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 | |
021e9299 | 82 | struct inet_peer_base { |
b914c4ea | 83 | struct inet_peer __rcu *root; |
65e8354e | 84 | seqlock_t lock; |
d6cc1d64 | 85 | int total; |
021e9299 DM |
86 | }; |
87 | ||
88 | static struct inet_peer_base v4_peers = { | |
b914c4ea | 89 | .root = peer_avl_empty_rcu, |
65e8354e | 90 | .lock = __SEQLOCK_UNLOCKED(v4_peers.lock), |
d6cc1d64 ED |
91 | .total = 0, |
92 | }; | |
021e9299 DM |
93 | |
94 | static struct inet_peer_base v6_peers = { | |
95 | .root = peer_avl_empty_rcu, | |
65e8354e | 96 | .lock = __SEQLOCK_UNLOCKED(v6_peers.lock), |
021e9299 DM |
97 | .total = 0, |
98 | }; | |
99 | ||
1da177e4 LT |
100 | #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */ |
101 | ||
1da177e4 | 102 | /* Exported for sysctl_net_ipv4. */ |
243bbcaa | 103 | int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more |
1da177e4 | 104 | * aggressively at this stage */ |
243bbcaa ED |
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; | |
1da177e4 | 109 | |
d6cc1d64 ED |
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 | }; | |
1da177e4 LT |
117 | |
118 | static void peer_check_expire(unsigned long dummy); | |
8d06afab | 119 | static DEFINE_TIMER(peer_periodic_timer, peer_check_expire, 0, 0); |
1da177e4 | 120 | |
1da177e4 LT |
121 | |
122 | /* Called from ip_output.c:ip_init */ | |
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), | |
317fe0e6 | 142 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, |
20c2df83 | 143 | NULL); |
1da177e4 | 144 | |
1da177e4 LT |
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. */ | |
155 | static void unlink_from_unused(struct inet_peer *p) | |
156 | { | |
686a7e32 ED |
157 | spin_lock_bh(&unused_peers.lock); |
158 | list_del_init(&p->unused); | |
159 | spin_unlock_bh(&unused_peers.lock); | |
1da177e4 LT |
160 | } |
161 | ||
8790ca17 DM |
162 | static int addr_compare(const struct inetpeer_addr *a, |
163 | const struct inetpeer_addr *b) | |
02663045 DM |
164 | { |
165 | int i, n = (a->family == AF_INET ? 1 : 4); | |
166 | ||
167 | for (i = 0; i < n; i++) { | |
7a71ed89 | 168 | if (a->addr.a6[i] == b->addr.a6[i]) |
02663045 | 169 | continue; |
7a71ed89 | 170 | if (a->addr.a6[i] < b->addr.a6[i]) |
02663045 DM |
171 | return -1; |
172 | return 1; | |
173 | } | |
174 | ||
175 | return 0; | |
176 | } | |
177 | ||
65e8354e ED |
178 | #define rcu_deref_locked(X, BASE) \ |
179 | rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock)) | |
180 | ||
243bbcaa ED |
181 | /* |
182 | * Called with local BH disabled and the pool lock held. | |
243bbcaa | 183 | */ |
98158f5a | 184 | #define lookup(_daddr, _stack, _base) \ |
1da177e4 | 185 | ({ \ |
b914c4ea ED |
186 | struct inet_peer *u; \ |
187 | struct inet_peer __rcu **v; \ | |
aa1039e7 ED |
188 | \ |
189 | stackptr = _stack; \ | |
98158f5a | 190 | *stackptr++ = &_base->root; \ |
65e8354e | 191 | for (u = rcu_deref_locked(_base->root, _base); \ |
b914c4ea | 192 | u != peer_avl_empty; ) { \ |
02663045 DM |
193 | int cmp = addr_compare(_daddr, &u->daddr); \ |
194 | if (cmp == 0) \ | |
1da177e4 | 195 | break; \ |
02663045 | 196 | if (cmp == -1) \ |
