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 | ||
243bbcaa ED |
155 | /* |
156 | * Called with local BH disabled and the pool lock held. | |
243bbcaa | 157 | */ |
98158f5a | 158 | #define lookup(_daddr, _stack, _base) \ |
1da177e4 | 159 | ({ \ |
b914c4ea ED |
160 | struct inet_peer *u; \ |
161 | struct inet_peer __rcu **v; \ | |
aa1039e7 ED |
162 | \ |
163 | stackptr = _stack; \ | |
98158f5a DM |
164 | *stackptr++ = &_base->root; \ |
165 | for (u = rcu_dereference_protected(_base->root, \ | |
166 | lockdep_is_held(&_base->lock)); \ | |
b914c4ea | 167 | u != peer_avl_empty; ) { \ |
582a72da | 168 | if (_daddr == u->daddr.a4) \ |
1da177e4 | 169 | break; \ |
582a72da | 170 | if ((__force __u32)_daddr < (__force __u32)u->daddr.a4) \ |
1da177e4 LT |
171 | v = &u->avl_left; \ |
172 | else \ | |
173 | v = &u->avl_right; \ | |
aa1039e7 | 174 | *stackptr++ = v; \ |
b914c4ea | 175 | u = rcu_dereference_protected(*v, \ |
98158f5a | 176 | lockdep_is_held(&_base->lock)); \ |
1da177e4 LT |
177 | } \ |
178 | u; \ | |
179 | }) | |
180 | ||
aa1039e7 ED |
181 | /* |
182 | * Called with rcu_read_lock_bh() | |
183 | * Because we hold no lock against a writer, its quite possible we fall | |
184 | * in an endless loop. | |
185 | * But every pointer we follow is guaranteed to be valid thanks to RCU. | |
186 | * We exit from this function if number of links exceeds PEER_MAXDEPTH | |
187 | */ | |
98158f5a | 188 | static struct inet_peer *lookup_rcu_bh(__be32 daddr, struct inet_peer_base *base) |
aa1039e7 | 189 | { |
98158f5a | 190 | struct inet_peer *u = rcu_dereference_bh(base->root); |
aa1039e7 ED |
191 | int count = 0; |
192 | ||
193 | while (u != peer_avl_empty) { | |
582a72da | 194 | if (daddr == u->daddr.a4) { |
5f2f8920 ED |
195 | /* Before taking a reference, check if this entry was |
196 | * deleted, unlink_from_pool() sets refcnt=-1 to make | |
197 | * distinction between an unused entry (refcnt=0) and | |
198 | * a freed one. | |
199 | */ | |
200 | if (unlikely(!atomic_add_unless(&u->refcnt, 1, -1))) | |
aa1039e7 ED |
201 | u = NULL; |
202 | return u; | |
203 | } | |
582a72da | 204 | if ((__force __u32)daddr < (__force __u32)u->daddr.a4) |
aa1039e7 ED |
205 | u = rcu_dereference_bh(u->avl_left); |
206 | else | |
207 | u = rcu_dereference_bh(u->avl_right); | |
208 | if (unlikely(++count == PEER_MAXDEPTH)) | |
209 | break; | |
210 | } | |
211 | return NULL; | |
212 | } | |
213 | ||
214 | /* Called with local BH disabled and the pool lock held. */ | |
98158f5a | 215 | #define lookup_rightempty(start, base) \ |
1da177e4 | 216 | ({ \ |
b914c4ea ED |
217 | struct inet_peer *u; \ |
218 | struct inet_peer __rcu **v; \ | |
1da177e4 LT |
219 | *stackptr++ = &start->avl_left; \ |
220 | v = &start->avl_left; \ | |
b914c4ea | 221 | for (u = rcu_dereference_protected(*v, \ |
98158f5a | 222 | lockdep_is_held(&base->lock)); \ |
b914c4ea | 223 | u->avl_right != peer_avl_empty_rcu; ) { \ |
1da177e4 LT |
224 | v = &u->avl_right; \ |
225 | *stackptr++ = v; \ | |
b914c4ea | 226 | u = rcu_dereference_protected(*v, \ |
98158f5a | 227 | lockdep_is_held(&base->lock)); \ |
1da177e4 LT |
228 | } \ |
229 | u; \ | |
230 | }) | |
231 | ||
aa1039e7 | 232 | /* Called with local BH disabled and the pool lock held. |
1da177e4 | 233 | * Variable names are the proof of operation correctness. |
aa1039e7 ED |
234 | * Look into mm/map_avl.c for more detail description of the ideas. |
235 | */ | |
