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); | |
b48c80ec | 89 | bp->flush_seq = ~0U; |
c3426b47 DM |
90 | bp->total = 0; |
91 | } | |
92 | EXPORT_SYMBOL_GPL(inet_peer_base_init); | |
021e9299 | 93 | |
b48c80ec DM |
94 | static atomic_t v4_seq = ATOMIC_INIT(0); |
95 | static atomic_t v6_seq = ATOMIC_INIT(0); | |
96 | ||
97 | static atomic_t *inetpeer_seq_ptr(int family) | |
98 | { | |
99 | return (family == AF_INET ? &v4_seq : &v6_seq); | |
100 | } | |
101 | ||
102 | static inline void flush_check(struct inet_peer_base *base, int family) | |
103 | { | |
104 | atomic_t *fp = inetpeer_seq_ptr(family); | |
105 | ||
106 | if (unlikely(base->flush_seq != atomic_read(fp))) { | |
107 | inetpeer_invalidate_tree(base); | |
108 | base->flush_seq = atomic_read(fp); | |
109 | } | |
110 | } | |
111 | ||
112 | void inetpeer_invalidate_family(int family) | |
113 | { | |
114 | atomic_t *fp = inetpeer_seq_ptr(family); | |
115 | ||
116 | atomic_inc(fp); | |
117 | } | |
118 | ||
1da177e4 LT |
119 | #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */ |
120 | ||
1da177e4 | 121 | /* Exported for sysctl_net_ipv4. */ |
243bbcaa | 122 | int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more |
1da177e4 | 123 | * aggressively at this stage */ |
243bbcaa ED |
124 | int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */ |
125 | int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */ | |
1da177e4 | 126 | |
5faa5df1 SK |
127 | static void inetpeer_gc_worker(struct work_struct *work) |
128 | { | |
129 | struct inet_peer *p, *n; | |
130 | LIST_HEAD(list); | |
131 | ||
132 | spin_lock_bh(&gc_lock); | |
133 | list_replace_init(&gc_list, &list); | |
134 | spin_unlock_bh(&gc_lock); | |
135 | ||
136 | if (list_empty(&list)) | |
137 | return; | |
138 | ||
139 | list_for_each_entry_safe(p, n, &list, gc_list) { | |
140 | ||
141 | if(need_resched()) | |
142 | cond_resched(); | |
143 | ||
144 | if (p->avl_left != peer_avl_empty) { | |
145 | list_add_tail(&p->avl_left->gc_list, &list); | |
146 | p->avl_left = peer_avl_empty; | |
147 | } | |
148 | ||
149 | if (p->avl_right != peer_avl_empty) { | |
150 | list_add_tail(&p->avl_right->gc_list, &list); | |
151 | p->avl_right = peer_avl_empty; | |
152 | } | |
153 | ||
154 | n = list_entry(p->gc_list.next, struct inet_peer, gc_list); | |
155 | ||
156 | if (!atomic_read(&p->refcnt)) { | |
157 | list_del(&p->gc_list); | |
158 | kmem_cache_free(peer_cachep, p); | |
159 | } | |
160 | } | |
161 | ||
162 | if (list_empty(&list)) | |
163 | return; | |
164 | ||
165 | spin_lock_bh(&gc_lock); | |
166 | list_splice(&list, &gc_list); | |
167 | spin_unlock_bh(&gc_lock); | |
168 | ||
169 | schedule_delayed_work(&gc_work, gc_delay); | |
170 | } | |
1da177e4 LT |
171 | |
172 | /* Called from ip_output.c:ip_init */ | |
173 | void __init inet_initpeers(void) | |
174 | { | |
175 | struct sysinfo si; | |
176 | ||
177 | /* Use the straight interface to information about memory. */ | |
178 | si_meminfo(&si); | |
179 | /* The values below were suggested by Alexey Kuznetsov | |
180 | * <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values | |
181 | * myself. --SAW | |
182 | */ | |
183 | if (si.totalram <= (32768*1024)/PAGE_SIZE) | |
184 | inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */ | |
185 | if (si.totalram <= (16384*1024)/PAGE_SIZE) | |
186 | inet_peer_threshold >>= 1; /* about 512KB */ | |
187 | if (si.totalram <= (8192*1024)/PAGE_SIZE) | |
188 | inet_peer_threshold >>= 2; /* about 128KB */ | |
