Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
1da177e4 LT |
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> | |
1da177e4 LT |
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> | |
20380731 | 20 | #include <net/ip.h> |
1da177e4 LT |
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. | |
aa1039e7 ED |
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 | |
1da177e4 LT |
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 | |
d71209de | 62 | * unused: unused node list lock |
1da177e4 LT |
63 | * refcnt: atomically against modifications on other CPU; |
64 | * usually under some other lock to prevent node disappearing | |
65 | * dtime: unused node list lock | |
582a72da | 66 | * daddr: unchangeable |
317fe0e6 | 67 | * ip_id_count: atomic value (no lock needed) |
1da177e4 LT |
68 | */ |
69 | ||
e18b890b | 70 | static struct kmem_cache *peer_cachep __read_mostly; |
1da177e4 LT |
71 | |
72 | #define node_height(x) x->avl_height | |
d6cc1d64 ED |
73 | |
74 | #define peer_avl_empty ((struct inet_peer *)&peer_fake_node) | |
b914c4ea | 75 | #define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node) |
d6cc1d64 | 76 | static const struct inet_peer peer_fake_node = { |
b914c4ea ED |
77 | .avl_left = peer_avl_empty_rcu, |
78 | .avl_right = peer_avl_empty_rcu, | |
1da177e4 LT |
79 | .avl_height = 0 |
80 | }; | |
d6cc1d64 | 81 | |
98158f5a | 82 | static struct inet_peer_base { |
b914c4ea | 83 | struct inet_peer __rcu *root; |
aa1039e7 | 84 | spinlock_t lock; |
d6cc1d64 | 85 | int total; |
98158f5a | 86 | } v4_peers = { |
b914c4ea | 87 | .root = peer_avl_empty_rcu, |
98158f5a | 88 | .lock = __SPIN_LOCK_UNLOCKED(v4_peers.lock), |
d6cc1d64 ED |
89 | .total = 0, |
90 | }; | |
1da177e4 LT |
91 | #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */ |
92 | ||
1da177e4 | 93 | /* Exported for sysctl_net_ipv4. */ |
243bbcaa | 94 | int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more |
1da177e4 | 95 | * aggressively at this stage */ |
243bbcaa ED |
96 | int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */ |
97 | int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */ | |
98 | int inet_peer_gc_mintime __read_mostly = 10 * HZ; | |
99 | int inet_peer_gc_maxtime __read_mostly = 120 * HZ; | |
1da177e4 | 100 | |
d6cc1d64 ED |
101 | static struct { |
102 | struct list_head list; | |
103 | spinlock_t lock; | |
104 | } unused_peers = { | |
105 | .list = LIST_HEAD_INIT(unused_peers.list), | |
106 | .lock = __SPIN_LOCK_UNLOCKED(unused_peers.lock), | |
107 | }; | |
1da177e4 LT |
108 | |
109 | static void peer_check_expire(unsigned long dummy); | |
8d06afab | 110 | static DEFINE_TIMER(peer_periodic_timer, peer_check_expire, 0, 0); |
1da177e4 | 111 | |
1da177e4 LT |
112 | |
113 | /* Called from ip_output.c:ip_init */ | |
114 | void __init inet_initpeers(void) | |
115 | { | |
116 | struct sysinfo si; | |
117 | ||
118 | /* Use the straight interface to information about memory. */ | |
119 | si_meminfo(&si); | |
120 | /* The values below were suggested by Alexey Kuznetsov | |
121 | * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values | |
122 | * myself. --SAW | |
