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 | ||
be991971 HFS |
31 | /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements |
32 | * Value : 0xff if frame should be dropped. | |
33 | * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field | |
34 | */ | |
35 | const u8 ip_frag_ecn_table[16] = { | |
36 | /* at least one fragment had CE, and others ECT_0 or ECT_1 */ | |
37 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE, | |
38 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
39 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
40 | ||
41 | /* invalid combinations : drop frame */ | |
42 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff, | |
43 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff, | |
44 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff, | |
45 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
46 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff, | |
47 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff, | |
48 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
49 | }; | |
50 | EXPORT_SYMBOL(ip_frag_ecn_table); | |
7eb95156 | 51 | |
fb3cfe6e FW |
52 | static unsigned int |
53 | inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q) | |
54 | { | |
55 | return f->hashfn(q) & (INETFRAGS_HASHSZ - 1); | |
56 | } | |
57 | ||
321a3a99 PE |
58 | static void inet_frag_secret_rebuild(unsigned long dummy) |
59 | { | |
60 | struct inet_frags *f = (struct inet_frags *)dummy; | |
61 | unsigned long now = jiffies; | |
62 | int i; | |
63 | ||
19952cc4 | 64 | /* Per bucket lock NOT needed here, due to write lock protection */ |
321a3a99 | 65 | write_lock(&f->lock); |
19952cc4 | 66 | |
321a3a99 PE |
67 | get_random_bytes(&f->rnd, sizeof(u32)); |
68 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { | |
19952cc4 | 69 | struct inet_frag_bucket *hb; |
321a3a99 | 70 | struct inet_frag_queue *q; |
b67bfe0d | 71 | struct hlist_node *n; |
321a3a99 | 72 | |
19952cc4 JDB |
73 | hb = &f->hash[i]; |
74 | hlist_for_each_entry_safe(q, n, &hb->chain, list) { | |
fb3cfe6e | 75 | unsigned int hval = inet_frag_hashfn(f, q); |
321a3a99 PE |
76 | |
77 | if (hval != i) { | |
19952cc4 JDB |
78 | struct inet_frag_bucket *hb_dest; |
79 | ||
321a3a99 PE |
80 | hlist_del(&q->list); |
81 | ||
82 | /* Relink to new hash chain. */ | |
19952cc4 JDB |
83 | hb_dest = &f->hash[hval]; |
84 | hlist_add_head(&q->list, &hb_dest->chain); | |
321a3a99 PE |
85 | } |
86 | } | |
87 | } | |
88 | write_unlock(&f->lock); | |
89 | ||
3b4bc4a2 | 90 | mod_timer(&f->secret_timer, now + f->secret_interval); |
321a3a99 PE |
91 | } |
92 | ||
b13d3cbf FW |
93 | static bool inet_fragq_should_evict(const struct inet_frag_queue *q) |
94 | { | |
95 | return q->net->low_thresh == 0 || | |
96 | frag_mem_limit(q->net) >= q->net->low_thresh; | |
97 | } | |
98 | ||
99 | static unsigned int | |
100 | inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb) | |
101 | { | |
102 | struct inet_frag_queue *fq; | |
103 | struct hlist_node *n; | |
104 | unsigned int evicted = 0; | |
105 | HLIST_HEAD(expired); | |
106 | ||
107 | evict_again: | |
108 | spin_lock(&hb->chain_lock); | |
109 | ||
110 | hlist_for_each_entry_safe(fq, n, &hb->chain, list) { | |
111 | if (!inet_fragq_should_evict(fq)) | |
112 | continue; | |
113 | ||
114 | if (!del_timer(&fq->timer)) { | |
115 | /* q expiring right now thus increment its refcount so | |
116 | * it won't be freed under us and wait until the timer | |
117 | * has finished executing then destroy it | |
118 | */ | |
119 | atomic_inc(&fq->refcnt); | |