189 | ||
190 | peer_cachep = kmem_cache_create("inet_peer_cache", | |
191 | sizeof(struct inet_peer), | |
317fe0e6 | 192 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, |
20c2df83 | 193 | NULL); |
1da177e4 | 194 | |
5faa5df1 | 195 | INIT_DELAYED_WORK_DEFERRABLE(&gc_work, inetpeer_gc_worker); |
1da177e4 LT |
196 | } |
197 | ||
8790ca17 DM |
198 | static int addr_compare(const struct inetpeer_addr *a, |
199 | const struct inetpeer_addr *b) | |
02663045 DM |
200 | { |
201 | int i, n = (a->family == AF_INET ? 1 : 4); | |
202 | ||
203 | for (i = 0; i < n; i++) { | |
7a71ed89 | 204 | if (a->addr.a6[i] == b->addr.a6[i]) |
02663045 | 205 | continue; |
747465ef | 206 | if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i]) |
02663045 DM |
207 | return -1; |
208 | return 1; | |
209 | } | |
210 | ||
211 | return 0; | |
212 | } | |
213 | ||
65e8354e ED |
214 | #define rcu_deref_locked(X, BASE) \ |
215 | rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock)) | |
216 | ||
243bbcaa ED |
217 | /* |
218 | * Called with local BH disabled and the pool lock held. | |
243bbcaa | 219 | */ |
98158f5a | 220 | #define lookup(_daddr, _stack, _base) \ |
1da177e4 | 221 | ({ \ |
b914c4ea ED |
222 | struct inet_peer *u; \ |
223 | struct inet_peer __rcu **v; \ | |
aa1039e7 ED |
224 | \ |
225 | stackptr = _stack; \ | |
98158f5a | 226 | *stackptr++ = &_base->root; \ |
65e8354e | 227 | for (u = rcu_deref_locked(_base->root, _base); \ |
b914c4ea | 228 | u != peer_avl_empty; ) { \ |
02663045 DM |
229 | int cmp = addr_compare(_daddr, &u->daddr); \ |
230 | if (cmp == 0) \ | |
1da177e4 | 231 | break; \ |
02663045 | 232 | if (cmp == -1) \ |
1da177e4 LT |
233 | v = &u->avl_left; \ |
234 | else \ | |
235 | v = &u->avl_right; \ | |
aa1039e7 | 236 | *stackptr++ = v; \ |
65e8354e | 237 | u = rcu_deref_locked(*v, _base); \ |
1da177e4 LT |
238 | } \ |
239 | u; \ | |
240 | }) | |
241 | ||
aa1039e7 | 242 | /* |
7b46ac4e | 243 | * Called with rcu_read_lock() |
aa1039e7 ED |
244 | * Because we hold no lock against a writer, its quite possible we fall |
245 | * in an endless loop. | |
246 | * But every pointer we follow is guaranteed to be valid thanks to RCU. | |
247 | * We exit from this function if number of links exceeds PEER_MAXDEPTH | |
248 | */ | |
7b46ac4e | 249 | static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr, |
4b9d9be8 | 250 | struct inet_peer_base *base) |
aa1039e7 | 251 | { |
7b46ac4e | 252 | struct inet_peer *u = rcu_dereference(base->root); |
aa1039e7 ED |
253 | int count = 0; |
254 | ||
255 | while (u != peer_avl_empty) { | |
02663045 DM |
256 | int cmp = addr_compare(daddr, &u->daddr); |
257 | if (cmp == 0) { | |
5f2f8920 | 258 | /* Before taking a reference, check if this entry was |
4b9d9be8 | 259 | * deleted (refcnt=-1) |
5f2f8920 | 260 | */ |
4b9d9be8 | 261 | if (!atomic_add_unless(&u->refcnt, 1, -1)) |
aa1039e7 ED |
262 | u = NULL; |
263 | return u; | |
264 | } | |
02663045 | 265 | if (cmp == -1) |
7b46ac4e | 266 | u = rcu_dereference(u->avl_left); |
aa1039e7 | 267 | else |
7b46ac4e | 268 | u = rcu_dereference(u->avl_right); |
aa1039e7 ED |
269 | if (unlikely(++count == PEER_MAXDEPTH)) |
270 | break; | |
271 | } | |
272 | return NULL; | |
273 | } | |
274 | ||
275 | /* Called with local BH disabled and the pool lock held. */ | |
98158f5a | 276 | #define lookup_rightempty(start, base) \ |
1da177e4 | 277 | ({ \ |
b914c4ea ED |
278 | struct inet_peer *u; \ |
279 | struct inet_peer __rcu **v; \ | |