123 | */ | |
124 | if (si.totalram <= (32768*1024)/PAGE_SIZE) | |
125 | inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */ | |
126 | if (si.totalram <= (16384*1024)/PAGE_SIZE) | |
127 | inet_peer_threshold >>= 1; /* about 512KB */ | |
128 | if (si.totalram <= (8192*1024)/PAGE_SIZE) | |
129 | inet_peer_threshold >>= 2; /* about 128KB */ | |
130 | ||
131 | peer_cachep = kmem_cache_create("inet_peer_cache", | |
132 | sizeof(struct inet_peer), | |
317fe0e6 | 133 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, |
20c2df83 | 134 | NULL); |
1da177e4 | 135 | |
1da177e4 LT |
136 | /* All the timers, started at system startup tend |
137 | to synchronize. Perturb it a bit. | |
138 | */ | |
139 | peer_periodic_timer.expires = jiffies | |
140 | + net_random() % inet_peer_gc_maxtime | |
141 | + inet_peer_gc_maxtime; | |
142 | add_timer(&peer_periodic_timer); | |
143 | } | |
144 | ||
145 | /* Called with or without local BH being disabled. */ | |
146 | static void unlink_from_unused(struct inet_peer *p) | |
147 | { | |
d6cc1d64 ED |
148 | if (!list_empty(&p->unused)) { |
149 | spin_lock_bh(&unused_peers.lock); | |
150 | list_del_init(&p->unused); | |
151 | spin_unlock_bh(&unused_peers.lock); | |
152 | } | |
1da177e4 LT |
153 | } |
154 | ||
02663045 DM |
155 | static int addr_compare(const inet_peer_address_t *a, |
156 | const inet_peer_address_t *b) | |
157 | { | |
158 | int i, n = (a->family == AF_INET ? 1 : 4); | |
159 | ||
160 | for (i = 0; i < n; i++) { | |
161 | if (a->a6[i] == b->a6[i]) | |
162 | continue; | |
163 | if (a->a6[i] < b->a6[i]) | |
164 | return -1; | |
165 | return 1; | |
166 | } | |
167 | ||
168 | return 0; | |
169 | } | |
170 | ||
243bbcaa ED |
171 | /* |
172 | * Called with local BH disabled and the pool lock held. | |
243bbcaa | 173 | */ |
98158f5a | 174 | #define lookup(_daddr, _stack, _base) \ |
1da177e4 | 175 | ({ \ |
b914c4ea ED |
176 | struct inet_peer *u; \ |
177 | struct inet_peer __rcu **v; \ | |
aa1039e7 ED |
178 | \ |
179 | stackptr = _stack; \ | |
98158f5a DM |
180 | *stackptr++ = &_base->root; \ |
181 | for (u = rcu_dereference_protected(_base->root, \ | |
182 | lockdep_is_held(&_base->lock)); \ | |
b914c4ea | 183 | u != peer_avl_empty; ) { \ |
02663045 DM |
184 | int cmp = addr_compare(_daddr, &u->daddr); \ |
185 | if (cmp == 0) \ | |
1da177e4 | 186 | break; \ |
02663045 | 187 | if (cmp == -1) \ |
1da177e4 LT |
188 | v = &u->avl_left; \ |
189 | else \ | |
190 | v = &u->avl_right; \ | |
aa1039e7 | 191 | *stackptr++ = v; \ |
b914c4ea | 192 | u = rcu_dereference_protected(*v, \ |
98158f5a | 193 | lockdep_is_held(&_base->lock)); \ |
1da177e4 LT |
194 | } \ |
195 | u; \ | |
196 | }) | |
197 | ||
aa1039e7 ED |
198 | /* |
199 | * Called with rcu_read_lock_bh() | |
200 | * Because we hold no lock against a writer, its quite possible we fall | |
201 | * in an endless loop. | |
202 | * But every pointer we follow is guaranteed to be valid thanks to RCU. | |
203 | * We exit from this function if number of links exceeds PEER_MAXDEPTH | |
204 | */ | |
02663045 DM |
205 | static struct inet_peer *lookup_rcu_bh(const inet_peer_address_t *daddr, |