1da177e4 LT |
197 | v = &u->avl_left; \ |
198 | else \ | |
199 | v = &u->avl_right; \ | |
aa1039e7 | 200 | *stackptr++ = v; \ |
65e8354e | 201 | u = rcu_deref_locked(*v, _base); \ |
1da177e4 LT |
202 | } \ |
203 | u; \ | |
204 | }) | |
205 | ||
686a7e32 ED |
206 | static bool atomic_add_unless_return(atomic_t *ptr, int a, int u, int *newv) |
207 | { | |
208 | int cur, old = atomic_read(ptr); | |
209 | ||
210 | while (old != u) { | |
211 | *newv = old + a; | |
212 | cur = atomic_cmpxchg(ptr, old, *newv); | |
213 | if (cur == old) | |
214 | return true; | |
215 | old = cur; | |
216 | } | |
217 | return false; | |
218 | } | |
219 | ||
aa1039e7 | 220 | /* |
7b46ac4e | 221 | * Called with rcu_read_lock() |
aa1039e7 ED |
222 | * Because we hold no lock against a writer, its quite possible we fall |
223 | * in an endless loop. | |
224 | * But every pointer we follow is guaranteed to be valid thanks to RCU. | |
225 | * We exit from this function if number of links exceeds PEER_MAXDEPTH | |
226 | */ | |
7b46ac4e | 227 | static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr, |
686a7e32 ED |
228 | struct inet_peer_base *base, |
229 | int *newrefcnt) | |
aa1039e7 | 230 | { |
7b46ac4e | 231 | struct inet_peer *u = rcu_dereference(base->root); |
aa1039e7 ED |
232 | int count = 0; |
233 | ||
234 | while (u != peer_avl_empty) { | |
02663045 DM |
235 | int cmp = addr_compare(daddr, &u->daddr); |
236 | if (cmp == 0) { | |
5f2f8920 ED |
237 | /* Before taking a reference, check if this entry was |
238 | * deleted, unlink_from_pool() sets refcnt=-1 to make | |
239 | * distinction between an unused entry (refcnt=0) and | |
240 | * a freed one. | |
241 | */ | |
686a7e32 | 242 | if (!atomic_add_unless_return(&u->refcnt, 1, -1, newrefcnt)) |
aa1039e7 ED |
243 | u = NULL; |
244 | return u; | |
245 | } | |
02663045 | 246 | if (cmp == -1) |
7b46ac4e | 247 | u = rcu_dereference(u->avl_left); |
aa1039e7 | 248 | else |
7b46ac4e | 249 | u = rcu_dereference(u->avl_right); |
aa1039e7 ED |
250 | if (unlikely(++count == PEER_MAXDEPTH)) |
251 | break; | |
252 | } | |
253 | return NULL; | |
254 | } | |
255 | ||
256 | /* Called with local BH disabled and the pool lock held. */ | |
98158f5a | 257 | #define lookup_rightempty(start, base) \ |
1da177e4 | 258 | ({ \ |
b914c4ea ED |
259 | struct inet_peer *u; \ |
260 | struct inet_peer __rcu **v; \ | |
1da177e4 LT |
261 | *stackptr++ = &start->avl_left; \ |
262 | v = &start->avl_left; \ | |
65e8354e | 263 | for (u = rcu_deref_locked(*v, base); \ |
b914c4ea | 264 | u->avl_right != peer_avl_empty_rcu; ) { \ |
1da177e4 LT |
265 | v = &u->avl_right; \ |
266 | *stackptr++ = v; \ | |
65e8354e | 267 | u = rcu_deref_locked(*v, base); \ |
1da177e4 LT |
268 | } \ |
269 | u; \ | |
270 | }) | |
271 | ||
aa1039e7 | 272 | /* Called with local BH disabled and the pool lock held. |
1da177e4 | 273 | * Variable names are the proof of operation correctness. |
aa1039e7 ED |
274 | * Look into mm/map_avl.c for more detail description of the ideas. |
275 | */ | |
b914c4ea | 276 | static void peer_avl_rebalance(struct inet_peer __rcu **stack[], |
98158f5a DM |
277 | struct inet_peer __rcu ***stackend, |