1da177e4 LT |
280 | *stackptr++ = &start->avl_left; \ |
281 | v = &start->avl_left; \ | |
65e8354e | 282 | for (u = rcu_deref_locked(*v, base); \ |
b914c4ea | 283 | u->avl_right != peer_avl_empty_rcu; ) { \ |
1da177e4 LT |
284 | v = &u->avl_right; \ |
285 | *stackptr++ = v; \ | |
65e8354e | 286 | u = rcu_deref_locked(*v, base); \ |
1da177e4 LT |
287 | } \ |
288 | u; \ | |
289 | }) | |
290 | ||
aa1039e7 | 291 | /* Called with local BH disabled and the pool lock held. |
1da177e4 | 292 | * Variable names are the proof of operation correctness. |
aa1039e7 ED |
293 | * Look into mm/map_avl.c for more detail description of the ideas. |
294 | */ | |
b914c4ea | 295 | static void peer_avl_rebalance(struct inet_peer __rcu **stack[], |
98158f5a DM |
296 | struct inet_peer __rcu ***stackend, |
297 | struct inet_peer_base *base) | |
1da177e4 | 298 | { |
b914c4ea ED |
299 | struct inet_peer __rcu **nodep; |
300 | struct inet_peer *node, *l, *r; | |
1da177e4 LT |
301 | int lh, rh; |
302 | ||
303 | while (stackend > stack) { | |
304 | nodep = *--stackend; | |
65e8354e ED |
305 | node = rcu_deref_locked(*nodep, base); |
306 | l = rcu_deref_locked(node->avl_left, base); | |
307 | r = rcu_deref_locked(node->avl_right, base); | |
1da177e4 LT |
308 | lh = node_height(l); |
309 | rh = node_height(r); | |
310 | if (lh > rh + 1) { /* l: RH+2 */ | |
311 | struct inet_peer *ll, *lr, *lrl, *lrr; | |
312 | int lrh; | |
65e8354e ED |
313 | ll = rcu_deref_locked(l->avl_left, base); |
314 | lr = rcu_deref_locked(l->avl_right, base); | |
1da177e4 LT |
315 | lrh = node_height(lr); |
316 | if (lrh <= node_height(ll)) { /* ll: RH+1 */ | |
b914c4ea ED |
317 | RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */ |
318 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 319 | node->avl_height = lrh + 1; /* RH+1 or RH+2 */ |
b914c4ea ED |
320 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */ |
321 | RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */ | |
1da177e4 | 322 | l->avl_height = node->avl_height + 1; |
b914c4ea | 323 | RCU_INIT_POINTER(*nodep, l); |
1da177e4 | 324 | } else { /* ll: RH, lr: RH+1 */ |
65e8354e ED |
325 | lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */ |
326 | lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */ | |
b914c4ea ED |
327 | RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */ |
328 | RCU_INIT_POINTER(node->avl_right, r); /* r: RH */ | |
1da177e4 | 329 | node->avl_height = rh + 1; /* node: RH+1 */ |
b914c4ea ED |
330 | RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */ |
331 | RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */ | |
1da177e4 | 332 | l->avl_height = rh + 1; /* l: RH+1 */ |
b914c4ea ED |
333 | RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */ |
334 | RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */ | |
1da177e4 | 335 | lr->avl_height = rh + 2; |
b914c4ea | 336 | RCU_INIT_POINTER(*nodep, lr); |
1da177e4 LT |
337 | } |
338 | } else if (rh > lh + 1) { /* r: LH+2 */ | |
339 | struct inet_peer *rr, *rl, *rlr, *rll; | |
340 | int rlh; | |
65e8354e ED |
341 | rr = rcu_deref_locked(r->avl_right, base); |
342 | rl = rcu_deref_locked(r->avl_left, base); | |
1da177e4 LT |
343 | rlh = node_height(rl); |
344 | if (rlh <= node_height(rr)) { /* rr: LH+1 */ | |
b914c4ea ED |
345 | RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */ |
346 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 347 | node->avl_height = rlh + 1; /* LH+1 or LH+2 */ |
