]>
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> | |
5faa5df1 | 20 | #include <linux/workqueue.h> |
20380731 | 21 | #include <net/ip.h> |
1da177e4 | 22 | #include <net/inetpeer.h> |
6e5714ea | 23 | #include <net/secure_seq.h> |
1da177e4 LT |
24 | |
25 | /* | |
26 | * Theory of operations. | |
27 | * We keep one entry for each peer IP address. The nodes contains long-living | |
28 | * information about the peer which doesn't depend on routes. | |
29 | * At this moment this information consists only of ID field for the next | |
30 | * outgoing IP packet. This field is incremented with each packet as encoded | |
31 | * in inet_getid() function (include/net/inetpeer.h). | |
32 | * At the moment of writing this notes identifier of IP packets is generated | |
33 | * to be unpredictable using this code only for packets subjected | |
34 | * (actually or potentially) to defragmentation. I.e. DF packets less than | |
35 | * PMTU in size uses a constant ID and do not use this code (see | |
36 | * ip_select_ident() in include/net/ip.h). | |
37 | * | |
38 | * Route cache entries hold references to our nodes. | |
39 | * New cache entries get references via lookup by destination IP address in | |
40 | * the avl tree. The reference is grabbed only when it's needed i.e. only | |
41 | * when we try to output IP packet which needs an unpredictable ID (see | |
42 | * __ip_select_ident() in net/ipv4/route.c). | |
43 | * Nodes are removed only when reference counter goes to 0. | |
44 | * When it's happened the node may be removed when a sufficient amount of | |
45 | * time has been passed since its last use. The less-recently-used entry can | |
46 | * also be removed if the pool is overloaded i.e. if the total amount of | |
47 | * entries is greater-or-equal than the threshold. | |
48 | * | |
49 | * Node pool is organised as an AVL tree. | |
50 | * Such an implementation has been chosen not just for fun. It's a way to | |
51 | * prevent easy and efficient DoS attacks by creating hash collisions. A huge | |
52 | * amount of long living nodes in a single hash slot would significantly delay | |
53 | * lookups performed with disabled BHs. | |
54 | * | |
55 | * Serialisation issues. | |
aa1039e7 ED |
56 | * 1. Nodes may appear in the tree only with the pool lock held. |
57 | * 2. Nodes may disappear from the tree only with the pool lock held | |
1da177e4 | 58 | * AND reference count being 0. |
4b9d9be8 ED |
59 | * 3. Global variable peer_total is modified under the pool lock. |
60 | * 4. struct inet_peer fields modification: | |
1da177e4 | 61 | * avl_left, avl_right, avl_parent, avl_height: pool lock |
1da177e4 LT |
62 | * refcnt: atomically against modifications on other CPU; |
63 | * usually under some other lock to prevent node disappearing | |
582a72da | 64 | * daddr: unchangeable |
317fe0e6 | 65 | * ip_id_count: atomic value (no lock needed) |
1da177e4 LT |
66 | */ |
67 | ||
e18b890b | 68 | static struct kmem_cache *peer_cachep __read_mostly; |
1da177e4 | 69 | |
5faa5df1 SK |
70 | static LIST_HEAD(gc_list); |
71 | static const int gc_delay = 60 * HZ; | |
72 | static struct delayed_work gc_work; | |
73 | static DEFINE_SPINLOCK(gc_lock); | |
74 | ||
1da177e4 | 75 | #define node_height(x) x->avl_height |
d6cc1d64 ED |
76 | |
77 | #define peer_avl_empty ((struct inet_peer *)&peer_fake_node) | |
b914c4ea | 78 | #define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node) |