206 | struct inet_peer_base *base) | |
aa1039e7 | 207 | { |
98158f5a | 208 | struct inet_peer *u = rcu_dereference_bh(base->root); |
aa1039e7 ED |
209 | int count = 0; |
210 | ||
211 | while (u != peer_avl_empty) { | |
02663045 DM |
212 | int cmp = addr_compare(daddr, &u->daddr); |
213 | if (cmp == 0) { | |
5f2f8920 ED |
214 | /* Before taking a reference, check if this entry was |
215 | * deleted, unlink_from_pool() sets refcnt=-1 to make | |
216 | * distinction between an unused entry (refcnt=0) and | |
217 | * a freed one. | |
218 | */ | |
219 | if (unlikely(!atomic_add_unless(&u->refcnt, 1, -1))) | |
aa1039e7 ED |
220 | u = NULL; |
221 | return u; | |
222 | } | |
02663045 | 223 | if (cmp == -1) |
aa1039e7 ED |
224 | u = rcu_dereference_bh(u->avl_left); |
225 | else | |
226 | u = rcu_dereference_bh(u->avl_right); | |
227 | if (unlikely(++count == PEER_MAXDEPTH)) | |
228 | break; | |
229 | } | |
230 | return NULL; | |
231 | } | |
232 | ||
233 | /* Called with local BH disabled and the pool lock held. */ | |
98158f5a | 234 | #define lookup_rightempty(start, base) \ |
1da177e4 | 235 | ({ \ |
b914c4ea ED |
236 | struct inet_peer *u; \ |
237 | struct inet_peer __rcu **v; \ | |
1da177e4 LT |
238 | *stackptr++ = &start->avl_left; \ |
239 | v = &start->avl_left; \ | |
b914c4ea | 240 | for (u = rcu_dereference_protected(*v, \ |
98158f5a | 241 | lockdep_is_held(&base->lock)); \ |
b914c4ea | 242 | u->avl_right != peer_avl_empty_rcu; ) { \ |
1da177e4 LT |
243 | v = &u->avl_right; \ |
244 | *stackptr++ = v; \ | |
b914c4ea | 245 | u = rcu_dereference_protected(*v, \ |
98158f5a | 246 | lockdep_is_held(&base->lock)); \ |
1da177e4 LT |
247 | } \ |
248 | u; \ | |
249 | }) | |
250 | ||
aa1039e7 | 251 | /* Called with local BH disabled and the pool lock held. |
1da177e4 | 252 | * Variable names are the proof of operation correctness. |
aa1039e7 ED |
253 | * Look into mm/map_avl.c for more detail description of the ideas. |
254 | */ | |
b914c4ea | 255 | static void peer_avl_rebalance(struct inet_peer __rcu **stack[], |
98158f5a DM |
256 | struct inet_peer __rcu ***stackend, |
257 | struct inet_peer_base *base) | |
1da177e4 | 258 | { |
b914c4ea ED |
259 | struct inet_peer __rcu **nodep; |
260 | struct inet_peer *node, *l, *r; | |
1da177e4 LT |
261 | int lh, rh; |
262 | ||
263 | while (stackend > stack) { | |
264 | nodep = *--stackend; | |
b914c4ea | 265 | node = rcu_dereference_protected(*nodep, |
98158f5a | 266 | lockdep_is_held(&base->lock)); |
b914c4ea | 267 | l = rcu_dereference_protected(node->avl_left, |
98158f5a | 268 | lockdep_is_held(&base->lock)); |
b914c4ea | 269 | r = rcu_dereference_protected(node->avl_right, |
98158f5a | 270 | lockdep_is_held(&base->lock)); |
1da177e4 LT |
271 | lh = node_height(l); |
272 | rh = node_height(r); | |
273 | if (lh > rh + 1) { /* l: RH+2 */ | |
274 | struct inet_peer *ll, *lr, *lrl, *lrr; | |
275 | int lrh; | |
b914c4ea | 276 | ll = rcu_dereference_protected(l->avl_left, |
98158f5a | 277 | lockdep_is_held(&base->lock)); |
b914c4ea | 278 | lr = rcu_dereference_protected(l->avl_right, |
98158f5a | 279 | lockdep_is_held(&base->lock)); |
1da177e4 LT |
280 | lrh = node_height(lr); |