b914c4ea | 236 | static void peer_avl_rebalance(struct inet_peer __rcu **stack[], |
98158f5a DM |
237 | struct inet_peer __rcu ***stackend, |
238 | struct inet_peer_base *base) | |
1da177e4 | 239 | { |
b914c4ea ED |
240 | struct inet_peer __rcu **nodep; |
241 | struct inet_peer *node, *l, *r; | |
1da177e4 LT |
242 | int lh, rh; |
243 | ||
244 | while (stackend > stack) { | |
245 | nodep = *--stackend; | |
b914c4ea | 246 | node = rcu_dereference_protected(*nodep, |
98158f5a | 247 | lockdep_is_held(&base->lock)); |
b914c4ea | 248 | l = rcu_dereference_protected(node->avl_left, |
98158f5a | 249 | lockdep_is_held(&base->lock)); |
b914c4ea | 250 | r = rcu_dereference_protected(node->avl_right, |
98158f5a | 251 | lockdep_is_held(&base->lock)); |
1da177e4 LT |
252 | lh = node_height(l); |
253 | rh = node_height(r); | |
254 | if (lh > rh + 1) { /* l: RH+2 */ | |
255 | struct inet_peer *ll, *lr, *lrl, *lrr; | |
256 | int lrh; | |
b914c4ea | 257 | ll = rcu_dereference_protected(l->avl_left, |
98158f5a | 258 | lockdep_is_held(&base->lock)); |
b914c4ea | 259 | lr = rcu_dereference_protected(l->avl_right, |
98158f5a | 260 | lockdep_is_held(&base->lock)); |
1da177e4 LT |
261 | lrh = node_height(lr); |
262 | if (lrh <= node_height(ll)) { /* ll: RH+1 */ | |
b914c4ea ED |
263 | RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */ |
264 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 265 | node->avl_height = lrh + 1; /* RH+1 or RH+2 */ |
b914c4ea ED |
266 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */ |
267 | RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */ | |
1da177e4 | 268 | l->avl_height = node->avl_height + 1; |
b914c4ea | 269 | RCU_INIT_POINTER(*nodep, l); |
1da177e4 | 270 | } else { /* ll: RH, lr: RH+1 */ |
b914c4ea | 271 | lrl = rcu_dereference_protected(lr->avl_left, |
98158f5a | 272 | lockdep_is_held(&base->lock)); /* lrl: RH or RH-1 */ |
b914c4ea | 273 | lrr = rcu_dereference_protected(lr->avl_right, |
98158f5a | 274 | lockdep_is_held(&base->lock)); /* lrr: RH or RH-1 */ |
b914c4ea ED |
275 | RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */ |
276 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 277 | node->avl_height = rh + 1; /* node: RH+1 */ |
b914c4ea ED |
278 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */ |
279 | RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */ | |
1da177e4 | 280 | l->avl_height = rh + 1; /* l: RH+1 */ |
b914c4ea ED |
281 | RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */ |
282 | RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */ | |
1da177e4 | 283 | lr->avl_height = rh + 2; |
b914c4ea | 284 | RCU_INIT_POINTER(*nodep, lr); |
1da177e4 LT |
285 | } |
286 | } else if (rh > lh + 1) { /* r: LH+2 */ | |
287 | struct inet_peer *rr, *rl, *rlr, *rll; | |
288 | int rlh; | |
b914c4ea | 289 | rr = rcu_dereference_protected(r->avl_right, |
98158f5a | 290 | lockdep_is_held(&base->lock)); |
b914c4ea | 291 | rl = rcu_dereference_protected(r->avl_left, |
98158f5a | 292 | lockdep_is_held(&base->lock)); |
1da177e4 LT |
293 | rlh = node_height(rl); |
294 | if (rlh <= node_height(rr)) { /* rr: LH+1 */ | |
b914c4ea ED |
295 | RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */ |
296 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 297 | node->avl_height = rlh + 1; /* LH+1 or LH+2 */ |
b914c4ea ED |
298 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */ |
299 | RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */ | |
1da177e4 | 300 | r->avl_height = node->avl_height + 1; |
b914c4ea | 301 | RCU_INIT_POINTER(*nodep, r); |
1da177e4 | 302 | } else { /* rr: RH, rl: RH+1 */ |
b914c4ea | 303 | rlr = rcu_dereference_protected(rl->avl_right, |
98158f5a | 304 | lockdep_is_held(&base->lock)); /* rlr: LH or LH-1 */ |
b914c4ea | 305 | rll = rcu_dereference_protected(rl->avl_left, |
98158f5a | 306 | lockdep_is_held(&base->lock)); /* rll: LH or LH-1 */ |
b914c4ea ED |
307 | RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */ |
308 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 309 | node->avl_height = lh + 1; /* node: LH+1 */ |
b914c4ea ED |
310 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */ |
311 | RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */ | |
1da177e4 | 312 | r->avl_height = lh + 1; /* r: LH+1 */ |
b914c4ea ED |
313 | RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */ |
314 | RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */ | |
1da177e4 | 315 | rl->avl_height = lh + 2; |
b914c4ea | 316 | RCU_INIT_POINTER(*nodep, rl); |
1da177e4 LT |
317 | } |
318 | } else { | |
319 | node->avl_height = (lh > rh ? lh : rh) + 1; | |
320 | } | |
321 | } | |
322 | } | |
323 | ||
aa1039e7 | 324 | /* Called with local BH disabled and the pool lock held. */ |
98158f5a | 325 | #define link_to_pool(n, base) \ |
1da177e4 LT |
326 | do { \ |
327 | n->avl_height = 1; \ | |
b914c4ea ED |
328 | n->avl_left = peer_avl_empty_rcu; \ |
329 | n->avl_right = peer_avl_empty_rcu; \ | |
330 | /* lockless readers can catch us now */ \ | |
331 | rcu_assign_pointer(**--stackptr, n); \ | |
98158f5a | 332 | peer_avl_rebalance(stack, stackptr, base); \ |
d6cc1d64 | 333 | } while (0) |
1da177e4 | 334 | |
aa1039e7 ED |
335 | static void inetpeer_free_rcu(struct rcu_head *head) |
336 | { | |
337 | kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu)); | |
338 | } | |
339 | ||
1da177e4 | 340 | /* May be called with local BH enabled. */ |
98158f5a | 341 | static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base) |
1da177e4 LT |
342 | { |
343 | int do_free; | |
344 | ||
345 | do_free = 0; | |
346 | ||
98158f5a | 347 | spin_lock_bh(&base->lock); |
1da177e4 | 348 | /* Check the reference counter. It was artificially incremented by 1 |
aa1039e7 ED |
349 | * in cleanup() function to prevent sudden disappearing. If we can |
350 | * atomically (because of lockless readers) take this last reference, | |
351 | * it's safe to remove the node and free it later. | |
5f2f8920 | 352 | * We use refcnt=-1 to alert lockless readers this entry is deleted. |
aa1039e7 | 353 | */ |
5f2f8920 | 354 | if (atomic_cmpxchg(&p->refcnt, 1, -1) == 1) { |
b914c4ea ED |
355 | struct inet_peer __rcu **stack[PEER_MAXDEPTH]; |
356 | struct inet_peer __rcu ***stackptr, ***delp; | |
582a72da | 357 | if (lookup(p->daddr.a4, stack, base) != p) |
1da177e4 LT |
358 | BUG(); |
359 | delp = stackptr - 1; /* *delp[0] == p */ | |
b914c4ea | 360 | if (p->avl_left == peer_avl_empty_rcu) { |
1da177e4 LT |
361 | *delp[0] = p->avl_right; |
362 | --stackptr; | |
363 | } else { | |
364 | /* look for a node to insert instead of p */ | |
365 | struct inet_peer *t; | |
98158f5a | 366 | t = lookup_rightempty(p, base); |
b914c4ea | 367 | BUG_ON(rcu_dereference_protected(*stackptr[-1], |
98158f5a | 368 | lockdep_is_held(&base->lock)) != t); |
1da177e4 | 369 | **--stackptr = t->avl_left; |
582a72da | 370 | /* t is removed, t->daddr > x->daddr for any |
1da177e4 LT |
371 | * x in p->avl_left subtree. |