278 | struct inet_peer_base *base) | |
1da177e4 | 279 | { |
b914c4ea ED |
280 | struct inet_peer __rcu **nodep; |
281 | struct inet_peer *node, *l, *r; | |
1da177e4 LT |
282 | int lh, rh; |
283 | ||
284 | while (stackend > stack) { | |
285 | nodep = *--stackend; | |
65e8354e ED |
286 | node = rcu_deref_locked(*nodep, base); |
287 | l = rcu_deref_locked(node->avl_left, base); | |
288 | r = rcu_deref_locked(node->avl_right, base); | |
1da177e4 LT |
289 | lh = node_height(l); |
290 | rh = node_height(r); | |
291 | if (lh > rh + 1) { /* l: RH+2 */ | |
292 | struct inet_peer *ll, *lr, *lrl, *lrr; | |
293 | int lrh; | |
65e8354e ED |
294 | ll = rcu_deref_locked(l->avl_left, base); |
295 | lr = rcu_deref_locked(l->avl_right, base); | |
1da177e4 LT |
296 | lrh = node_height(lr); |
297 | if (lrh <= node_height(ll)) { /* ll: RH+1 */ | |
b914c4ea ED |
298 | RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */ |
299 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 300 | node->avl_height = lrh + 1; /* RH+1 or RH+2 */ |
b914c4ea ED |
301 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */ |
302 | RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */ | |
1da177e4 | 303 | l->avl_height = node->avl_height + 1; |
b914c4ea | 304 | RCU_INIT_POINTER(*nodep, l); |
1da177e4 | 305 | } else { /* ll: RH, lr: RH+1 */ |
65e8354e ED |
306 | lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */ |
307 | lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */ | |
b914c4ea ED |
308 | RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */ |
309 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 310 | node->avl_height = rh + 1; /* node: RH+1 */ |
b914c4ea ED |
311 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */ |
312 | RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */ | |
1da177e4 | 313 | l->avl_height = rh + 1; /* l: RH+1 */ |
b914c4ea ED |
314 | RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */ |
315 | RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */ | |
1da177e4 | 316 | lr->avl_height = rh + 2; |
b914c4ea | 317 | RCU_INIT_POINTER(*nodep, lr); |
1da177e4 LT |
318 | } |
319 | } else if (rh > lh + 1) { /* r: LH+2 */ | |
320 | struct inet_peer *rr, *rl, *rlr, *rll; | |
321 | int rlh; | |
65e8354e ED |
322 | rr = rcu_deref_locked(r->avl_right, base); |
323 | rl = rcu_deref_locked(r->avl_left, base); | |
1da177e4 LT |
324 | rlh = node_height(rl); |
325 | if (rlh <= node_height(rr)) { /* rr: LH+1 */ | |
b914c4ea ED |
326 | RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */ |
327 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 328 | node->avl_height = rlh + 1; /* LH+1 or LH+2 */ |
b914c4ea ED |
329 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */ |
330 | RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */ | |
1da177e4 | 331 | r->avl_height = node->avl_height + 1; |
b914c4ea | 332 | RCU_INIT_POINTER(*nodep, r); |
1da177e4 | 333 | } else { /* rr: RH, rl: RH+1 */ |
65e8354e ED |
334 | rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */ |
335 | rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */ | |
b914c4ea ED |
336 | RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */ |
337 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 338 | node->avl_height = lh + 1; /* node: LH+1 */ |
b914c4ea ED |
339 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */ |
340 | RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */ | |
1da177e4 | 341 | r->avl_height = lh + 1; /* r: LH+1 */ |
b914c4ea ED |
342 | RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */ |
343 | RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */ | |
1da177e4 | 344 | rl->avl_height = lh + 2; |
b914c4ea | 345 | RCU_INIT_POINTER(*nodep, rl); |
1da177e4 LT |
346 | } |
347 | } else { | |
348 | node->avl_height = (lh > rh ? lh : rh) + 1; | |
349 | } | |
350 | } | |
351 | } | |
352 | ||
aa1039e7 | 353 | /* Called with local BH disabled and the pool lock held. */ |
98158f5a | 354 | #define link_to_pool(n, base) \ |
1da177e4 LT |
355 | do { \ |
356 | n->avl_height = 1; \ | |
b914c4ea ED |
357 | n->avl_left = peer_avl_empty_rcu; \ |
358 | n->avl_right = peer_avl_empty_rcu; \ | |
359 | /* lockless readers can catch us now */ \ | |
360 | rcu_assign_pointer(**--stackptr, n); \ | |
98158f5a | 361 | peer_avl_rebalance(stack, stackptr, base); \ |
d6cc1d64 | 362 | } while (0) |
1da177e4 | 363 | |
aa1039e7 ED |
364 | static void inetpeer_free_rcu(struct rcu_head *head) |
365 | { | |
366 | kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu)); | |
367 | } | |
368 | ||
1da177e4 | 369 | /* May be called with local BH enabled. */ |
66944e1c ED |
370 | static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base, |
371 | struct inet_peer __rcu **stack[PEER_MAXDEPTH]) | |
1da177e4 LT |
372 | { |
373 | int do_free; | |
374 | ||
375 | do_free = 0; | |
376 | ||
65e8354e | 377 | write_seqlock_bh(&base->lock); |
1da177e4 | 378 | /* Check the reference counter. It was artificially incremented by 1 |
aa1039e7 ED |
379 | * in cleanup() function to prevent sudden disappearing. If we can |
380 | * atomically (because of lockless readers) take this last reference, | |
381 | * it's safe to remove the node and free it later. | |
5f2f8920 | 382 | * We use refcnt=-1 to alert lockless readers this entry is deleted. |
aa1039e7 | 383 | */ |
5f2f8920 | 384 | if (atomic_cmpxchg(&p->refcnt, 1, -1) == 1) { |
b914c4ea | 385 | struct inet_peer __rcu ***stackptr, ***delp; |
02663045 | 386 | if (lookup(&p->daddr, stack, base) != p) |
1da177e4 LT |
387 | BUG(); |
388 | delp = stackptr - 1; /* *delp[0] == p */ | |
b914c4ea | 389 | if (p->avl_left == peer_avl_empty_rcu) { |
1da177e4 LT |
390 | *delp[0] = p->avl_right; |
391 | --stackptr; | |
392 | } else { | |
393 | /* look for a node to insert instead of p */ | |
394 | struct inet_peer *t; | |
98158f5a | 395 | t = lookup_rightempty(p, base); |
65e8354e | 396 | BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t); |
1da177e4 | 397 | **--stackptr = t->avl_left; |
582a72da | 398 | /* t is removed, t->daddr > x->daddr for any |
1da177e4 LT |
399 | * x in p->avl_left subtree. |
400 | * Put t in the old place of p. */ | |
b914c4ea | 401 | RCU_INIT_POINTER(*delp[0], t); |
1da177e4 LT |
402 | t->avl_left = p->avl_left; |
403 | t->avl_right = p->avl_right; | |