d6cc1d64 | 79 | static const struct inet_peer peer_fake_node = { |
b914c4ea ED |
80 | .avl_left = peer_avl_empty_rcu, |
81 | .avl_right = peer_avl_empty_rcu, | |
1da177e4 LT |
82 | .avl_height = 0 |
83 | }; | |
d6cc1d64 | 84 | |
c3426b47 DM |
85 | void inet_peer_base_init(struct inet_peer_base *bp) |
86 | { | |
87 | bp->root = peer_avl_empty_rcu; | |
88 | seqlock_init(&bp->lock); | |
89 | bp->total = 0; | |
90 | } | |
91 | EXPORT_SYMBOL_GPL(inet_peer_base_init); | |
021e9299 | 92 | |
1da177e4 LT |
93 | #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */ |
94 | ||
1da177e4 | 95 | /* Exported for sysctl_net_ipv4. */ |
243bbcaa | 96 | int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more |
1da177e4 | 97 | * aggressively at this stage */ |
243bbcaa ED |
98 | int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */ |
99 | int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */ | |
1da177e4 | 100 | |
5faa5df1 SK |
101 | static void inetpeer_gc_worker(struct work_struct *work) |
102 | { | |
103 | struct inet_peer *p, *n; | |
104 | LIST_HEAD(list); | |
105 | ||
106 | spin_lock_bh(&gc_lock); | |
107 | list_replace_init(&gc_list, &list); | |
108 | spin_unlock_bh(&gc_lock); | |
109 | ||
110 | if (list_empty(&list)) | |
111 | return; | |
112 | ||
113 | list_for_each_entry_safe(p, n, &list, gc_list) { | |
114 | ||
115 | if(need_resched()) | |
116 | cond_resched(); | |
117 | ||
118 | if (p->avl_left != peer_avl_empty) { | |
119 | list_add_tail(&p->avl_left->gc_list, &list); | |
120 | p->avl_left = peer_avl_empty; | |
121 | } | |
122 | ||
123 | if (p->avl_right != peer_avl_empty) { | |
124 | list_add_tail(&p->avl_right->gc_list, &list); | |
125 | p->avl_right = peer_avl_empty; | |
126 | } | |
127 | ||
128 | n = list_entry(p->gc_list.next, struct inet_peer, gc_list); | |
129 | ||
130 | if (!atomic_read(&p->refcnt)) { | |
131 | list_del(&p->gc_list); | |
132 | kmem_cache_free(peer_cachep, p); | |
133 | } | |
134 | } | |
135 | ||
136 | if (list_empty(&list)) | |
137 | return; | |
138 | ||
139 | spin_lock_bh(&gc_lock); | |
140 | list_splice(&list, &gc_list); | |
141 | spin_unlock_bh(&gc_lock); | |
142 | ||
143 | schedule_delayed_work(&gc_work, gc_delay); | |
144 | } | |
1da177e4 LT |
145 | |
146 | /* Called from ip_output.c:ip_init */ | |
147 | void __init inet_initpeers(void) | |
148 | { | |
149 | struct sysinfo si; | |
150 | ||
151 | /* Use the straight interface to information about memory. */ | |
152 | si_meminfo(&si); | |
153 | /* The values below were suggested by Alexey Kuznetsov | |
154 | * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values | |
155 | * myself. --SAW | |
156 | */ | |
157 | if (si.totalram <= (32768*1024)/PAGE_SIZE) | |
158 | inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */ | |
159 | if (si.totalram <= (16384*1024)/PAGE_SIZE) | |
160 | inet_peer_threshold >>= 1; /* about 512KB */ | |
161 | if (si.totalram <= (8192*1024)/PAGE_SIZE) | |
162 | inet_peer_threshold >>= 2; /* about 128KB */ | |
163 | ||
164 | peer_cachep = kmem_cache_create("inet_peer_cache", | |
165 | sizeof(struct inet_peer), | |
317fe0e6 | 166 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, |
20c2df83 | 167 | NULL); |
1da177e4 | 168 | |
5faa5df1 | 169 | INIT_DELAYED_WORK_DEFERRABLE(&gc_work, inetpeer_gc_worker); |
1da177e4 LT |
170 | } |
171 | ||
8790ca17 DM |
172 | static int addr_compare(const struct inetpeer_addr *a, |