120 | spin_unlock(&hb->chain_lock); | |
121 | del_timer_sync(&fq->timer); | |
122 | WARN_ON(atomic_read(&fq->refcnt) != 1); | |
123 | inet_frag_put(fq, f); | |
124 | goto evict_again; | |
125 | } | |
126 | ||
127 | /* suppress xmit of (icmp) error packet */ | |
128 | fq->last_in &= ~INET_FRAG_FIRST_IN; | |
129 | fq->last_in |= INET_FRAG_EVICTED; | |
130 | hlist_del(&fq->list); | |
131 | hlist_add_head(&fq->list, &expired); | |
132 | ++evicted; | |
133 | } | |
134 | ||
135 | spin_unlock(&hb->chain_lock); | |
136 | ||
137 | hlist_for_each_entry_safe(fq, n, &expired, list) | |
138 | f->frag_expire((unsigned long) fq); | |
139 | ||
140 | return evicted; | |
141 | } | |
142 | ||
143 | static void inet_frag_worker(struct work_struct *work) | |
144 | { | |
145 | unsigned int budget = INETFRAGS_EVICT_BUCKETS; | |
146 | unsigned int i, evicted = 0; | |
147 | struct inet_frags *f; | |
148 | ||
149 | f = container_of(work, struct inet_frags, frags_work); | |
150 | ||
151 | BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ); | |
152 | ||
153 | read_lock_bh(&f->lock); | |
154 | ||
155 | for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) { | |
156 | evicted += inet_evict_bucket(f, &f->hash[i]); | |
157 | i = (i + 1) & (INETFRAGS_HASHSZ - 1); | |
158 | if (evicted > INETFRAGS_EVICT_MAX) | |
159 | break; | |
160 | } | |
161 | ||
162 | f->next_bucket = i; | |
163 | ||
164 | read_unlock_bh(&f->lock); | |
165 | } | |
166 | ||
167 | static void inet_frag_schedule_worker(struct inet_frags *f) | |
168 | { | |
169 | if (unlikely(!work_pending(&f->frags_work))) | |
170 | schedule_work(&f->frags_work); | |
171 | } | |
172 | ||
7eb95156 PE |
173 | void inet_frags_init(struct inet_frags *f) |
174 | { | |
175 | int i; | |
176 | ||
b13d3cbf FW |
177 | INIT_WORK(&f->frags_work, inet_frag_worker); |
178 | ||
19952cc4 JDB |
179 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { |
180 | struct inet_frag_bucket *hb = &f->hash[i]; | |
7eb95156 | 181 | |
19952cc4 JDB |
182 | spin_lock_init(&hb->chain_lock); |
183 | INIT_HLIST_HEAD(&hb->chain); | |
184 | } | |
7eb95156 PE |
185 | rwlock_init(&f->lock); |
186 | ||
b24b8a24 PE |
187 | setup_timer(&f->secret_timer, inet_frag_secret_rebuild, |
188 | (unsigned long)f); | |
3b4bc4a2 | 189 | f->secret_timer.expires = jiffies + f->secret_interval; |
321a3a99 | 190 | add_timer(&f->secret_timer); |
7eb95156 PE |
191 | } |
192 | EXPORT_SYMBOL(inet_frags_init); | |
193 | ||
e5a2bb84 PE |
194 | void inet_frags_init_net(struct netns_frags *nf) |
195 | { | |
d433673e | 196 | init_frag_mem_limit(nf); |
3140c25c | 197 | INIT_LIST_HEAD(&nf->lru_list); |
3ef0eb0d | 198 | spin_lock_init(&nf->lru_lock); |
e5a2bb84 PE |
199 | } |
200 | EXPORT_SYMBOL(inet_frags_init_net); | |
201 | ||
7eb95156 PE |
202 | void inet_frags_fini(struct inet_frags *f) |
203 | { | |
321a3a99 | 204 | del_timer(&f->secret_timer); |
b13d3cbf | 205 | cancel_work_sync(&f->frags_work); |
7eb95156 PE |
206 | } |
207 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 208 | |
81566e83 PE |
209 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
210 | { | |
b13d3cbf FW |
211 | int i; |
212 | ||
81566e83 | 213 | nf->low_thresh = 0; |
e8e16b70 | 214 | |
b13d3cbf FW |
215 | read_lock_bh(&f->lock); |
216 | ||
217 | for (i = 0; i < INETFRAGS_HASHSZ ; i++) | |
218 | inet_evict_bucket(f, &f->hash[i]); | |
219 | ||
220 | read_unlock_bh(&f->lock); | |
6d7b857d JDB |
221 | |
222 | percpu_counter_destroy(&nf->mem); | |
81566e83 PE |
223 | } |
224 | EXPORT_SYMBOL(inet_frags_exit_net); | |
225 | ||
277e650d PE |