404 | t->avl_height = p->avl_height; | |
09a62660 | 405 | BUG_ON(delp[1] != &p->avl_left); |
1da177e4 LT |
406 | delp[1] = &t->avl_left; /* was &p->avl_left */ |
407 | } | |
98158f5a DM |
408 | peer_avl_rebalance(stack, stackptr, base); |
409 | base->total--; | |
1da177e4 LT |
410 | do_free = 1; |
411 | } | |
65e8354e | 412 | write_sequnlock_bh(&base->lock); |
1da177e4 LT |
413 | |
414 | if (do_free) | |
4e75db2e | 415 | call_rcu(&p->rcu, inetpeer_free_rcu); |
1da177e4 LT |
416 | else |
417 | /* The node is used again. Decrease the reference counter | |
418 | * back. The loop "cleanup -> unlink_from_unused | |
419 | * -> unlink_from_pool -> putpeer -> link_to_unused | |
420 | * -> cleanup (for the same node)" | |
421 | * doesn't really exist because the entry will have a | |
aa1039e7 ED |
422 | * recent deletion time and will not be cleaned again soon. |
423 | */ | |
1da177e4 LT |
424 | inet_putpeer(p); |
425 | } | |
426 | ||
021e9299 DM |
427 | static struct inet_peer_base *family_to_base(int family) |
428 | { | |
429 | return (family == AF_INET ? &v4_peers : &v6_peers); | |
430 | } | |
431 | ||
98158f5a DM |
432 | static struct inet_peer_base *peer_to_base(struct inet_peer *p) |
433 | { | |
021e9299 | 434 | return family_to_base(p->daddr.family); |
98158f5a DM |
435 | } |
436 | ||
1da177e4 | 437 | /* May be called with local BH enabled. */ |
66944e1c | 438 | static int cleanup_once(unsigned long ttl, struct inet_peer __rcu **stack[PEER_MAXDEPTH]) |
1da177e4 | 439 | { |
d71209de | 440 | struct inet_peer *p = NULL; |
1da177e4 LT |
441 | |
442 | /* Remove the first entry from the list of unused nodes. */ | |
d6cc1d64 ED |
443 | spin_lock_bh(&unused_peers.lock); |
444 | if (!list_empty(&unused_peers.list)) { | |
d71209de PE |
445 | __u32 delta; |
446 | ||
d6cc1d64 | 447 | p = list_first_entry(&unused_peers.list, struct inet_peer, unused); |
d71209de PE |
448 | delta = (__u32)jiffies - p->dtime; |
449 | ||
4663afe2 | 450 | if (delta < ttl) { |
1da177e4 | 451 | /* Do not prune fresh entries. */ |
d6cc1d64 | 452 | spin_unlock_bh(&unused_peers.lock); |
1da177e4 LT |
453 | return -1; |
454 | } | |
d71209de PE |
455 | |
456 | list_del_init(&p->unused); | |
457 | ||
1da177e4 LT |
458 | /* Grab an extra reference to prevent node disappearing |
459 | * before unlink_from_pool() call. */ | |
460 | atomic_inc(&p->refcnt); | |
461 | } | |
d6cc1d64 | 462 | spin_unlock_bh(&unused_peers.lock); |
1da177e4 LT |
463 | |
464 | if (p == NULL) | |
465 | /* It means that the total number of USED entries has | |
466 | * grown over inet_peer_threshold. It shouldn't really | |
467 | * happen because of entry limits in route cache. */ | |
468 | return -1; | |
469 | ||
66944e1c | 470 | unlink_from_pool(p, peer_to_base(p), stack); |
1da177e4 LT |
471 | return 0; |
472 | } | |
473 | ||
474 | /* Called with or without local BH being disabled. */ | |
8790ca17 | 475 | struct inet_peer *inet_getpeer(struct inetpeer_addr *daddr, int create) |
1da177e4 | 476 | { |
b914c4ea | 477 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr; |
3408404a | 478 | struct inet_peer_base *base = family_to_base(daddr->family); |
98158f5a | 479 | struct inet_peer *p; |
