]>
Commit | Line | Data |
---|---|---|
7eb95156 PE |
1 | /* |
2 | * inet fragments management | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Authors: Pavel Emelyanov <xemul@openvz.org> | |
10 | * Started as consolidation of ipv4/ip_fragment.c, | |
11 | * ipv6/reassembly. and ipv6 nf conntrack reassembly | |
12 | */ | |
13 | ||
14 | #include <linux/list.h> | |
15 | #include <linux/spinlock.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/timer.h> | |
18 | #include <linux/mm.h> | |
321a3a99 | 19 | #include <linux/random.h> |
1e4b8287 PE |
20 | #include <linux/skbuff.h> |
21 | #include <linux/rtnetlink.h> | |
5a0e3ad6 | 22 | #include <linux/slab.h> |
7eb95156 | 23 | |
5a3da1fe | 24 | #include <net/sock.h> |
7eb95156 | 25 | #include <net/inet_frag.h> |
be991971 HFS |
26 | #include <net/inet_ecn.h> |
27 | ||
b13d3cbf FW |
28 | #define INETFRAGS_EVICT_BUCKETS 128 |
29 | #define INETFRAGS_EVICT_MAX 512 | |
30 | ||
e3a57d18 FW |
31 | /* don't rebuild inetfrag table with new secret more often than this */ |
32 | #define INETFRAGS_MIN_REBUILD_INTERVAL (5 * HZ) | |
33 | ||
be991971 HFS |
34 | /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements |
35 | * Value : 0xff if frame should be dropped. | |
36 | * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field | |
37 | */ | |
38 | const u8 ip_frag_ecn_table[16] = { | |
39 | /* at least one fragment had CE, and others ECT_0 or ECT_1 */ | |
40 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE, | |
41 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
42 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
43 | ||
44 | /* invalid combinations : drop frame */ | |
45 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff, | |
46 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff, | |
47 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff, | |
48 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
49 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff, | |
50 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff, | |
51 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
52 | }; | |
53 | EXPORT_SYMBOL(ip_frag_ecn_table); | |
7eb95156 | 54 | |
fb3cfe6e FW |
55 | static unsigned int |
56 | inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q) | |
57 | { | |
58 | return f->hashfn(q) & (INETFRAGS_HASHSZ - 1); | |
59 | } | |
60 | ||
e3a57d18 FW |
61 | static bool inet_frag_may_rebuild(struct inet_frags *f) |
62 | { | |
63 | return time_after(jiffies, | |
64 | f->last_rebuild_jiffies + INETFRAGS_MIN_REBUILD_INTERVAL); | |
65 | } | |
66 | ||
67 | static void inet_frag_secret_rebuild(struct inet_frags *f) | |
321a3a99 | 68 | { |
321a3a99 PE |
69 | int i; |
70 | ||
19952cc4 | 71 | /* Per bucket lock NOT needed here, due to write lock protection */ |
e3a57d18 FW |
72 | write_lock_bh(&f->lock); |
73 | ||
74 | if (!inet_frag_may_rebuild(f)) | |
75 | goto out; | |
19952cc4 | 76 | |
321a3a99 | 77 | get_random_bytes(&f->rnd, sizeof(u32)); |
e3a57d18 | 78 | |
321a3a99 | 79 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { |
19952cc4 | 80 | struct inet_frag_bucket *hb; |
