]>
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 | 21 | #include <net/inetpeer.h> |
6e5714ea | 22 | #include <net/secure_seq.h> |
1da177e4 LT |
23 | |
24 | /* | |
25 | * Theory of operations. | |
26 | * We keep one entry for each peer IP address. The nodes contains long-living | |
27 | * information about the peer which doesn't depend on routes. | |
28 | * At this moment this information consists only of ID field for the next | |
29 | * outgoing IP packet. This field is incremented with each packet as encoded | |
30 | * in inet_getid() function (include/net/inetpeer.h). | |
31 | * At the moment of writing this notes identifier of IP packets is generated | |
32 | * to be unpredictable using this code only for packets subjected | |
33 | * (actually or potentially) to defragmentation. I.e. DF packets less than | |
34 | * PMTU in size uses a constant ID and do not use this code (see | |
35 | * ip_select_ident() in include/net/ip.h). | |
36 | * | |
37 | * Route cache entries hold references to our nodes. | |
38 | * New cache entries get references via lookup by destination IP address in | |
39 | * the avl tree. The reference is grabbed only when it's needed i.e. only | |
40 | * when we try to output IP packet which needs an unpredictable ID (see | |
41 | * __ip_select_ident() in net/ipv4/route.c). | |
42 | * Nodes are removed only when reference counter goes to 0. | |
43 | * When it's happened the node may be removed when a sufficient amount of | |
44 | * time has been passed since its last use. The less-recently-used entry can | |
45 | * also be removed if the pool is overloaded i.e. if the total amount of | |
46 | * entries is greater-or-equal than the threshold. | |
47 | * | |
48 | * Node pool is organised as an AVL tree. | |
49 | * Such an implementation has been chosen not just for fun. It's a way to | |
50 | * prevent easy and efficient DoS attacks by creating hash collisions. A huge | |
51 | * amount of long living nodes in a single hash slot would significantly delay | |
52 | * lookups performed with disabled BHs. | |
53 | * | |
54 | * Serialisation issues. | |
aa1039e7 ED |
55 | * 1. Nodes may appear in the tree only with the pool lock held. |
56 | * 2. Nodes may disappear from the tree only with the pool lock held | |
1da177e4 | 57 | * AND reference count being 0. |
4b9d9be8 ED |
58 | * 3. Global variable peer_total is modified under the pool lock. |
59 | * 4. struct inet_peer fields modification: | |
1da177e4 | 60 | * avl_left, avl_right, avl_parent, avl_height: pool lock |
1da177e4 LT |
61 | * refcnt: atomically against modifications on other CPU; |
62 | * usually under some other lock to prevent node disappearing | |
582a72da | 63 | * daddr: unchangeable |
317fe0e6 | 64 | * ip_id_count: atomic value (no lock needed) |
1da177e4 LT |
65 | */ |
66 | ||
e18b890b | 67 | static struct kmem_cache *peer_cachep __read_mostly; |
1da177e4 LT |
68 | |
69 | #define node_height(x) x->avl_height | |
d6cc1d64 ED |
70 | |
71 | #define peer_avl_empty ((struct inet_peer *)&peer_fake_node) | |
b914c4ea | 72 | #define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node) |
d6cc1d64 | 73 | static const struct inet_peer peer_fake_node = { |
b914c4ea ED |
74 | .avl_left = peer_avl_empty_rcu, |
75 | .avl_right = peer_avl_empty_rcu, | |
1da177e4 LT |
76 | .avl_height = 0 |
77 | }; | |
d6cc1d64 | 78 | |
021e9299 | 79 | struct inet_peer_base { |
b914c4ea | 80 | struct inet_peer __rcu *root; |
65e8354e | 81 | seqlock_t lock; |
d6cc1d64 | 82 | int total; |
021e9299 DM |
83 | }; |
84 | ||