281 | if (lrh <= node_height(ll)) { /* ll: RH+1 */ | |
b914c4ea ED |
282 | RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */ |
283 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 284 | node->avl_height = lrh + 1; /* RH+1 or RH+2 */ |
b914c4ea ED |
285 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */ |
286 | RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */ | |
1da177e4 | 287 | l->avl_height = node->avl_height + 1; |
b914c4ea | 288 | RCU_INIT_POINTER(*nodep, l); |
1da177e4 | 289 | } else { /* ll: RH, lr: RH+1 */ |
b914c4ea | 290 | lrl = rcu_dereference_protected(lr->avl_left, |
98158f5a | 291 | lockdep_is_held(&base->lock)); /* lrl: RH or RH-1 */ |
b914c4ea | 292 | lrr = rcu_dereference_protected(lr->avl_right, |
98158f5a | 293 | lockdep_is_held(&base->lock)); /* lrr: RH or RH-1 */ |
b914c4ea ED |
294 | RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */ |
295 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 296 | node->avl_height = rh + 1; /* node: RH+1 */ |
b914c4ea ED |
297 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */ |
298 | RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */ | |
1da177e4 | 299 | l->avl_height = rh + 1; /* l: RH+1 */ |
b914c4ea ED |
300 | RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */ |
301 | RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */ | |
1da177e4 | 302 | lr->avl_height = rh + 2; |
b914c4ea | 303 | RCU_INIT_POINTER(*nodep, lr); |
1da177e4 LT |
304 | } |
305 | } else if (rh > lh + 1) { /* r: LH+2 */ | |
306 | struct inet_peer *rr, *rl, *rlr, *rll; | |
307 | int rlh; | |
b914c4ea | 308 | rr = rcu_dereference_protected(r->avl_right, |
98158f5a | 309 | lockdep_is_held(&base->lock)); |
b914c4ea | 310 | rl = rcu_dereference_protected(r->avl_left, |
98158f5a | 311 | lockdep_is_held(&base->lock)); |
1da177e4 LT |
312 | rlh = node_height(rl); |
313 | if (rlh <= node_height(rr)) { /* rr: LH+1 */ | |
b914c4ea ED |
314 | RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */ |
315 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 316 | node->avl_height = rlh + 1; /* LH+1 or LH+2 */ |
b914c4ea ED |
317 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */ |
318 | RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */ | |
1da177e4 | 319 | r->avl_height = node->avl_height + 1; |
b914c4ea | 320 | RCU_INIT_POINTER(*nodep, r); |
1da177e4 | 321 | } else { /* rr: RH, rl: RH+1 */ |
b914c4ea | 322 | rlr = rcu_dereference_protected(rl->avl_right, |
98158f5a | 323 | lockdep_is_held(&base->lock)); /* rlr: LH or LH-1 */ |
b914c4ea | 324 | rll = rcu_dereference_protected(rl->avl_left, |
98158f5a | 325 | lockdep_is_held(&base->lock)); /* rll: LH or LH-1 */ |
b914c4ea ED |
326 | RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */ |
327 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 328 | node->avl_height = lh + 1; /* node: LH+1 */ |
b914c4ea ED |
329 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */ |
330 | RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */ | |
1da177e4 | 331 | r->avl_height = lh + 1; /* r: LH+1 */ |
b914c4ea ED |