b914c4ea ED |
348 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */ |
349 | RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */ | |
1da177e4 | 350 | r->avl_height = node->avl_height + 1; |
b914c4ea | 351 | RCU_INIT_POINTER(*nodep, r); |
1da177e4 | 352 | } else { /* rr: RH, rl: RH+1 */ |
65e8354e ED |
353 | rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */ |
354 | rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */ | |
b914c4ea ED |
355 | RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */ |
356 | RCU_INIT_POINTER(node->avl_left, l); /* l: LH */ | |
1da177e4 | 357 | node->avl_height = lh + 1; /* node: LH+1 */ |
b914c4ea ED |
358 | RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */ |
359 | RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */ | |
1da177e4 | 360 | r->avl_height = lh + 1; /* r: LH+1 */ |
b914c4ea ED |
361 | RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */ |
362 | RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */ | |
1da177e4 | 363 | rl->avl_height = lh + 2; |
b914c4ea | 364 | RCU_INIT_POINTER(*nodep, rl); |
1da177e4 LT |
365 | } |
366 | } else { | |
367 | node->avl_height = (lh > rh ? lh : rh) + 1; | |
368 | } | |
369 | } | |
370 | } | |
371 | ||
aa1039e7 | 372 | /* Called with local BH disabled and the pool lock held. */ |
98158f5a | 373 | #define link_to_pool(n, base) \ |
1da177e4 LT |
374 | do { \ |
375 | n->avl_height = 1; \ | |
b914c4ea ED |
376 | n->avl_left = peer_avl_empty_rcu; \ |
377 | n->avl_right = peer_avl_empty_rcu; \ | |
378 | /* lockless readers can catch us now */ \ | |
379 | rcu_assign_pointer(**--stackptr, n); \ | |
98158f5a | 380 | peer_avl_rebalance(stack, stackptr, base); \ |
d6cc1d64 | 381 | } while (0) |
1da177e4 | 382 | |
aa1039e7 ED |
383 | static void inetpeer_free_rcu(struct rcu_head *head) |
384 | { | |
385 | kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu)); | |
386 | } | |
387 | ||
66944e1c ED |
388 | static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base, |
389 | struct inet_peer __rcu **stack[PEER_MAXDEPTH]) | |
1da177e4 | 390 | { |
4b9d9be8 ED |
391 | struct inet_peer __rcu ***stackptr, ***delp; |
392 | ||
393 | if (lookup(&p->daddr, stack, base) != p) | |
394 | BUG(); | |
395 | delp = stackptr - 1; /* *delp[0] == p */ | |
396 | if (p->avl_left == peer_avl_empty_rcu) { | |
397 | *delp[0] = p->avl_right; | |
398 | --stackptr; | |
399 | } else { | |
400 | /* look for a node to insert instead of p */ | |
401 | struct inet_peer *t; | |
402 | t = lookup_rightempty(p, base); | |
403 | BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t); | |
404 | **--stackptr = t->avl_left; | |
405 | /* t is removed, t->daddr > x->daddr for any | |
406 | * x in p->avl_left subtree. | |
407 | * Put t in the old place of p. */ | |
408 | RCU_INIT_POINTER(*delp[0], t); | |
409 | t->avl_left = p->avl_left; | |
410 | t->avl_right = p->avl_right; | |
411 | t->avl_height = p->avl_height; | |
412 | BUG_ON(delp[1] != &p->avl_left); | |
413 | delp[1] = &t->avl_left; /* was &p->avl_left */ | |
1da177e4 | 414 | } |
4b9d9be8 ED |
415 | peer_avl_rebalance(stack, stackptr, base); |
416 | base->total--; | |
417 | call_rcu(&p->rcu, inetpeer_free_rcu); | |
1da177e4 LT |
418 | } |
419 | ||
4b9d9be8 ED |
420 | /* perform garbage collect on all items stacked during a lookup */ |
421 | static int inet_peer_gc(struct inet_peer_base *base, | |
422 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], | |