85 | static struct inet_peer_base v4_peers = { | |
b914c4ea | 86 | .root = peer_avl_empty_rcu, |
65e8354e | 87 | .lock = __SEQLOCK_UNLOCKED(v4_peers.lock), |
d6cc1d64 ED |
88 | .total = 0, |
89 | }; | |
021e9299 DM |
90 | |
91 | static struct inet_peer_base v6_peers = { | |
92 | .root = peer_avl_empty_rcu, | |
65e8354e | 93 | .lock = __SEQLOCK_UNLOCKED(v6_peers.lock), |
021e9299 DM |
94 | .total = 0, |
95 | }; | |
96 | ||
1da177e4 LT |
97 | #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */ |
98 | ||
1da177e4 | 99 | /* Exported for sysctl_net_ipv4. */ |
243bbcaa | 100 | int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more |
1da177e4 | 101 | * aggressively at this stage */ |
243bbcaa ED |
102 | int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */ |
103 | int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */ | |
1da177e4 | 104 | |
1da177e4 LT |
105 | |
106 | /* Called from ip_output.c:ip_init */ | |
107 | void __init inet_initpeers(void) | |
108 | { | |
109 | struct sysinfo si; | |
110 | ||
111 | /* Use the straight interface to information about memory. */ | |
112 | si_meminfo(&si); | |
113 | /* The values below were suggested by Alexey Kuznetsov | |
114 | * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values | |
115 | * myself. --SAW | |
116 | */ | |
117 | if (si.totalram <= (32768*1024)/PAGE_SIZE) | |
118 | inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */ | |
119 | if (si.totalram <= (16384*1024)/PAGE_SIZE) | |
120 | inet_peer_threshold >>= 1; /* about 512KB */ | |
121 | if (si.totalram <= (8192*1024)/PAGE_SIZE) | |
122 | inet_peer_threshold >>= 2; /* about 128KB */ | |
123 | ||
124 | peer_cachep = kmem_cache_create("inet_peer_cache", | |
125 | sizeof(struct inet_peer), | |
317fe0e6 | 126 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, |
20c2df83 | 127 | NULL); |
1da177e4 | 128 | |
1da177e4 LT |
129 | } |
130 | ||
8790ca17 DM |
131 | static int addr_compare(const struct inetpeer_addr *a, |
132 | const struct inetpeer_addr *b) | |
02663045 DM |
133 | { |
134 | int i, n = (a->family == AF_INET ? 1 : 4); | |
135 | ||
136 | for (i = 0; i < n; i++) { | |
7a71ed89 | 137 | if (a->addr.a6[i] == b->addr.a6[i]) |
02663045 | 138 | continue; |
7a71ed89 | 139 | if (a->addr.a6[i] < b->addr.a6[i]) |
02663045 DM |
140 | return -1; |
141 | return 1; | |
142 | } | |
143 | ||
144 | return 0; | |
145 | } | |
146 | ||
65e8354e ED |
147 | #define rcu_deref_locked(X, BASE) \ |
148 | rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock)) | |
149 | ||
243bbcaa ED |
150 | /* |
151 | * Called with local BH disabled and the pool lock held. | |
243bbcaa | 152 | */ |
98158f5a | 153 | #define lookup(_daddr, _stack, _base) \ |
1da177e4 | 154 | ({ \ |
b914c4ea ED |
155 | struct inet_peer *u; \ |
156 | struct inet_peer __rcu **v; \ | |
aa1039e7 ED |
157 | \ |
158 | stackptr = _stack; \ | |
98158f5a | 159 | *stackptr++ = &_base->root; \ |
65e8354e | 160 | for (u = rcu_deref_locked(_base->root, _base); \ |
b914c4ea | 161 | u != peer_avl_empty; ) { \ |
02663045 DM |
162 | int cmp = addr_compare(_daddr, &u->daddr); \ |
163 | if (cmp == 0) \ | |
1da177e4 | 164 | break; \ |
02663045 | 165 | if (cmp == -1) \ |
1da177e4 LT |
166 | v = &u->avl_left; \ |
167 | else \ | |
168 | v = &u->avl_right; \ | |
aa1039e7 | 169 | *stackptr++ = v; \ |
65e8354e | 170 | u = rcu_deref_locked(*v, _base); \ |
1da177e4 LT |
171 | } \ |
172 | u; \ | |
173 | }) | |
174 | ||
aa1039e7 | 175 | /* |
7b46ac4e | 176 | * Called with rcu_read_lock() |
aa1039e7 ED |
177 | * Because we hold no lock against a writer, its quite possible we fall |