321a3a99 | 81 | struct inet_frag_queue *q; |
b67bfe0d | 82 | struct hlist_node *n; |
321a3a99 | 83 | |
19952cc4 JDB |
84 | hb = &f->hash[i]; |
85 | hlist_for_each_entry_safe(q, n, &hb->chain, list) { | |
fb3cfe6e | 86 | unsigned int hval = inet_frag_hashfn(f, q); |
321a3a99 PE |
87 | |
88 | if (hval != i) { | |
19952cc4 JDB |
89 | struct inet_frag_bucket *hb_dest; |
90 | ||
321a3a99 PE |
91 | hlist_del(&q->list); |
92 | ||
93 | /* Relink to new hash chain. */ | |
19952cc4 JDB |
94 | hb_dest = &f->hash[hval]; |
95 | hlist_add_head(&q->list, &hb_dest->chain); | |
321a3a99 PE |
96 | } |
97 | } | |
98 | } | |
321a3a99 | 99 | |
e3a57d18 FW |
100 | f->rebuild = false; |
101 | f->last_rebuild_jiffies = jiffies; | |
102 | out: | |
103 | write_unlock_bh(&f->lock); | |
321a3a99 PE |
104 | } |
105 | ||
b13d3cbf FW |
106 | static bool inet_fragq_should_evict(const struct inet_frag_queue *q) |
107 | { | |
108 | return q->net->low_thresh == 0 || | |
109 | frag_mem_limit(q->net) >= q->net->low_thresh; | |
110 | } | |
111 | ||
112 | static unsigned int | |
113 | inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb) | |
114 | { | |
115 | struct inet_frag_queue *fq; | |
116 | struct hlist_node *n; | |
117 | unsigned int evicted = 0; | |
118 | HLIST_HEAD(expired); | |
119 | ||
120 | evict_again: | |
121 | spin_lock(&hb->chain_lock); | |
122 | ||
123 | hlist_for_each_entry_safe(fq, n, &hb->chain, list) { | |
124 | if (!inet_fragq_should_evict(fq)) | |
125 | continue; | |
126 | ||
127 | if (!del_timer(&fq->timer)) { | |
128 | /* q expiring right now thus increment its refcount so | |
129 | * it won't be freed under us and wait until the timer | |
130 | * has finished executing then destroy it | |
131 | */ | |
132 | atomic_inc(&fq->refcnt); | |
133 | spin_unlock(&hb->chain_lock); | |
134 | del_timer_sync(&fq->timer); | |
135 | WARN_ON(atomic_read(&fq->refcnt) != 1); | |
136 | inet_frag_put(fq, f); | |
137 | goto evict_again; | |
138 | } | |
139 | ||
140 | /* suppress xmit of (icmp) error packet */ | |
141 | fq->last_in &= ~INET_FRAG_FIRST_IN; | |
142 | fq->last_in |= INET_FRAG_EVICTED; | |
143 | hlist_del(&fq->list); | |
144 | hlist_add_head(&fq->list, &expired); | |
145 | ++evicted; | |
146 | } | |
147 | ||
148 | spin_unlock(&hb->chain_lock); | |
149 | ||
150 | hlist_for_each_entry_safe(fq, n, &expired, list) | |
151 | f->frag_expire((unsigned long) fq); | |
152 | ||
153 | return evicted; | |
154 | } | |
155 | ||
156 | static void inet_frag_worker(struct work_struct *work) | |
157 | { | |
158 | unsigned int budget = INETFRAGS_EVICT_BUCKETS; | |
159 | unsigned int i, evicted = 0; | |
160 | struct inet_frags *f; | |
161 | ||
162 | f = container_of(work, struct inet_frags, frags_work); | |
163 | ||
164 | BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ); | |
165 | ||
166 | read_lock_bh(&f->lock); | |
167 | ||
168 | for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) { | |
169 | evicted += inet_evict_bucket(f, &f->hash[i]); | |
170 | i = (i + 1) & (INETFRAGS_HASHSZ - 1); | |
171 | if (evicted > INETFRAGS_EVICT_MAX) | |