332 | RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */ |
333 | RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */ | |
1da177e4 | 334 | rl->avl_height = lh + 2; |
b914c4ea | 335 | RCU_INIT_POINTER(*nodep, rl); |
1da177e4 LT |
336 | } |
337 | } else { | |
338 | node->avl_height = (lh > rh ? lh : rh) + 1; | |
339 | } | |
340 | } | |
341 | } | |
342 | ||
aa1039e7 | 343 | /* Called with local BH disabled and the pool lock held. */ |
98158f5a | 344 | #define link_to_pool(n, base) \ |
1da177e4 LT |
345 | do { \ |
346 | n->avl_height = 1; \ | |
b914c4ea ED |
347 | n->avl_left = peer_avl_empty_rcu; \ |
348 | n->avl_right = peer_avl_empty_rcu; \ | |
349 | /* lockless readers can catch us now */ \ | |
350 | rcu_assign_pointer(**--stackptr, n); \ | |
98158f5a | 351 | peer_avl_rebalance(stack, stackptr, base); \ |
d6cc1d64 | 352 | } while (0) |
1da177e4 | 353 | |
aa1039e7 ED |
354 | static void inetpeer_free_rcu(struct rcu_head *head) |
355 | { | |
356 | kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu)); | |
357 | } | |
358 | ||
1da177e4 | 359 | /* May be called with local BH enabled. */ |
98158f5a | 360 | static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base) |
1da177e4 LT |
361 | { |
362 | int do_free; | |
363 | ||
364 | do_free = 0; | |
365 | ||
98158f5a | 366 | spin_lock_bh(&base->lock); |
1da177e4 | 367 | /* Check the reference counter. It was artificially incremented by 1 |
aa1039e7 ED |
368 | * in cleanup() function to prevent sudden disappearing. If we can |
369 | * atomically (because of lockless readers) take this last reference, | |
370 | * it's safe to remove the node and free it later. | |
5f2f8920 | 371 | * We use refcnt=-1 to alert lockless readers this entry is deleted. |
aa1039e7 | 372 | */ |
5f2f8920 | 373 | if (atomic_cmpxchg(&p->refcnt, 1, -1) == 1) { |
b914c4ea ED |
374 | struct inet_peer __rcu **stack[PEER_MAXDEPTH]; |
375 | struct inet_peer __rcu ***stackptr, ***delp; | |
02663045 | 376 | if (lookup(&p->daddr, stack, base) != p) |
1da177e4 LT |
377 | BUG(); |
378 | delp = stackptr - 1; /* *delp[0] == p */ | |
b914c4ea | 379 | if (p->avl_left == peer_avl_empty_rcu) { |
1da177e4 LT |
380 | *delp[0] = p->avl_right; |
381 | --stackptr; | |
382 | } else { | |
383 | /* look for a node to insert instead of p */ | |
384 | struct inet_peer *t; | |
98158f5a | 385 | t = lookup_rightempty(p, base); |
b914c4ea | 386 | BUG_ON(rcu_dereference_protected(*stackptr[-1], |
98158f5a | 387 | lockdep_is_held(&base->lock)) != t); |
1da177e4 | 388 | **--stackptr = t->avl_left; |
582a72da | 389 | /* t is removed, t->daddr > x->daddr for any |
1da177e4 LT |
390 | * x in p->avl_left subtree. |
391 | * Put t in the old place of p. */ | |
b914c4ea | 392 | RCU_INIT_POINTER(*delp[0], t); |
1da177e4 LT |
393 | t->avl_left = p->avl_left; |
394 | t->avl_right = p->avl_right; | |
395 | t->avl_height = p->avl_height; | |
09a62660 | 396 | BUG_ON(delp[1] != &p->avl_left); |
1da177e4 LT |
397 | delp[1] = &t->avl_left; /* was &p->avl_left */ |
398 | } | |
98158f5a DM |
399 | peer_avl_rebalance(stack, stackptr, base); |
400 | base->total--; | |
1da177e4 LT |
401 | do_free = 1; |
402 | } | |