178 | * in an endless loop. | |
179 | * But every pointer we follow is guaranteed to be valid thanks to RCU. | |
180 | * We exit from this function if number of links exceeds PEER_MAXDEPTH | |
181 | */ | |
7b46ac4e | 182 | static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr, |
4b9d9be8 | 183 | struct inet_peer_base *base) |
aa1039e7 | 184 | { |
7b46ac4e | 185 | struct inet_peer *u = rcu_dereference(base->root); |
aa1039e7 ED |
186 | int count = 0; |
187 | ||
188 | while (u != peer_avl_empty) { | |
02663045 DM |
189 | int cmp = addr_compare(daddr, &u->daddr); |
190 | if (cmp == 0) { | |
5f2f8920 | 191 | /* Before taking a reference, check if this entry was |
4b9d9be8 | 192 | * deleted (refcnt=-1) |
5f2f8920 | 193 | */ |
4b9d9be8 | 194 | if (!atomic_add_unless(&u->refcnt, 1, -1)) |
aa1039e7 ED |
195 | u = NULL; |
196 | return u; | |
197 | } | |
02663045 | 198 | if (cmp == -1) |
7b46ac4e | 199 | u = rcu_dereference(u->avl_left); |
aa1039e7 | 200 | else |
7b46ac4e | 201 | u = rcu_dereference(u->avl_right); |
aa1039e7 ED |
202 | if (unlikely(++count == PEER_MAXDEPTH)) |
203 | break; | |
204 | } | |
205 | return NULL; | |
206 | } | |
207 | ||
208 | /* Called with local BH disabled and the pool lock held. */ | |
98158f5a | 209 | #define lookup_rightempty(start, base) \ |
1da177e4 | 210 | ({ \ |
b914c4ea ED |
211 | struct inet_peer *u; \ |
212 | struct inet_peer __rcu **v; \ | |
1da177e4 LT |
213 | *stackptr++ = &start->avl_left; \ |
214 | v = &start->avl_left; \ | |
65e8354e | 215 | for (u = rcu_deref_locked(*v, base); \ |
b914c4ea | 216 | u->avl_right != peer_avl_empty_rcu; ) { \ |
1da177e4 LT |
217 | v = &u->avl_right; \ |
218 | *stackptr++ = v; \ | |
65e8354e | 219 | u = rcu_deref_locked(*v, base); \ |
1da177e4 LT |
220 | } \ |
221 | u; \ | |
222 | }) | |
223 | ||
aa1039e7 | 224 | /* Called with local BH disabled and the pool lock held. |
1da177e4 | 225 | * Variable names are the proof of operation correctness. |
aa1039e7 ED |
226 | * Look into mm/map_avl.c for more detail description of the ideas. |
227 | */ | |
b914c4ea | 228 | static void peer_avl_rebalance(struct inet_peer __rcu **stack[], |
98158f5a DM |
229 | struct inet_peer __rcu ***stackend, |
230 | struct inet_peer_base *base) | |
1da177e4 | 231 | { |
b914c4ea ED |
232 | struct inet_peer __rcu **nodep; |
233 | struct inet_peer *node, *l, *r; | |
1da177e4 LT |
234 | int lh, rh; |
235 | ||
236 | while (stackend > stack) { | |
237 | nodep = *--stackend; | |
65e8354e ED |
238 | node = rcu_deref_locked(*nodep, base); |
239 | l = rcu_deref_locked(node->avl_left, base); | |
240 | r = rcu_deref_locked(node->avl_right, base); | |
1da177e4 LT |
241 | lh = node_height(l); |
242 | rh = node_height(r); | |
243 | if (lh > rh + 1) { /* l: RH+2 */ | |
244 | struct inet_peer *ll, *lr, *lrl, *lrr; | |
245 | int lrh; | |
65e8354e ED |
246 | ll = rcu_deref_locked(l->avl_left, base); |
247 | lr = rcu_deref_locked(l->avl_right, base); | |
1da177e4 LT |
248 | lrh = node_height(lr); |
249 | if (lrh <= node_height(ll)) { /* ll: RH+1 */ | |
b914c4ea ED |
250 | RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */ |
251 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 252 | node->avl_height = lrh + 1; /* RH+1 or RH+2 */ |
b914c4ea ED |
253 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */ |
254 | RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */ | |
1da177e4 | 255 | l->avl_height = node->avl_height + 1; |
b914c4ea | 256 | RCU_INIT_POINTER(*nodep, l); |
1da177e4 | 257 | } else { /* ll: RH, lr: RH+1 */ |
65e8354e ED |