172 | break; | |
173 | } | |
174 | ||
175 | f->next_bucket = i; | |
176 | ||
177 | read_unlock_bh(&f->lock); | |
e3a57d18 FW |
178 | if (f->rebuild && inet_frag_may_rebuild(f)) |
179 | inet_frag_secret_rebuild(f); | |
b13d3cbf FW |
180 | } |
181 | ||
182 | static void inet_frag_schedule_worker(struct inet_frags *f) | |
183 | { | |
184 | if (unlikely(!work_pending(&f->frags_work))) | |
185 | schedule_work(&f->frags_work); | |
186 | } | |
187 | ||
7eb95156 PE |
188 | void inet_frags_init(struct inet_frags *f) |
189 | { | |
190 | int i; | |
191 | ||
b13d3cbf FW |
192 | INIT_WORK(&f->frags_work, inet_frag_worker); |
193 | ||
19952cc4 JDB |
194 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { |
195 | struct inet_frag_bucket *hb = &f->hash[i]; | |
7eb95156 | 196 | |
19952cc4 JDB |
197 | spin_lock_init(&hb->chain_lock); |
198 | INIT_HLIST_HEAD(&hb->chain); | |
199 | } | |
7eb95156 | 200 | rwlock_init(&f->lock); |
e3a57d18 | 201 | f->last_rebuild_jiffies = 0; |
7eb95156 PE |
202 | } |
203 | EXPORT_SYMBOL(inet_frags_init); | |
204 | ||
e5a2bb84 PE |
205 | void inet_frags_init_net(struct netns_frags *nf) |
206 | { | |
d433673e | 207 | init_frag_mem_limit(nf); |
e5a2bb84 PE |
208 | } |
209 | EXPORT_SYMBOL(inet_frags_init_net); | |
210 | ||
7eb95156 PE |
211 | void inet_frags_fini(struct inet_frags *f) |
212 | { | |
b13d3cbf | 213 | cancel_work_sync(&f->frags_work); |
7eb95156 PE |
214 | } |
215 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 216 | |
81566e83 PE |
217 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
218 | { | |
b13d3cbf FW |
219 | int i; |
220 | ||
81566e83 | 221 | nf->low_thresh = 0; |
e8e16b70 | 222 | |
b13d3cbf FW |
223 | read_lock_bh(&f->lock); |
224 | ||
225 | for (i = 0; i < INETFRAGS_HASHSZ ; i++) | |
226 | inet_evict_bucket(f, &f->hash[i]); | |
227 | ||
228 | read_unlock_bh(&f->lock); | |
6d7b857d JDB |
229 | |
230 | percpu_counter_destroy(&nf->mem); | |
81566e83 PE |
231 | } |
232 | EXPORT_SYMBOL(inet_frags_exit_net); | |
233 | ||
277e650d PE |
234 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) |
235 | { | |
19952cc4 JDB |
236 | struct inet_frag_bucket *hb; |
237 | unsigned int hash; | |
238 | ||
239 | read_lock(&f->lock); | |
fb3cfe6e | 240 | hash = inet_frag_hashfn(f, fq); |
19952cc4 JDB |
241 | hb = &f->hash[hash]; |
242 | ||
243 | spin_lock(&hb->chain_lock); | |
277e650d | 244 | hlist_del(&fq->list); |
19952cc4 JDB |
245 | spin_unlock(&hb->chain_lock); |
246 | ||
247 | read_unlock(&f->lock); | |
277e650d PE |
248 | } |
249 | ||
250 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
251 | { | |
252 | if (del_timer(&fq->timer)) | |
253 | atomic_dec(&fq->refcnt); | |
254 | ||
bc578a54 | 255 | if (!(fq->last_in & INET_FRAG_COMPLETE)) { |
277e650d PE |
256 | fq_unlink(fq, f); |
257 | atomic_dec(&fq->refcnt); | |
bc578a54 | 258 | fq->last_in |= INET_FRAG_COMPLETE; |
277e650d PE |
259 | } |
260 | } | |
277e650d | 261 | EXPORT_SYMBOL(inet_frag_kill); |
1e4b8287 | 262 | |
6ddc0822 | 263 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
d433673e | 264 | struct sk_buff *skb) |
