

/* Called with local BH disabled. */
static void peer_check_expire(unsigned long dummy)
{
425
	unsigned long now = jiffies;
L
Linus Torvalds 已提交
426 427 428 429 430 431 432 433
	int ttl;

	if (peer_total >= inet_peer_threshold)
		ttl = inet_peer_minttl;
	else
		ttl = inet_peer_maxttl
				- (inet_peer_maxttl - inet_peer_minttl) / HZ *
					peer_total / inet_peer_threshold * HZ;
434 435 436 437
	while (!cleanup_once(ttl)) {
		if (jiffies != now)
			break;
	}
L
Linus Torvalds 已提交
438 439 440 441

	/* Trigger the timer after inet_peer_gc_mintime .. inet_peer_gc_maxtime
	 * interval depending on the total number of entries (more entries,
	 * less interval). */
442 443 444 445 446 447 448
	if (peer_total >= inet_peer_threshold)
		peer_periodic_timer.expires = jiffies + inet_peer_gc_mintime;
	else
		peer_periodic_timer.expires = jiffies
			+ inet_peer_gc_maxtime
			- (inet_peer_gc_maxtime - inet_peer_gc_mintime) / HZ *
				peer_total / inet_peer_threshold * HZ;
L
Linus Torvalds 已提交
449 450
	add_timer(&peer_periodic_timer);
}
451 452 453 454 455

void inet_putpeer(struct inet_peer *p)
{
	spin_lock_bh(&inet_peer_unused_lock);
	if (atomic_dec_and_test(&p->refcnt)) {
456
		list_add_tail(&p->unused, &unused_peers);
457 458 459 460
		p->dtime = (__u32)jiffies;
	}
	spin_unlock_bh(&inet_peer_unused_lock);
}