258 | lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */ |
259 | lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */ | |
b914c4ea ED |
260 | RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */ |
261 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 262 | node->avl_height = rh + 1; /* node: RH+1 */ |
b914c4ea ED |
263 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */ |
264 | RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */ | |
1da177e4 | 265 | l->avl_height = rh + 1; /* l: RH+1 */ |
b914c4ea ED |
266 | RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */ |
267 | RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */ | |
1da177e4 | 268 | lr->avl_height = rh + 2; |
b914c4ea | 269 | RCU_INIT_POINTER(*nodep, lr); |
1da177e4 LT |
270 | } |
271 | } else if (rh > lh + 1) { /* r: LH+2 */ | |
272 | struct inet_peer *rr, *rl, *rlr, *rll; | |
273 | int rlh; | |
65e8354e ED |
274 | rr = rcu_deref_locked(r->avl_right, base); |
275 | rl = rcu_deref_locked(r->avl_left, base); | |
1da177e4 LT |
276 | rlh = node_height(rl); |
277 | if (rlh <= node_height(rr)) { /* rr: LH+1 */ | |
b914c4ea ED |
278 | RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */ |
279 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 280 | node->avl_height = rlh + 1; /* LH+1 or LH+2 */ |
b914c4ea ED |
281 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */ |
282 | RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */ | |
1da177e4 | 283 | r->avl_height = node->avl_height + 1; |
b914c4ea | 284 | RCU_INIT_POINTER(*nodep, r); |
1da177e4 | 285 | } else { /* rr: RH, rl: RH+1 */ |
65e8354e ED |
286 | rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */ |
287 | rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */ | |
b914c4ea ED |
288 | RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */ |
289 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 290 | node->avl_height = lh + 1; /* node: LH+1 */ |
b914c4ea ED |
291 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */ |
292 | RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */ | |
1da177e4 | 293 | r->avl_height = lh + 1; /* r: LH+1 */ |
b914c4ea ED |
294 | RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */ |
295 | RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */ | |
1da177e4 | 296 | rl->avl_height = lh + 2; |
b914c4ea | 297 | RCU_INIT_POINTER(*nodep, rl); |
1da177e4 LT |
298 | } |
299 | } else { | |
300 | node->avl_height = (lh > rh ? lh : rh) + 1; | |
301 | } | |
302 | } | |
303 | } | |
304 | ||
aa1039e7 | 305 | /* Called with local BH disabled and the pool lock held. */ |
98158f5a | 306 | #define link_to_pool(n, base) \ |
1da177e4 LT |
307 | do { \ |
308 | n->avl_height = 1; \ | |
b914c4ea ED |
309 | n->avl_left = peer_avl_empty_rcu; \ |
310 | n->avl_right = peer_avl_empty_rcu; \ | |
311 | /* lockless readers can catch us now */ \ | |
312 | rcu_assign_pointer(**--stackptr, n); \ | |
98158f5a | 313 | peer_avl_rebalance(stack, stackptr, base); \ |
d6cc1d64 | 314 | } while (0) |
1da177e4 | 315 | |
aa1039e7 ED |
316 | static void inetpeer_free_rcu(struct rcu_head *head) |
317 | { | |
318 | kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu)); | |
319 | } | |
320 | ||
66944e1c ED |
321 | static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base, |
322 | struct inet_peer __rcu **stack[PEER_MAXDEPTH]) | |
1da177e4 | 323 | { |
4b9d9be8 ED |
324 | struct inet_peer __rcu ***stackptr, ***delp; |
325 | ||
326 | if (lookup(&p->daddr, stack, base) != p) | |
327 | BUG(); | |
328 | delp = stackptr - 1; /* *delp[0] == p */ | |