1e4b8287 | 265 | { |
1e4b8287 PE |
266 | if (f->skb_free) |
267 | f->skb_free(skb); | |
268 | kfree_skb(skb); | |
269 | } | |
270 | ||
3fd588eb | 271 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f) |
1e4b8287 PE |
272 | { |
273 | struct sk_buff *fp; | |
6ddc0822 | 274 | struct netns_frags *nf; |
d433673e | 275 | unsigned int sum, sum_truesize = 0; |
1e4b8287 | 276 | |
547b792c IJ |
277 | WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); |
278 | WARN_ON(del_timer(&q->timer) != 0); | |
1e4b8287 PE |
279 | |
280 | /* Release all fragment data. */ | |
281 | fp = q->fragments; | |
6ddc0822 | 282 | nf = q->net; |
1e4b8287 PE |
283 | while (fp) { |
284 | struct sk_buff *xp = fp->next; | |
285 | ||
d433673e JDB |
286 | sum_truesize += fp->truesize; |
287 | frag_kfree_skb(nf, f, fp); | |
1e4b8287 PE |
288 | fp = xp; |
289 | } | |
d433673e | 290 | sum = sum_truesize + f->qsize; |
d433673e | 291 | sub_frag_mem_limit(q, sum); |
1e4b8287 | 292 | |
c9547709 PE |
293 | if (f->destructor) |
294 | f->destructor(q); | |
295 | kfree(q); | |
1e4b8287 PE |
296 | } |
297 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 298 | |
ac18e750 PE |
299 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
300 | struct inet_frag_queue *qp_in, struct inet_frags *f, | |
9a375803 | 301 | void *arg) |
2588fe1d | 302 | { |
19952cc4 | 303 | struct inet_frag_bucket *hb; |
2588fe1d | 304 | struct inet_frag_queue *qp; |
9a375803 | 305 | unsigned int hash; |
2588fe1d | 306 | |
19952cc4 | 307 | read_lock(&f->lock); /* Protects against hash rebuild */ |
9a375803 PE |
308 | /* |
309 | * While we stayed w/o the lock other CPU could update | |
310 | * the rnd seed, so we need to re-calculate the hash | |
311 | * chain. Fortunatelly the qp_in can be used to get one. | |
312 | */ | |
fb3cfe6e | 313 | hash = inet_frag_hashfn(f, qp_in); |
19952cc4 JDB |
314 | hb = &f->hash[hash]; |
315 | spin_lock(&hb->chain_lock); | |
316 | ||
2588fe1d PE |
317 | #ifdef CONFIG_SMP |
318 | /* With SMP race we have to recheck hash table, because | |
319 | * such entry could be created on other cpu, while we | |
19952cc4 | 320 | * released the hash bucket lock. |
2588fe1d | 321 | */ |
19952cc4 | 322 | hlist_for_each_entry(qp, &hb->chain, list) { |
ac18e750 | 323 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d | 324 | atomic_inc(&qp->refcnt); |
19952cc4 JDB |
325 | spin_unlock(&hb->chain_lock); |
326 | read_unlock(&f->lock); | |
bc578a54 | 327 | qp_in->last_in |= INET_FRAG_COMPLETE; |
2588fe1d PE |
328 | inet_frag_put(qp_in, f); |
329 | return qp; | |
330 | } | |
331 | } | |
332 | #endif | |
333 | qp = qp_in; | |
b2fd5321 | 334 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
335 | atomic_inc(&qp->refcnt); |
336 | ||
337 | atomic_inc(&qp->refcnt); | |
19952cc4 | 338 | hlist_add_head(&qp->list, &hb->chain); |
3fd588eb | 339 | |
19952cc4 JDB |
340 | spin_unlock(&hb->chain_lock); |
341 | read_unlock(&f->lock); | |
24b9bf43 | 342 | |
2588fe1d PE |
343 | return qp; |
344 | } | |
e521db9d | 345 | |
ac18e750 PE |
346 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