98158f5a | 403 | spin_unlock_bh(&base->lock); |
1da177e4 LT |
404 | |
405 | if (do_free) | |
aa1039e7 | 406 | call_rcu_bh(&p->rcu, inetpeer_free_rcu); |
1da177e4 LT |
407 | else |
408 | /* The node is used again. Decrease the reference counter | |
409 | * back. The loop "cleanup -> unlink_from_unused | |
410 | * -> unlink_from_pool -> putpeer -> link_to_unused | |
411 | * -> cleanup (for the same node)" | |
412 | * doesn't really exist because the entry will have a | |
aa1039e7 ED |
413 | * recent deletion time and will not be cleaned again soon. |
414 | */ | |
1da177e4 LT |
415 | inet_putpeer(p); |
416 | } | |
417 | ||
98158f5a DM |
418 | static struct inet_peer_base *peer_to_base(struct inet_peer *p) |
419 | { | |
420 | return &v4_peers; | |
421 | } | |
422 | ||
1da177e4 LT |
423 | /* May be called with local BH enabled. */ |
424 | static int cleanup_once(unsigned long ttl) | |
425 | { | |
d71209de | 426 | struct inet_peer *p = NULL; |
1da177e4 LT |
427 | |
428 | /* Remove the first entry from the list of unused nodes. */ | |
d6cc1d64 ED |
429 | spin_lock_bh(&unused_peers.lock); |
430 | if (!list_empty(&unused_peers.list)) { | |
d71209de PE |
431 | __u32 delta; |
432 | ||
d6cc1d64 | 433 | p = list_first_entry(&unused_peers.list, struct inet_peer, unused); |
d71209de PE |
434 | delta = (__u32)jiffies - p->dtime; |
435 | ||
4663afe2 | 436 | if (delta < ttl) { |
1da177e4 | 437 | /* Do not prune fresh entries. */ |
d6cc1d64 | 438 | spin_unlock_bh(&unused_peers.lock); |
1da177e4 LT |
439 | return -1; |
440 | } | |
d71209de PE |
441 | |
442 | list_del_init(&p->unused); | |
443 | ||
1da177e4 LT |
444 | /* Grab an extra reference to prevent node disappearing |
445 | * before unlink_from_pool() call. */ | |
446 | atomic_inc(&p->refcnt); | |
447 | } | |
d6cc1d64 | 448 | spin_unlock_bh(&unused_peers.lock); |
1da177e4 LT |
449 | |
450 | if (p == NULL) | |
451 | /* It means that the total number of USED entries has | |
452 | * grown over inet_peer_threshold. It shouldn't really | |
453 | * happen because of entry limits in route cache. */ | |
454 | return -1; | |
455 | ||
98158f5a | 456 | unlink_from_pool(p, peer_to_base(p)); |
1da177e4 LT |
457 | return 0; |
458 | } | |
459 | ||
98158f5a DM |
460 | static struct inet_peer_base *family_to_base(int family) |
461 | { | |
462 | return &v4_peers; | |
463 | } | |
464 | ||
1da177e4 | 465 | /* Called with or without local BH being disabled. */ |
b534ecf1 | 466 | struct inet_peer *inet_getpeer(inet_peer_address_t *daddr, int create) |
1da177e4 | 467 | { |
b914c4ea | 468 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr; |
98158f5a DM |
469 | struct inet_peer_base *base = family_to_base(AF_INET); |
470 | struct inet_peer *p; | |
1da177e4 | 471 | |
aa1039e7 ED |
472 | /* Look up for the address quickly, lockless. |
473 | * Because of a concurrent writer, we might not find an existing entry. | |
474 | */ | |
475 | rcu_read_lock_bh(); | |
02663045 | 476 | p = lookup_rcu_bh(daddr, base); |
aa1039e7 ED |
477 | rcu_read_unlock_bh(); |
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 | } | |
1da177e4 | 486 | |
aa1039e7 ED |
487 | /* retry an exact lookup, taking the lock before. |