173 | const struct inetpeer_addr *b) | |
02663045 DM |
174 | { |
175 | int i, n = (a->family == AF_INET ? 1 : 4); | |
176 | ||
177 | for (i = 0; i < n; i++) { | |
7a71ed89 | 178 | if (a->addr.a6[i] == b->addr.a6[i]) |
02663045 | 179 | continue; |
747465ef | 180 | if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i]) |
02663045 DM |
181 | return -1; |
182 | return 1; | |
183 | } | |
184 | ||
185 | return 0; | |
186 | } | |
187 | ||
65e8354e ED |
188 | #define rcu_deref_locked(X, BASE) \ |
189 | rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock)) | |
190 | ||
243bbcaa ED |
191 | /* |
192 | * Called with local BH disabled and the pool lock held. | |
243bbcaa | 193 | */ |
98158f5a | 194 | #define lookup(_daddr, _stack, _base) \ |
1da177e4 | 195 | ({ \ |
b914c4ea ED |
196 | struct inet_peer *u; \ |
197 | struct inet_peer __rcu **v; \ | |
aa1039e7 ED |
198 | \ |
199 | stackptr = _stack; \ | |
98158f5a | 200 | *stackptr++ = &_base->root; \ |
65e8354e | 201 | for (u = rcu_deref_locked(_base->root, _base); \ |
b914c4ea | 202 | u != peer_avl_empty; ) { \ |
02663045 DM |
203 | int cmp = addr_compare(_daddr, &u->daddr); \ |
204 | if (cmp == 0) \ | |
1da177e4 | 205 | break; \ |
02663045 | 206 | if (cmp == -1) \ |
1da177e4 LT |
207 | v = &u->avl_left; \ |
208 | else \ | |
209 | v = &u->avl_right; \ | |
aa1039e7 | 210 | *stackptr++ = v; \ |
65e8354e | 211 | u = rcu_deref_locked(*v, _base); \ |
1da177e4 LT |
212 | } \ |
213 | u; \ | |
214 | }) | |
215 | ||
aa1039e7 | 216 | /* |
7b46ac4e | 217 | * Called with rcu_read_lock() |
aa1039e7 ED |
218 | * Because we hold no lock against a writer, its quite possible we fall |
219 | * in an endless loop. | |
220 | * But every pointer we follow is guaranteed to be valid thanks to RCU. | |
221 | * We exit from this function if number of links exceeds PEER_MAXDEPTH | |
222 | */ | |
7b46ac4e | 223 | static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr, |
4b9d9be8 | 224 | struct inet_peer_base *base) |
aa1039e7 | 225 | { |
7b46ac4e | 226 | struct inet_peer *u = rcu_dereference(base->root); |
aa1039e7 ED |
227 | int count = 0; |
228 | ||
229 | while (u != peer_avl_empty) { | |
02663045 DM |
230 | int cmp = addr_compare(daddr, &u->daddr); |
231 | if (cmp == 0) { | |
5f2f8920 | 232 | /* Before taking a reference, check if this entry was |
4b9d9be8 | 233 | * deleted (refcnt=-1) |
5f2f8920 | 234 | */ |
4b9d9be8 | 235 | if (!atomic_add_unless(&u->refcnt, 1, -1)) |
aa1039e7 ED |
236 | u = NULL; |
237 | return u; | |
238 | } | |
02663045 | 239 | if (cmp == -1) |
7b46ac4e | 240 | u = rcu_dereference(u->avl_left); |
aa1039e7 | 241 | else |
7b46ac4e | 242 | u = rcu_dereference(u->avl_right); |
aa1039e7 ED |
243 | if (unlikely(++count == PEER_MAXDEPTH)) |
244 | break; | |
245 | } | |
246 | return NULL; | |
247 | } | |
248 | ||
249 | /* Called with local BH disabled and the pool lock held. */ | |
98158f5a | 250 | #define lookup_rightempty(start, base) \ |
1da177e4 | 251 | ({ \ |
b914c4ea ED |
252 | struct inet_peer *u; \ |
253 | struct inet_peer __rcu **v; \ | |
1da177e4 LT |
254 | *stackptr++ = &start->avl_left; \ |
255 | v = &start->avl_left; \ | |
65e8354e | 256 | for (u = rcu_deref_locked(*v, base); \ |
b914c4ea | 257 | u->avl_right != peer_avl_empty_rcu; ) { \ |
1da177e4 LT |
258 | v = &u->avl_right; \ |
259 | *stackptr++ = v; \ | |
65e8354e | 260 | u = rcu_deref_locked(*v, base); \ |
1da177e4 LT |
261 | } \ |
262 | u; \ | |
263 | }) | |