347 | struct inet_frags *f, void *arg) | |
e521db9d PE |
348 | { |
349 | struct inet_frag_queue *q; | |
350 | ||
b13d3cbf FW |
351 | if (frag_mem_limit(nf) > nf->high_thresh) { |
352 | inet_frag_schedule_worker(f); | |
86e93e47 | 353 | return NULL; |
b13d3cbf | 354 | } |
86e93e47 | 355 | |
e521db9d PE |
356 | q = kzalloc(f->qsize, GFP_ATOMIC); |
357 | if (q == NULL) | |
358 | return NULL; | |
359 | ||
54db0cc2 | 360 | q->net = nf; |
c6fda282 | 361 | f->constructor(q, arg); |
d433673e JDB |
362 | add_frag_mem_limit(q, f->qsize); |
363 | ||
e521db9d PE |
364 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
365 | spin_lock_init(&q->lock); | |
366 | atomic_set(&q->refcnt, 1); | |
367 | ||
368 | return q; | |
369 | } | |
c6fda282 | 370 | |
ac18e750 | 371 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
9a375803 | 372 | struct inet_frags *f, void *arg) |
c6fda282 PE |
373 | { |
374 | struct inet_frag_queue *q; | |
375 | ||
ac18e750 | 376 | q = inet_frag_alloc(nf, f, arg); |
c6fda282 PE |
377 | if (q == NULL) |
378 | return NULL; | |
379 | ||
9a375803 | 380 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 381 | } |
abd6523d | 382 | |
ac18e750 PE |
383 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
384 | struct inet_frags *f, void *key, unsigned int hash) | |
56bca31f | 385 | __releases(&f->lock) |
abd6523d | 386 | { |
19952cc4 | 387 | struct inet_frag_bucket *hb; |
abd6523d | 388 | struct inet_frag_queue *q; |
5a3da1fe | 389 | int depth = 0; |
abd6523d | 390 | |
b13d3cbf FW |
391 | if (frag_mem_limit(nf) > nf->low_thresh) |
392 | inet_frag_schedule_worker(f); | |
86e93e47 | 393 | |
fb3cfe6e | 394 | hash &= (INETFRAGS_HASHSZ - 1); |
19952cc4 JDB |
395 | hb = &f->hash[hash]; |
396 | ||
397 | spin_lock(&hb->chain_lock); | |
398 | hlist_for_each_entry(q, &hb->chain, list) { | |
ac18e750 | 399 | if (q->net == nf && f->match(q, key)) { |
abd6523d | 400 | atomic_inc(&q->refcnt); |
19952cc4 | 401 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
402 | read_unlock(&f->lock); |
403 | return q; | |
404 | } | |
5a3da1fe | 405 | depth++; |
abd6523d | 406 | } |
19952cc4 | 407 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
408 | read_unlock(&f->lock); |
409 | ||
5a3da1fe HFS |
410 | if (depth <= INETFRAGS_MAXDEPTH) |
411 | return inet_frag_create(nf, f, key); | |
e3a57d18 FW |
412 | |
413 | if (inet_frag_may_rebuild(f)) { | |
414 | f->rebuild = true; | |
415 | inet_frag_schedule_worker(f); | |
416 | } | |
417 | ||
418 | return ERR_PTR(-ENOBUFS); | |
abd6523d PE |
419 | } |
420 | EXPORT_SYMBOL(inet_frag_find); | |
5a3da1fe HFS |
421 | |
422 | void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, | |
423 | const char *prefix) | |
424 | { | |
425 | static const char msg[] = "inet_frag_find: Fragment hash bucket" | |
426 | " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) | |
427 | ". Dropping fragment.\n"; | |
428 | ||
429 | if (PTR_ERR(q) == -ENOBUFS) | |
430 | LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg); | |
431 | } | |
432 | EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); |