329 | if (p->avl_left == peer_avl_empty_rcu) { | |
330 | *delp[0] = p->avl_right; | |
331 | --stackptr; | |
332 | } else { | |
333 | /* look for a node to insert instead of p */ | |
334 | struct inet_peer *t; | |
335 | t = lookup_rightempty(p, base); | |
336 | BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t); | |
337 | **--stackptr = t->avl_left; | |
338 | /* t is removed, t->daddr > x->daddr for any | |
339 | * x in p->avl_left subtree. | |
340 | * Put t in the old place of p. */ | |
341 | RCU_INIT_POINTER(*delp[0], t); | |
342 | t->avl_left = p->avl_left; | |
343 | t->avl_right = p->avl_right; | |
344 | t->avl_height = p->avl_height; | |
345 | BUG_ON(delp[1] != &p->avl_left); | |
346 | delp[1] = &t->avl_left; /* was &p->avl_left */ | |
1da177e4 | 347 | } |
4b9d9be8 ED |
348 | peer_avl_rebalance(stack, stackptr, base); |
349 | base->total--; | |
350 | call_rcu(&p->rcu, inetpeer_free_rcu); | |
1da177e4 LT |
351 | } |
352 | ||
021e9299 DM |
353 | static struct inet_peer_base *family_to_base(int family) |
354 | { | |
4b9d9be8 | 355 | return family == AF_INET ? &v4_peers : &v6_peers; |
021e9299 DM |
356 | } |
357 | ||
4b9d9be8 ED |
358 | /* perform garbage collect on all items stacked during a lookup */ |
359 | static int inet_peer_gc(struct inet_peer_base *base, | |
360 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], | |
361 | struct inet_peer __rcu ***stackptr) | |
98158f5a | 362 | { |
4b9d9be8 ED |
363 | struct inet_peer *p, *gchead = NULL; |
364 | __u32 delta, ttl; | |
365 | int cnt = 0; | |
d71209de | 366 | |
4b9d9be8 ED |
367 | if (base->total >= inet_peer_threshold) |
368 | ttl = 0; /* be aggressive */ | |
369 | else | |
370 | ttl = inet_peer_maxttl | |
371 | - (inet_peer_maxttl - inet_peer_minttl) / HZ * | |
372 | base->total / inet_peer_threshold * HZ; | |
373 | stackptr--; /* last stack slot is peer_avl_empty */ | |
374 | while (stackptr > stack) { | |
375 | stackptr--; | |
376 | p = rcu_deref_locked(**stackptr, base); | |
6d1a3e04 ED |
377 | if (atomic_read(&p->refcnt) == 0) { |
378 | smp_rmb(); | |
379 | delta = (__u32)jiffies - p->dtime; | |
380 | if (delta >= ttl && | |
381 | atomic_cmpxchg(&p->refcnt, 0, -1) == 0) { | |
382 | p->gc_next = gchead; | |
383 | gchead = p; | |
384 | } | |
1da177e4 | 385 | } |
1da177e4 | 386 | } |
4b9d9be8 ED |
387 | while ((p = gchead) != NULL) { |
388 | gchead = p->gc_next; | |
389 | cnt++; | |
390 | unlink_from_pool(p, base, stack); | |
391 | } | |
392 | return cnt; | |
1da177e4 LT |
393 | } |
394 | ||
87c48fa3 | 395 | struct inet_peer *inet_getpeer(const struct inetpeer_addr *daddr, int create) |
1da177e4 | 396 | { |
b914c4ea | 397 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr; |
3408404a | 398 | struct inet_peer_base *base = family_to_base(daddr->family); |
98158f5a | 399 | struct inet_peer *p; |
65e8354e | 400 | unsigned int sequence; |
4b9d9be8 | 401 | int invalidated, gccnt = 0; |
1da177e4 | 402 | |
4b9d9be8 | 403 | /* Attempt a lockless lookup first. |
aa1039e7 ED |
404 | * Because of a concurrent writer, we might not find an existing entry. |
405 | */ | |
7b46ac4e | 406 | rcu_read_lock(); |
65e8354e | 407 | sequence = read_seqbegin(&base->lock); |
4b9d9be8 | 408 | p = lookup_rcu(daddr, base); |
65e8354e | 409 | invalidated = read_seqretry(&base->lock, sequence); |
7b46ac4e | 410 | rcu_read_unlock(); |
aa1039e7 | 411 | |
4b9d9be8 | 412 | if (p) |
aa1039e7 | 413 | return p; |
1da177e4 | 414 | |
65e8354e ED |
415 | /* If no writer did a change during our lookup, we can return early. */ |
416 | if (!create && !invalidated) | |