264 | ||
aa1039e7 | 265 | /* Called with local BH disabled and the pool lock held. |
1da177e4 | 266 | * Variable names are the proof of operation correctness. |
aa1039e7 ED |
267 | * Look into mm/map_avl.c for more detail description of the ideas. |
268 | */ | |
b914c4ea | 269 | static void peer_avl_rebalance(struct inet_peer __rcu **stack[], |
98158f5a DM |
270 | struct inet_peer __rcu ***stackend, |
271 | struct inet_peer_base *base) | |
1da177e4 | 272 | { |
b914c4ea ED |
273 | struct inet_peer __rcu **nodep; |
274 | struct inet_peer *node, *l, *r; | |
1da177e4 LT |
275 | int lh, rh; |
276 | ||
277 | while (stackend > stack) { | |
278 | nodep = *--stackend; | |
65e8354e ED |
279 | node = rcu_deref_locked(*nodep, base); |
280 | l = rcu_deref_locked(node->avl_left, base); | |
281 | r = rcu_deref_locked(node->avl_right, base); | |
1da177e4 LT |
282 | lh = node_height(l); |
283 | rh = node_height(r); | |
284 | if (lh > rh + 1) { /* l: RH+2 */ | |
285 | struct inet_peer *ll, *lr, *lrl, *lrr; | |
286 | int lrh; | |
65e8354e ED |
287 | ll = rcu_deref_locked(l->avl_left, base); |
288 | lr = rcu_deref_locked(l->avl_right, base); | |
1da177e4 LT |
289 | lrh = node_height(lr); |
290 | if (lrh <= node_height(ll)) { /* ll: RH+1 */ | |
b914c4ea ED |
291 | RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */ |
292 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 293 | node->avl_height = lrh + 1; /* RH+1 or RH+2 */ |
b914c4ea ED |
294 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */ |
295 | RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */ | |
1da177e4 | 296 | l->avl_height = node->avl_height + 1; |
b914c4ea | 297 | RCU_INIT_POINTER(*nodep, l); |
1da177e4 | 298 | } else { /* ll: RH, lr: RH+1 */ |
65e8354e ED |
299 | lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */ |
300 | lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */ | |
b914c4ea ED |
301 | RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */ |
302 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 303 | node->avl_height = rh + 1; /* node: RH+1 */ |
b914c4ea ED |
304 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */ |
305 | RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */ | |
1da177e4 | 306 | l->avl_height = rh + 1; /* l: RH+1 */ |
b914c4ea ED |
307 | RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */ |
308 | RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */ | |
1da177e4 | 309 | lr->avl_height = rh + 2; |
b914c4ea | 310 | RCU_INIT_POINTER(*nodep, lr); |
1da177e4 LT |
311 | } |
312 | } else if (rh > lh + 1) { /* r: LH+2 */ | |
313 | struct inet_peer *rr, *rl, *rlr, *rll; | |
314 | int rlh; | |
65e8354e ED |
315 | rr = rcu_deref_locked(r->avl_right, base); |
316 | rl = rcu_deref_locked(r->avl_left, base); | |
1da177e4 LT |
317 | rlh = node_height(rl); |
318 | if (rlh <= node_height(rr)) { /* rr: LH+1 */ | |
b914c4ea ED |
319 | RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */ |
320 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 321 | node->avl_height = rlh + 1; /* LH+1 or LH+2 */ |
b914c4ea ED |
322 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */ |
323 | RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */ | |
1da177e4 | 324 | r->avl_height = node->avl_height + 1; |
b914c4ea | 325 | RCU_INIT_POINTER(*nodep, r); |
1da177e4 | 326 | } else { /* rr: RH, rl: RH+1 */ |
65e8354e ED |
327 | rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */ |
328 | rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */ | |
b914c4ea ED |
329 | RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */ |
330 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 331 | node->avl_height = lh + 1; /* node: LH+1 */ |
b914c4ea ED |
332 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */ |
333 | RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */ | |
1da177e4 | 334 | r->avl_height = lh + 1; /* r: LH+1 */ |
b914c4ea ED |
335 | RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */ |
336 | RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */ | |
1da177e4 | 337 | rl->avl_height = lh + 2; |
b914c4ea | 338 | RCU_INIT_POINTER(*nodep, rl); |
1da177e4 LT |
339 | } |
340 | } else { | |
341 | node->avl_height = (lh > rh ? lh : rh) + 1; | |
342 | } | |
343 | } | |
344 | } | |
345 | ||
aa1039e7 | 346 | /* Called with local BH disabled and the pool lock held. */ |
98158f5a | 347 | #define link_to_pool(n, base) \ |
1da177e4 LT |
348 | do { \ |
349 | n->avl_height = 1; \ | |
b914c4ea ED |
350 | n->avl_left = peer_avl_empty_rcu; \ |
351 | n->avl_right = peer_avl_empty_rcu; \ | |
352 | /* lockless readers can catch us now */ \ | |
353 | rcu_assign_pointer(**--stackptr, n); \ | |
98158f5a | 354 | peer_avl_rebalance(stack, stackptr, base); \ |
d6cc1d64 | 355 | } while (0) |
1da177e4 | 356 | |
aa1039e7 ED |
357 | static void inetpeer_free_rcu(struct rcu_head *head) |
358 | { | |
359 | kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu)); | |
360 | } | |
361 | ||
66944e1c ED |
362 | static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base, |
363 | struct inet_peer __rcu **stack[PEER_MAXDEPTH]) | |
1da177e4 | 364 | { |
4b9d9be8 ED |
365 | struct inet_peer __rcu ***stackptr, ***delp; |
366 | ||
367 | if (lookup(&p->daddr, stack, base) != p) | |
368 | BUG(); | |
369 | delp = stackptr - 1; /* *delp[0] == p */ | |
370 | if (p->avl_left == peer_avl_empty_rcu) { | |
371 | *delp[0] = p->avl_right; | |
372 | --stackptr; | |
373 | } else { | |
374 | /* look for a node to insert instead of p */ | |
375 | struct inet_peer *t; | |
376 | t = lookup_rightempty(p, base); | |
377 | BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t); | |
378 | **--stackptr = t->avl_left; | |
379 | /* t is removed, t->daddr > x->daddr for any | |
380 | * x in p->avl_left subtree. | |
381 | * Put t in the old place of p. */ | |
382 | RCU_INIT_POINTER(*delp[0], t); | |
383 | t->avl_left = p->avl_left; | |
384 | t->avl_right = p->avl_right; | |
385 | t->avl_height = p->avl_height; | |
386 | BUG_ON(delp[1] != &p->avl_left); | |
387 | delp[1] = &t->avl_left; /* was &p->avl_left */ | |
1da177e4 | 388 | } |
4b9d9be8 ED |
389 | peer_avl_rebalance(stack, stackptr, base); |
390 | base->total--; | |
391 | call_rcu(&p->rcu, inetpeer_free_rcu); | |
1da177e4 LT |
392 | } |
393 | ||
c8a627ed G |
394 | static struct inet_peer_base *family_to_base(struct net *net, |
395 | int family) | |
021e9299 | 396 | { |
c8a627ed | 397 | return family == AF_INET ? net->ipv4.peers : net->ipv6.peers; |
021e9299 DM |
398 | } |
399 | ||
4b9d9be8 ED |
400 | /* perform garbage collect on all items stacked during a lookup */ |
401 | static int inet_peer_gc(struct inet_peer_base *base, | |
402 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], | |
403 | struct inet_peer __rcu ***stackptr) | |
98158f5a | 404 | { |
4b9d9be8 ED |
405 | struct inet_peer *p, *gchead = NULL; |