65e8354e | 480 | unsigned int sequence; |
686a7e32 | 481 | int invalidated, newrefcnt = 0; |
1da177e4 | 482 | |
aa1039e7 ED |
483 | /* Look up for the address quickly, lockless. |
484 | * Because of a concurrent writer, we might not find an existing entry. | |
485 | */ | |
7b46ac4e | 486 | rcu_read_lock(); |
65e8354e | 487 | sequence = read_seqbegin(&base->lock); |
686a7e32 | 488 | p = lookup_rcu(daddr, base, &newrefcnt); |
65e8354e | 489 | invalidated = read_seqretry(&base->lock, sequence); |
7b46ac4e | 490 | rcu_read_unlock(); |
aa1039e7 ED |
491 | |
492 | if (p) { | |
686a7e32 | 493 | found: /* The existing node has been found. |
aa1039e7 ED |
494 | * Remove the entry from unused list if it was there. |
495 | */ | |
686a7e32 ED |
496 | if (newrefcnt == 1) |
497 | unlink_from_unused(p); | |
aa1039e7 ED |
498 | return p; |
499 | } | |
1da177e4 | 500 | |
65e8354e ED |
501 | /* If no writer did a change during our lookup, we can return early. */ |
502 | if (!create && !invalidated) | |
503 | return NULL; | |
504 | ||
aa1039e7 ED |
505 | /* retry an exact lookup, taking the lock before. |
506 | * At least, nodes should be hot in our cache. | |
507 | */ | |
65e8354e | 508 | write_seqlock_bh(&base->lock); |
02663045 | 509 | p = lookup(daddr, stack, base); |
1da177e4 | 510 | if (p != peer_avl_empty) { |
686a7e32 | 511 | newrefcnt = atomic_inc_return(&p->refcnt); |
65e8354e | 512 | write_sequnlock_bh(&base->lock); |
686a7e32 | 513 | goto found; |
1da177e4 | 514 | } |
aa1039e7 ED |
515 | p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL; |
516 | if (p) { | |
b534ecf1 | 517 | p->daddr = *daddr; |
aa1039e7 ED |
518 | atomic_set(&p->refcnt, 1); |
519 | atomic_set(&p->rid, 0); | |
7a71ed89 | 520 | atomic_set(&p->ip_id_count, secure_ip_id(daddr->addr.a4)); |
aa1039e7 | 521 | p->tcp_ts_stamp = 0; |
144001bd | 522 | p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW; |
92d86829 DM |
523 | p->rate_tokens = 0; |
524 | p->rate_last = 0; | |
ddd4aa42 | 525 | p->pmtu_expires = 0; |
46af3180 | 526 | p->pmtu_orig = 0; |
ddd4aa42 | 527 | memset(&p->redirect_learned, 0, sizeof(p->redirect_learned)); |
aa1039e7 ED |
528 | INIT_LIST_HEAD(&p->unused); |
529 | ||
530 | ||
531 | /* Link the node. */ | |
98158f5a DM |
532 | link_to_pool(p, base); |
533 | base->total++; | |
aa1039e7 | 534 | } |
65e8354e | 535 | write_sequnlock_bh(&base->lock); |
1da177e4 | 536 | |
98158f5a | 537 | if (base->total >= inet_peer_threshold) |
1da177e4 | 538 | /* Remove one less-recently-used entry. */ |
66944e1c | 539 | cleanup_once(0, stack); |
1da177e4 | 540 | |
1da177e4 LT |
541 | return p; |
542 | } | |
543 | ||
98158f5a DM |
544 | static int compute_total(void) |
545 | { | |
021e9299 | 546 | return v4_peers.total + v6_peers.total; |
98158f5a | 547 | } |
b3419363 | 548 | EXPORT_SYMBOL_GPL(inet_getpeer); |
98158f5a | 549 | |
1da177e4 LT |
550 | /* Called with local BH disabled. */ |
551 | static void peer_check_expire(unsigned long dummy) | |
552 | { | |
4663afe2 | 553 | unsigned long now = jiffies; |
98158f5a | 554 | int ttl, total; |
66944e1c | 555 | struct inet_peer __rcu **stack[PEER_MAXDEPTH]; |
1da177e4 | 556 | |
98158f5a DM |