417 | return NULL; | |
418 | ||
aa1039e7 ED |
419 | /* retry an exact lookup, taking the lock before. |
420 | * At least, nodes should be hot in our cache. | |
421 | */ | |
65e8354e | 422 | write_seqlock_bh(&base->lock); |
4b9d9be8 | 423 | relookup: |
02663045 | 424 | p = lookup(daddr, stack, base); |
1da177e4 | 425 | if (p != peer_avl_empty) { |
4b9d9be8 | 426 | atomic_inc(&p->refcnt); |
65e8354e | 427 | write_sequnlock_bh(&base->lock); |
4b9d9be8 ED |
428 | return p; |
429 | } | |
430 | if (!gccnt) { | |
431 | gccnt = inet_peer_gc(base, stack, stackptr); | |
432 | if (gccnt && create) | |
433 | goto relookup; | |
1da177e4 | 434 | } |
aa1039e7 ED |
435 | p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL; |
436 | if (p) { | |
b534ecf1 | 437 | p->daddr = *daddr; |
aa1039e7 ED |
438 | atomic_set(&p->refcnt, 1); |
439 | atomic_set(&p->rid, 0); | |
87c48fa3 ED |
440 | atomic_set(&p->ip_id_count, |
441 | (daddr->family == AF_INET) ? | |
442 | secure_ip_id(daddr->addr.a4) : | |
443 | secure_ipv6_id(daddr->addr.a6)); | |
aa1039e7 | 444 | p->tcp_ts_stamp = 0; |
144001bd | 445 | p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW; |
92d86829 DM |
446 | p->rate_tokens = 0; |
447 | p->rate_last = 0; | |
ddd4aa42 | 448 | p->pmtu_expires = 0; |
46af3180 | 449 | p->pmtu_orig = 0; |
ddd4aa42 | 450 | memset(&p->redirect_learned, 0, sizeof(p->redirect_learned)); |
aa1039e7 ED |
451 | |
452 | ||
453 | /* Link the node. */ | |
98158f5a DM |
454 | link_to_pool(p, base); |
455 | base->total++; | |
aa1039e7 | 456 | } |
65e8354e | 457 | write_sequnlock_bh(&base->lock); |
1da177e4 | 458 | |
1da177e4 LT |
459 | return p; |
460 | } | |
b3419363 | 461 | EXPORT_SYMBOL_GPL(inet_getpeer); |
98158f5a | 462 | |
4663afe2 ED |
463 | void inet_putpeer(struct inet_peer *p) |
464 | { | |
4b9d9be8 | 465 | p->dtime = (__u32)jiffies; |
6d1a3e04 | 466 | smp_mb__before_atomic_dec(); |
4b9d9be8 | 467 | atomic_dec(&p->refcnt); |
4663afe2 | 468 | } |
b3419363 | 469 | EXPORT_SYMBOL_GPL(inet_putpeer); |
92d86829 DM |
470 | |
471 | /* | |
472 | * Check transmit rate limitation for given message. | |
473 | * The rate information is held in the inet_peer entries now. | |
474 | * This function is generic and could be used for other purposes | |
475 | * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. | |
476 | * | |
477 | * Note that the same inet_peer fields are modified by functions in | |
478 | * route.c too, but these work for packet destinations while xrlim_allow | |
479 | * works for icmp destinations. This means the rate limiting information | |
480 | * for one "ip object" is shared - and these ICMPs are twice limited: | |
481 | * by source and by destination. | |
482 | * | |
483 | * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate | |
484 | * SHOULD allow setting of rate limits | |
485 | * | |
486 | * Shared between ICMPv4 and ICMPv6. | |
487 | */ | |
488 | #define XRLIM_BURST_FACTOR 6 | |
489 | bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout) | |
490 | { | |
491 | unsigned long now, token; | |
492 | bool rc = false; | |
493 | ||
494 | if (!peer) | |
495 | return true; | |
496 | ||
497 | token = peer->rate_tokens; | |
498 | now = jiffies; | |
499 | token += now - peer->rate_last; | |
500 | peer->rate_last = now; | |
501 | if (token > XRLIM_BURST_FACTOR * timeout) | |
502 | token = XRLIM_BURST_FACTOR * timeout; | |
503 | if (token >= timeout) { | |
504 | token -= timeout; | |
505 | rc = true; | |
506 | } | |
507 | peer->rate_tokens = token; | |
508 | return rc; | |
509 | } | |
510 | EXPORT_SYMBOL(inet_peer_xrlim_allow); |