488 | * At least, nodes should be hot in our cache. | |
489 | */ | |
98158f5a | 490 | spin_lock_bh(&base->lock); |
02663045 | 491 | p = lookup(daddr, stack, base); |
1da177e4 | 492 | if (p != peer_avl_empty) { |
aa1039e7 | 493 | atomic_inc(&p->refcnt); |
98158f5a | 494 | spin_unlock_bh(&base->lock); |
1da177e4 LT |
495 | /* Remove the entry from unused list if it was there. */ |
496 | unlink_from_unused(p); | |
497 | return p; | |
498 | } | |
aa1039e7 ED |
499 | p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL; |
500 | if (p) { | |
b534ecf1 | 501 | p->daddr = *daddr; |
aa1039e7 ED |
502 | atomic_set(&p->refcnt, 1); |
503 | atomic_set(&p->rid, 0); | |
b534ecf1 | 504 | atomic_set(&p->ip_id_count, secure_ip_id(daddr->a4)); |
aa1039e7 ED |
505 | p->tcp_ts_stamp = 0; |
506 | INIT_LIST_HEAD(&p->unused); | |
507 | ||
508 | ||
509 | /* Link the node. */ | |
98158f5a DM |
510 | link_to_pool(p, base); |
511 | base->total++; | |
aa1039e7 | 512 | } |
98158f5a | 513 | spin_unlock_bh(&base->lock); |
1da177e4 | 514 | |
98158f5a | 515 | if (base->total >= inet_peer_threshold) |
1da177e4 LT |
516 | /* Remove one less-recently-used entry. */ |
517 | cleanup_once(0); | |
518 | ||
1da177e4 LT |
519 | return p; |
520 | } | |
521 | ||
98158f5a DM |
522 | static int compute_total(void) |
523 | { | |
524 | return v4_peers.total; | |
525 | } | |
526 | ||
1da177e4 LT |
527 | /* Called with local BH disabled. */ |
528 | static void peer_check_expire(unsigned long dummy) | |
529 | { | |
4663afe2 | 530 | unsigned long now = jiffies; |
98158f5a | 531 | int ttl, total; |
1da177e4 | 532 | |
98158f5a DM |
533 | total = compute_total(); |
534 | if (total >= inet_peer_threshold) | |
1da177e4 LT |
535 | ttl = inet_peer_minttl; |
536 | else | |
537 | ttl = inet_peer_maxttl | |
538 | - (inet_peer_maxttl - inet_peer_minttl) / HZ * | |
98158f5a | 539 | total / inet_peer_threshold * HZ; |
4663afe2 ED |
540 | while (!cleanup_once(ttl)) { |
541 | if (jiffies != now) | |
542 | break; | |
543 | } | |
1da177e4 LT |
544 | |
545 | /* Trigger the timer after inet_peer_gc_mintime .. inet_peer_gc_maxtime | |
546 | * interval depending on the total number of entries (more entries, | |
547 | * less interval). */ | |
98158f5a DM |
548 | total = compute_total(); |
549 | if (total >= inet_peer_threshold) | |
1344a416 DJ |
550 | peer_periodic_timer.expires = jiffies + inet_peer_gc_mintime; |
551 | else | |
552 | peer_periodic_timer.expires = jiffies | |
553 | + inet_peer_gc_maxtime | |
554 | - (inet_peer_gc_maxtime - inet_peer_gc_mintime) / HZ * | |
98158f5a | 555 | total / inet_peer_threshold * HZ; |
1da177e4 LT |
556 | add_timer(&peer_periodic_timer); |
557 | } | |
4663afe2 ED |
558 | |
559 | void inet_putpeer(struct inet_peer *p) | |
560 | { | |
d6cc1d64 ED |
561 | local_bh_disable(); |
562 | ||
563 | if (atomic_dec_and_lock(&p->refcnt, &unused_peers.lock)) { | |
564 | list_add_tail(&p->unused, &unused_peers.list); | |
4663afe2 | 565 | p->dtime = (__u32)jiffies; |
d6cc1d64 | 566 | spin_unlock(&unused_peers.lock); |
4663afe2 | 567 | } |
d6cc1d64 ED |
568 | |
569 | local_bh_enable(); | |
4663afe2 | 570 | } |