557 | total = compute_total(); |
558 | if (total >= inet_peer_threshold) | |
1da177e4 LT |
559 | ttl = inet_peer_minttl; |
560 | else | |
561 | ttl = inet_peer_maxttl | |
562 | - (inet_peer_maxttl - inet_peer_minttl) / HZ * | |
98158f5a | 563 | total / inet_peer_threshold * HZ; |
66944e1c | 564 | while (!cleanup_once(ttl, stack)) { |
4663afe2 ED |
565 | if (jiffies != now) |
566 | break; | |
567 | } | |
1da177e4 LT |
568 | |
569 | /* Trigger the timer after inet_peer_gc_mintime .. inet_peer_gc_maxtime | |
570 | * interval depending on the total number of entries (more entries, | |
571 | * less interval). */ | |
98158f5a DM |
572 | total = compute_total(); |
573 | if (total >= inet_peer_threshold) | |
1344a416 DJ |
574 | peer_periodic_timer.expires = jiffies + inet_peer_gc_mintime; |
575 | else | |
576 | peer_periodic_timer.expires = jiffies | |
577 | + inet_peer_gc_maxtime | |
578 | - (inet_peer_gc_maxtime - inet_peer_gc_mintime) / HZ * | |
98158f5a | 579 | total / inet_peer_threshold * HZ; |
1da177e4 LT |
580 | add_timer(&peer_periodic_timer); |
581 | } | |
4663afe2 ED |
582 | |
583 | void inet_putpeer(struct inet_peer *p) | |
584 | { | |
d6cc1d64 ED |
585 | local_bh_disable(); |
586 | ||
587 | if (atomic_dec_and_lock(&p->refcnt, &unused_peers.lock)) { | |
588 | list_add_tail(&p->unused, &unused_peers.list); | |
4663afe2 | 589 | p->dtime = (__u32)jiffies; |
d6cc1d64 | 590 | spin_unlock(&unused_peers.lock); |
4663afe2 | 591 | } |
d6cc1d64 ED |
592 | |
593 | local_bh_enable(); | |
4663afe2 | 594 | } |
b3419363 | 595 | EXPORT_SYMBOL_GPL(inet_putpeer); |
92d86829 DM |
596 | |
597 | /* | |
598 | * Check transmit rate limitation for given message. | |
599 | * The rate information is held in the inet_peer entries now. | |
600 | * This function is generic and could be used for other purposes | |
601 | * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. | |
602 | * | |
603 | * Note that the same inet_peer fields are modified by functions in | |
604 | * route.c too, but these work for packet destinations while xrlim_allow | |
605 | * works for icmp destinations. This means the rate limiting information | |
606 | * for one "ip object" is shared - and these ICMPs are twice limited: | |
607 | * by source and by destination. | |
608 | * | |
609 | * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate | |
610 | * SHOULD allow setting of rate limits | |
611 | * | |
612 | * Shared between ICMPv4 and ICMPv6. | |
613 | */ | |
614 | #define XRLIM_BURST_FACTOR 6 | |
615 | bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout) | |
616 | { | |
617 | unsigned long now, token; | |
618 | bool rc = false; | |
619 | ||
620 | if (!peer) | |
621 | return true; | |
622 | ||
623 | token = peer->rate_tokens; | |
624 | now = jiffies; | |
625 | token += now - peer->rate_last; | |
626 | peer->rate_last = now; | |
627 | if (token > XRLIM_BURST_FACTOR * timeout) | |
628 | token = XRLIM_BURST_FACTOR * timeout; | |
629 | if (token >= timeout) { | |
630 | token -= timeout; | |
631 | rc = true; | |
632 | } | |
633 | peer->rate_tokens = token; | |
634 | return rc; | |
635 | } | |
636 | EXPORT_SYMBOL(inet_peer_xrlim_allow); |