406 | __u32 delta, ttl; | |
407 | int cnt = 0; | |
d71209de | 408 | |
4b9d9be8 ED |
409 | if (base->total >= inet_peer_threshold) |
410 | ttl = 0; /* be aggressive */ | |
411 | else | |
412 | ttl = inet_peer_maxttl | |
413 | - (inet_peer_maxttl - inet_peer_minttl) / HZ * | |
414 | base->total / inet_peer_threshold * HZ; | |
415 | stackptr--; /* last stack slot is peer_avl_empty */ | |
416 | while (stackptr > stack) { | |
417 | stackptr--; | |
418 | p = rcu_deref_locked(**stackptr, base); | |
6d1a3e04 ED |
419 | if (atomic_read(&p->refcnt) == 0) { |
420 | smp_rmb(); | |
421 | delta = (__u32)jiffies - p->dtime; | |
422 | if (delta >= ttl && | |
423 | atomic_cmpxchg(&p->refcnt, 0, -1) == 0) { | |
424 | p->gc_next = gchead; | |
425 | gchead = p; | |
426 | } | |
1da177e4 | 427 | } |
1da177e4 | 428 | } |
4b9d9be8 ED |
429 | while ((p = gchead) != NULL) { |
430 | gchead = p->gc_next; | |
431 | cnt++; | |
432 | unlink_from_pool(p, base, stack); | |
433 | } | |
434 | return cnt; | |
1da177e4 LT |
435 | } |
436 | ||
c8a627ed G |
437 | struct inet_peer *inet_getpeer(struct net *net, |
438 | const struct inetpeer_addr *daddr, | |
439 | int create) | |
1da177e4 | 440 | { |
b914c4ea | 441 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr; |
c8a627ed | 442 | struct inet_peer_base *base = family_to_base(net, daddr->family); |
98158f5a | 443 | struct inet_peer *p; |
65e8354e | 444 | unsigned int sequence; |
4b9d9be8 | 445 | int invalidated, gccnt = 0; |
1da177e4 | 446 | |
4b9d9be8 | 447 | /* Attempt a lockless lookup first. |
aa1039e7 ED |
448 | * Because of a concurrent writer, we might not find an existing entry. |
449 | */ | |
7b46ac4e | 450 | rcu_read_lock(); |
65e8354e | 451 | sequence = read_seqbegin(&base->lock); |
4b9d9be8 | 452 | p = lookup_rcu(daddr, base); |
65e8354e | 453 | invalidated = read_seqretry(&base->lock, sequence); |
7b46ac4e | 454 | rcu_read_unlock(); |
aa1039e7 | 455 | |
4b9d9be8 | 456 | if (p) |
aa1039e7 | 457 | return p; |
1da177e4 | 458 | |
65e8354e ED |
459 | /* If no writer did a change during our lookup, we can return early. */ |
460 | if (!create && !invalidated) | |
461 | return NULL; | |
462 | ||
aa1039e7 ED |
463 | /* retry an exact lookup, taking the lock before. |
464 | * At least, nodes should be hot in our cache. | |
465 | */ | |
65e8354e | 466 | write_seqlock_bh(&base->lock); |
4b9d9be8 | 467 | relookup: |
02663045 | 468 | p = lookup(daddr, stack, base); |
1da177e4 | 469 | if (p != peer_avl_empty) { |
4b9d9be8 | 470 | atomic_inc(&p->refcnt); |
65e8354e | 471 | write_sequnlock_bh(&base->lock); |
4b9d9be8 ED |
472 | return p; |
473 | } | |
474 | if (!gccnt) { | |
475 | gccnt = inet_peer_gc(base, stack, stackptr); | |
476 | if (gccnt && create) | |
477 | goto relookup; | |
1da177e4 | 478 | } |
aa1039e7 ED |
479 | p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL; |
480 | if (p) { | |
b534ecf1 | 481 | p->daddr = *daddr; |
aa1039e7 ED |
482 | atomic_set(&p->refcnt, 1); |
483 | atomic_set(&p->rid, 0); | |
87c48fa3 ED |
484 | atomic_set(&p->ip_id_count, |
485 | (daddr->family == AF_INET) ? | |
486 | secure_ip_id(daddr->addr.a4) : | |
487 | secure_ipv6_id(daddr->addr.a6)); | |
aa1039e7 | 488 | p->tcp_ts_stamp = 0; |
144001bd | 489 | p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW; |
92d86829 DM |
490 | p->rate_tokens = 0; |
491 | p->rate_last = 0; | |
ddd4aa42 | 492 | p->pmtu_expires = 0; |