372 | * Put t in the old place of p. */ | |
b914c4ea | 373 | RCU_INIT_POINTER(*delp[0], t); |
1da177e4 LT |
374 | t->avl_left = p->avl_left; |
375 | t->avl_right = p->avl_right; | |
376 | t->avl_height = p->avl_height; | |
09a62660 | 377 | BUG_ON(delp[1] != &p->avl_left); |
1da177e4 LT |
378 | delp[1] = &t->avl_left; /* was &p->avl_left */ |
379 | } | |
98158f5a DM |
380 | peer_avl_rebalance(stack, stackptr, base); |
381 | base->total--; | |
1da177e4 LT |
382 | do_free = 1; |
383 | } | |
98158f5a | 384 | spin_unlock_bh(&base->lock); |
1da177e4 LT |
385 | |
386 | if (do_free) | |
aa1039e7 | 387 | call_rcu_bh(&p->rcu, inetpeer_free_rcu); |
1da177e4 LT |
388 | else |
389 | /* The node is used again. Decrease the reference counter | |
390 | * back. The loop "cleanup -> unlink_from_unused | |
391 | * -> unlink_from_pool -> putpeer -> link_to_unused | |
392 | * -> cleanup (for the same node)" | |
393 | * doesn't really exist because the entry will have a | |
aa1039e7 ED |
394 | * recent deletion time and will not be cleaned again soon. |
395 | */ | |
1da177e4 LT |
396 | inet_putpeer(p); |
397 | } | |
398 | ||
98158f5a DM |
399 | static struct inet_peer_base *peer_to_base(struct inet_peer *p) |
400 | { | |
401 | return &v4_peers; | |
402 | } | |
403 | ||
1da177e4 LT |
404 | /* May be called with local BH enabled. */ |
405 | static int cleanup_once(unsigned long ttl) | |
406 | { | |
d71209de | 407 | struct inet_peer *p = NULL; |
1da177e4 LT |
408 | |
409 | /* Remove the first entry from the list of unused nodes. */ | |
d6cc1d64 ED |
410 | spin_lock_bh(&unused_peers.lock); |
411 | if (!list_empty(&unused_peers.list)) { | |
d71209de PE |
412 | __u32 delta; |
413 | ||
d6cc1d64 | 414 | p = list_first_entry(&unused_peers.list, struct inet_peer, unused); |
d71209de PE |
415 | delta = (__u32)jiffies - p->dtime; |
416 | ||
4663afe2 | 417 | if (delta < ttl) { |
1da177e4 | 418 | /* Do not prune fresh entries. */ |
d6cc1d64 | 419 | spin_unlock_bh(&unused_peers.lock); |
1da177e4 LT |
420 | return -1; |
421 | } | |
d71209de PE |
422 | |
423 | list_del_init(&p->unused); | |
424 | ||
1da177e4 LT |
425 | /* Grab an extra reference to prevent node disappearing |
426 | * before unlink_from_pool() call. */ | |
427 | atomic_inc(&p->refcnt); | |
428 | } | |
d6cc1d64 | 429 | spin_unlock_bh(&unused_peers.lock); |
1da177e4 LT |
430 | |
431 | if (p == NULL) | |
432 | /* It means that the total number of USED entries has | |
433 | * grown over inet_peer_threshold. It shouldn't really | |
434 | * happen because of entry limits in route cache. */ | |
435 | return -1; | |
436 | ||
98158f5a | 437 | unlink_from_pool(p, peer_to_base(p)); |
1da177e4 LT |
438 | return 0; |
439 | } | |
440 | ||
98158f5a DM |
441 | static struct inet_peer_base *family_to_base(int family) |
442 | { | |
443 | return &v4_peers; | |
444 | } | |
445 | ||
1da177e4 | 446 | /* Called with or without local BH being disabled. */ |
53576d9b | 447 | struct inet_peer *inet_getpeer(__be32 daddr, int create) |
1da177e4 | 448 | { |
b914c4ea | 449 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr; |
98158f5a DM |
450 | struct inet_peer_base *base = family_to_base(AF_INET); |
451 | struct inet_peer *p; | |
1da177e4 | 452 | |
aa1039e7 ED |
453 | /* Look up for the address quickly, lockless. |
454 | * Because of a concurrent writer, we might not find an existing entry. | |
455 | */ | |
456 | rcu_read_lock_bh(); | |
98158f5a | 457 | p = lookup_rcu_bh(daddr, base); |