423 | struct inet_peer __rcu ***stackptr) | |
98158f5a | 424 | { |
4b9d9be8 ED |
425 | struct inet_peer *p, *gchead = NULL; |
426 | __u32 delta, ttl; | |
427 | int cnt = 0; | |
d71209de | 428 | |
4b9d9be8 ED |
429 | if (base->total >= inet_peer_threshold) |
430 | ttl = 0; /* be aggressive */ | |
431 | else | |
432 | ttl = inet_peer_maxttl | |
433 | - (inet_peer_maxttl - inet_peer_minttl) / HZ * | |
434 | base->total / inet_peer_threshold * HZ; | |
435 | stackptr--; /* last stack slot is peer_avl_empty */ | |
436 | while (stackptr > stack) { | |
437 | stackptr--; | |
438 | p = rcu_deref_locked(**stackptr, base); | |
6d1a3e04 ED |
439 | if (atomic_read(&p->refcnt) == 0) { |
440 | smp_rmb(); | |
441 | delta = (__u32)jiffies - p->dtime; | |
442 | if (delta >= ttl && | |
443 | atomic_cmpxchg(&p->refcnt, 0, -1) == 0) { | |
444 | p->gc_next = gchead; | |
445 | gchead = p; | |
446 | } | |
1da177e4 | 447 | } |
1da177e4 | 448 | } |
4b9d9be8 ED |
449 | while ((p = gchead) != NULL) { |
450 | gchead = p->gc_next; | |
451 | cnt++; | |
452 | unlink_from_pool(p, base, stack); | |
453 | } | |
454 | return cnt; | |
1da177e4 LT |
455 | } |
456 | ||
c0efc887 | 457 | struct inet_peer *inet_getpeer(struct inet_peer_base *base, |
c8a627ed G |
458 | const struct inetpeer_addr *daddr, |
459 | int create) | |
1da177e4 | 460 | { |
b914c4ea | 461 | struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr; |
98158f5a | 462 | struct inet_peer *p; |
65e8354e | 463 | unsigned int sequence; |
4b9d9be8 | 464 | int invalidated, gccnt = 0; |
1da177e4 | 465 | |
b48c80ec DM |
466 | flush_check(base, daddr->family); |
467 | ||
4b9d9be8 | 468 | /* Attempt a lockless lookup first. |
aa1039e7 ED |
469 | * Because of a concurrent writer, we might not find an existing entry. |
470 | */ | |
7b46ac4e | 471 | rcu_read_lock(); |
65e8354e | 472 | sequence = read_seqbegin(&base->lock); |
4b9d9be8 | 473 | p = lookup_rcu(daddr, base); |
65e8354e | 474 | invalidated = read_seqretry(&base->lock, sequence); |
7b46ac4e | 475 | rcu_read_unlock(); |
aa1039e7 | 476 | |
4b9d9be8 | 477 | if (p) |
aa1039e7 | 478 | return p; |
1da177e4 | 479 | |
65e8354e ED |
480 | /* If no writer did a change during our lookup, we can return early. */ |
481 | if (!create && !invalidated) | |
482 | return NULL; | |
483 | ||
aa1039e7 ED |
484 | /* retry an exact lookup, taking the lock before. |
485 | * At least, nodes should be hot in our cache. | |
486 | */ | |
65e8354e | 487 | write_seqlock_bh(&base->lock); |
4b9d9be8 | 488 | relookup: |
02663045 | 489 | p = lookup(daddr, stack, base); |
1da177e4 | 490 | if (p != peer_avl_empty) { |
4b9d9be8 | 491 | atomic_inc(&p->refcnt); |
65e8354e | 492 | write_sequnlock_bh(&base->lock); |
4b9d9be8 ED |
493 | return p; |
494 | } | |
495 | if (!gccnt) { | |
496 | gccnt = inet_peer_gc(base, stack, stackptr); | |
497 | if (gccnt && create) | |
498 | goto relookup; | |
1da177e4 | 499 | } |
aa1039e7 ED |
500 | p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL; |
501 | if (p) { | |
b534ecf1 | 502 | p->daddr = *daddr; |
aa1039e7 ED |
503 | atomic_set(&p->refcnt, 1); |
504 | atomic_set(&p->rid, 0); | |
87c48fa3 ED |
505 | atomic_set(&p->ip_id_count, |
506 | (daddr->family == AF_INET) ? | |
507 | secure_ip_id(daddr->addr.a4) : | |
508 | secure_ipv6_id(daddr->addr.a6)); | |
aa1039e7 | 509 | p->tcp_ts_stamp = 0; |
144001bd | 510 | p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW; |
92d86829 DM |
511 | p->rate_tokens = 0; |
512 | p->rate_last = 0; | |