46af3180 | 493 | p->pmtu_orig = 0; |
ddd4aa42 | 494 | memset(&p->redirect_learned, 0, sizeof(p->redirect_learned)); |
5faa5df1 | 495 | INIT_LIST_HEAD(&p->gc_list); |
aa1039e7 ED |
496 | |
497 | /* Link the node. */ | |
98158f5a DM |
498 | link_to_pool(p, base); |
499 | base->total++; | |
aa1039e7 | 500 | } |
65e8354e | 501 | write_sequnlock_bh(&base->lock); |
1da177e4 | 502 | |
1da177e4 LT |
503 | return p; |
504 | } | |
b3419363 | 505 | EXPORT_SYMBOL_GPL(inet_getpeer); |
98158f5a | 506 | |
4663afe2 ED |
507 | void inet_putpeer(struct inet_peer *p) |
508 | { | |
4b9d9be8 | 509 | p->dtime = (__u32)jiffies; |
6d1a3e04 | 510 | smp_mb__before_atomic_dec(); |
4b9d9be8 | 511 | atomic_dec(&p->refcnt); |
4663afe2 | 512 | } |
b3419363 | 513 | EXPORT_SYMBOL_GPL(inet_putpeer); |
92d86829 DM |
514 | |
515 | /* | |
516 | * Check transmit rate limitation for given message. | |
517 | * The rate information is held in the inet_peer entries now. | |
518 | * This function is generic and could be used for other purposes | |
519 | * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. | |
520 | * | |
521 | * Note that the same inet_peer fields are modified by functions in | |
522 | * route.c too, but these work for packet destinations while xrlim_allow | |
523 | * works for icmp destinations. This means the rate limiting information | |
524 | * for one "ip object" is shared - and these ICMPs are twice limited: | |
525 | * by source and by destination. | |
526 | * | |
527 | * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate | |
528 | * SHOULD allow setting of rate limits | |
529 | * | |
530 | * Shared between ICMPv4 and ICMPv6. | |
531 | */ | |
532 | #define XRLIM_BURST_FACTOR 6 | |
533 | bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout) | |
534 | { | |
535 | unsigned long now, token; | |
536 | bool rc = false; | |
537 | ||
538 | if (!peer) | |
539 | return true; | |
540 | ||
541 | token = peer->rate_tokens; | |
542 | now = jiffies; | |
543 | token += now - peer->rate_last; | |
544 | peer->rate_last = now; | |
545 | if (token > XRLIM_BURST_FACTOR * timeout) | |
546 | token = XRLIM_BURST_FACTOR * timeout; | |
547 | if (token >= timeout) { | |
548 | token -= timeout; | |
549 | rc = true; | |
550 | } | |
551 | peer->rate_tokens = token; | |
552 | return rc; | |
553 | } | |
554 | EXPORT_SYMBOL(inet_peer_xrlim_allow); | |
5faa5df1 | 555 | |
55432d2b ED |
556 | static void inetpeer_inval_rcu(struct rcu_head *head) |
557 | { | |
558 | struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu); | |
559 | ||
560 | spin_lock_bh(&gc_lock); | |
561 | list_add_tail(&p->gc_list, &gc_list); | |
562 | spin_unlock_bh(&gc_lock); | |
563 | ||
564 | schedule_delayed_work(&gc_work, gc_delay); | |
565 | } | |
566 | ||
56a6b248 | 567 | void inetpeer_invalidate_tree(struct inet_peer_base *base) |
5faa5df1 SK |
568 | { |
569 | struct inet_peer *old, *new, *prev; | |
5faa5df1 SK |
570 | |
571 | write_seqlock_bh(&base->lock); | |
572 | ||
573 | old = base->root; | |
574 | if (old == peer_avl_empty_rcu) | |
575 | goto out; | |
576 | ||
577 | new = peer_avl_empty_rcu; | |
578 | ||
579 | prev = cmpxchg(&base->root, old, new); | |
580 | if (prev == old) { | |
581 | base->total = 0; | |
55432d2b | 582 | call_rcu(&prev->gc_rcu, inetpeer_inval_rcu); |
5faa5df1 SK |
583 | } |
584 | ||
585 | out: | |
586 | write_sequnlock_bh(&base->lock); | |
587 | } | |
588 | EXPORT_SYMBOL(inetpeer_invalidate_tree); |