aa1039e7 ED |
458 | rcu_read_unlock_bh(); |
459 | ||
460 | if (p) { | |
461 | /* The existing node has been found. | |
462 | * Remove the entry from unused list if it was there. | |
463 | */ | |
464 | unlink_from_unused(p); | |
465 | return p; | |
466 | } | |
1da177e4 | 467 | |
aa1039e7 ED |
468 | /* retry an exact lookup, taking the lock before. |
469 | * At least, nodes should be hot in our cache. | |
470 | */ | |
98158f5a DM |
471 | spin_lock_bh(&base->lock); |
472 | p = lookup(daddr, stack, base); | |
1da177e4 | 473 | if (p != peer_avl_empty) { |
aa1039e7 | 474 | atomic_inc(&p->refcnt); |
98158f5a | 475 | spin_unlock_bh(&base->lock); |
1da177e4 LT |
476 | /* Remove the entry from unused list if it was there. */ |
477 | unlink_from_unused(p); | |
478 | return p; | |
479 | } | |
aa1039e7 ED |
480 | p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL; |
481 | if (p) { | |
582a72da | 482 | p->daddr.a4 = daddr; |
aa1039e7 ED |
483 | atomic_set(&p->refcnt, 1); |
484 | atomic_set(&p->rid, 0); | |
485 | atomic_set(&p->ip_id_count, secure_ip_id(daddr)); | |
486 | p->tcp_ts_stamp = 0; | |
487 | INIT_LIST_HEAD(&p->unused); | |
488 | ||
489 | ||
490 | /* Link the node. */ | |
98158f5a DM |
491 | link_to_pool(p, base); |
492 | base->total++; | |
aa1039e7 | 493 | } |
98158f5a | 494 | spin_unlock_bh(&base->lock); |
1da177e4 | 495 | |
98158f5a | 496 | if (base->total >= inet_peer_threshold) |
1da177e4 LT |
497 | /* Remove one less-recently-used entry. */ |
498 | cleanup_once(0); | |
499 | ||
1da177e4 LT |
500 | return p; |
501 | } | |
502 | ||
98158f5a DM |
503 | static int compute_total(void) |
504 | { | |
505 | return v4_peers.total; | |
506 | } | |
507 | ||
1da177e4 LT |
508 | /* Called with local BH disabled. */ |
509 | static void peer_check_expire(unsigned long dummy) | |
510 | { | |
4663afe2 | 511 | unsigned long now = jiffies; |
98158f5a | 512 | int ttl, total; |
1da177e4 | 513 | |
98158f5a DM |
514 | total = compute_total(); |
515 | if (total >= inet_peer_threshold) | |
1da177e4 LT |
516 | ttl = inet_peer_minttl; |
517 | else | |
518 | ttl = inet_peer_maxttl | |
519 | - (inet_peer_maxttl - inet_peer_minttl) / HZ * | |
98158f5a | 520 | total / inet_peer_threshold * HZ; |
4663afe2 ED |
521 | while (!cleanup_once(ttl)) { |
522 | if (jiffies != now) | |
523 | break; | |
524 | } | |
1da177e4 LT |
525 | |
526 | /* Trigger the timer after inet_peer_gc_mintime .. inet_peer_gc_maxtime | |
527 | * interval depending on the total number of entries (more entries, | |
528 | * less interval). */ | |
98158f5a DM |
529 | total = compute_total(); |
530 | if (total >= inet_peer_threshold) | |
1344a416 DJ |
531 | peer_periodic_timer.expires = jiffies + inet_peer_gc_mintime; |
532 | else | |
533 | peer_periodic_timer.expires = jiffies | |
534 | + inet_peer_gc_maxtime | |
535 | - (inet_peer_gc_maxtime - inet_peer_gc_mintime) / HZ * | |
98158f5a | 536 | total / inet_peer_threshold * HZ; |
1da177e4 LT |
537 | add_timer(&peer_periodic_timer); |
538 | } | |
4663afe2 ED |
539 | |
540 | void inet_putpeer(struct inet_peer *p) | |
541 | { | |
d6cc1d64 ED |
542 | local_bh_disable(); |
543 | ||
544 | if (atomic_dec_and_lock(&p->refcnt, &unused_peers.lock)) { | |
545 | list_add_tail(&p->unused, &unused_peers.list); | |
4663afe2 | 546 | p->dtime = (__u32)jiffies; |
d6cc1d64 | 547 | spin_unlock(&unused_peers.lock); |
4663afe2 | 548 | } |
d6cc1d64 ED |
549 | |
550 | local_bh_enable(); | |
4663afe2 | 551 | } |