ddd4aa42 | 513 | p->pmtu_expires = 0; |
46af3180 | 514 | p->pmtu_orig = 0; |
ddd4aa42 | 515 | memset(&p->redirect_learned, 0, sizeof(p->redirect_learned)); |
5faa5df1 | 516 | INIT_LIST_HEAD(&p->gc_list); |
aa1039e7 ED |
517 | |
518 | /* Link the node. */ | |
98158f5a DM |
519 | link_to_pool(p, base); |
520 | base->total++; | |
aa1039e7 | 521 | } |
65e8354e | 522 | write_sequnlock_bh(&base->lock); |
1da177e4 | 523 | |
1da177e4 LT |
524 | return p; |
525 | } | |
b3419363 | 526 | EXPORT_SYMBOL_GPL(inet_getpeer); |
98158f5a | 527 | |
4663afe2 ED |
528 | void inet_putpeer(struct inet_peer *p) |
529 | { | |
4b9d9be8 | 530 | p->dtime = (__u32)jiffies; |
6d1a3e04 | 531 | smp_mb__before_atomic_dec(); |
4b9d9be8 | 532 | atomic_dec(&p->refcnt); |
4663afe2 | 533 | } |
b3419363 | 534 | EXPORT_SYMBOL_GPL(inet_putpeer); |
92d86829 DM |
535 | |
536 | /* | |
537 | * Check transmit rate limitation for given message. | |
538 | * The rate information is held in the inet_peer entries now. | |
539 | * This function is generic and could be used for other purposes | |
540 | * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. | |
541 | * | |
542 | * Note that the same inet_peer fields are modified by functions in | |
543 | * route.c too, but these work for packet destinations while xrlim_allow | |
544 | * works for icmp destinations. This means the rate limiting information | |
545 | * for one "ip object" is shared - and these ICMPs are twice limited: | |
546 | * by source and by destination. | |
547 | * | |
548 | * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate | |
549 | * SHOULD allow setting of rate limits | |
550 | * | |
551 | * Shared between ICMPv4 and ICMPv6. | |
552 | */ | |
553 | #define XRLIM_BURST_FACTOR 6 | |
554 | bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout) | |
555 | { | |
556 | unsigned long now, token; | |
557 | bool rc = false; | |
558 | ||
559 | if (!peer) | |
560 | return true; | |
561 | ||
562 | token = peer->rate_tokens; | |
563 | now = jiffies; | |
564 | token += now - peer->rate_last; | |
565 | peer->rate_last = now; | |
566 | if (token > XRLIM_BURST_FACTOR * timeout) | |
567 | token = XRLIM_BURST_FACTOR * timeout; | |
568 | if (token >= timeout) { | |
569 | token -= timeout; | |
570 | rc = true; | |
571 | } | |
572 | peer->rate_tokens = token; | |
573 | return rc; | |
574 | } | |
575 | EXPORT_SYMBOL(inet_peer_xrlim_allow); | |
5faa5df1 | 576 | |
55432d2b ED |
577 | static void inetpeer_inval_rcu(struct rcu_head *head) |
578 | { | |
579 | struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu); | |
580 | ||
581 | spin_lock_bh(&gc_lock); | |
582 | list_add_tail(&p->gc_list, &gc_list); | |
583 | spin_unlock_bh(&gc_lock); | |
584 | ||
585 | schedule_delayed_work(&gc_work, gc_delay); | |
586 | } | |
587 | ||
56a6b248 | 588 | void inetpeer_invalidate_tree(struct inet_peer_base *base) |
5faa5df1 SK |
589 | { |
590 | struct inet_peer *old, *new, *prev; | |
5faa5df1 SK |
591 | |
592 | write_seqlock_bh(&base->lock); | |
593 | ||
594 | old = base->root; | |
595 | if (old == peer_avl_empty_rcu) | |
596 | goto out; | |
597 | ||
598 | new = peer_avl_empty_rcu; | |
599 | ||
600 | prev = cmpxchg(&base->root, old, new); | |
601 | if (prev == old) { | |
602 | base->total = 0; | |
55432d2b | 603 | call_rcu(&prev->gc_rcu, inetpeer_inval_rcu); |
5faa5df1 SK |
604 | } |
605 | ||
606 | out: | |
607 | write_sequnlock_bh(&base->lock); | |
608 | } | |
609 | EXPORT_SYMBOL(inetpeer_invalidate_tree); |