226 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) |
227 | { | |
19952cc4 JDB |
228 | struct inet_frag_bucket *hb; |
229 | unsigned int hash; | |
230 | ||
231 | read_lock(&f->lock); | |
fb3cfe6e | 232 | hash = inet_frag_hashfn(f, fq); |
19952cc4 JDB |
233 | hb = &f->hash[hash]; |
234 | ||
235 | spin_lock(&hb->chain_lock); | |
277e650d | 236 | hlist_del(&fq->list); |
19952cc4 JDB |
237 | spin_unlock(&hb->chain_lock); |
238 | ||
239 | read_unlock(&f->lock); | |
3ef0eb0d | 240 | inet_frag_lru_del(fq); |
277e650d PE |
241 | } |
242 | ||
243 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
244 | { | |
245 | if (del_timer(&fq->timer)) | |
246 | atomic_dec(&fq->refcnt); | |
247 | ||
bc578a54 | 248 | if (!(fq->last_in & INET_FRAG_COMPLETE)) { |
277e650d PE |
249 | fq_unlink(fq, f); |
250 | atomic_dec(&fq->refcnt); | |
bc578a54 | 251 | fq->last_in |= INET_FRAG_COMPLETE; |
277e650d PE |
252 | } |
253 | } | |
277e650d | 254 | EXPORT_SYMBOL(inet_frag_kill); |
1e4b8287 | 255 | |
6ddc0822 | 256 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
d433673e | 257 | struct sk_buff *skb) |
1e4b8287 | 258 | { |
1e4b8287 PE |
259 | if (f->skb_free) |
260 | f->skb_free(skb); | |
261 | kfree_skb(skb); | |
262 | } | |
263 | ||
264 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, | |
265 | int *work) | |
266 | { | |
267 | struct sk_buff *fp; | |
6ddc0822 | 268 | struct netns_frags *nf; |
d433673e | 269 | unsigned int sum, sum_truesize = 0; |
1e4b8287 | 270 | |
547b792c IJ |
271 | WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); |
272 | WARN_ON(del_timer(&q->timer) != 0); | |
1e4b8287 PE |
273 | |
274 | /* Release all fragment data. */ | |
275 | fp = q->fragments; | |
6ddc0822 | 276 | nf = q->net; |
1e4b8287 PE |
277 | while (fp) { |
278 | struct sk_buff *xp = fp->next; | |
279 | ||
d433673e JDB |
280 | sum_truesize += fp->truesize; |
281 | frag_kfree_skb(nf, f, fp); | |
1e4b8287 PE |
282 | fp = xp; |
283 | } | |
d433673e | 284 | sum = sum_truesize + f->qsize; |
1e4b8287 | 285 | if (work) |
d433673e JDB |
286 | *work -= sum; |
287 | sub_frag_mem_limit(q, sum); | |
1e4b8287 | 288 | |
c9547709 PE |
289 | if (f->destructor) |
290 | f->destructor(q); | |
291 | kfree(q); | |
1e4b8287 PE |
292 | |
293 | } | |
294 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 295 | |
ac18e750 PE |
296 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
297 | struct inet_frag_queue *qp_in, struct inet_frags *f, | |
9a375803 | 298 | void *arg) |
2588fe1d | 299 | { |
19952cc4 | 300 | struct inet_frag_bucket *hb; |
2588fe1d | 301 | struct inet_frag_queue *qp; |
9a375803 | 302 | unsigned int hash; |
2588fe1d | 303 | |
19952cc4 | 304 | read_lock(&f->lock); /* Protects against hash rebuild */ |
9a375803 PE |
305 | /* |
306 | * While we stayed w/o the lock other CPU could update | |
307 | * the rnd seed, so we need to re-calculate the hash | |
308 | * chain. Fortunatelly the qp_in can be used to get one. | |
309 | */ | |
fb3cfe6e | 310 | hash = inet_frag_hashfn(f, qp_in); |
19952cc4 JDB |
311 | hb = &f->hash[hash]; |
312 | spin_lock(&hb->chain_lock); | |
313 | ||
2588fe1d PE |
314 | #ifdef CONFIG_SMP |
315 | /* With SMP race we have to recheck hash table, because | |
316 | * such entry could be created on other cpu, while we | |
19952cc4 | 317 | * released the hash bucket lock. |
2588fe1d | 318 | */ |
19952cc4 | 319 | hlist_for_each_entry(qp, &hb->chain, list) { |
ac18e750 | 320 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d | 321 | atomic_inc(&qp->refcnt); |
19952cc4 JDB |
322 | spin_unlock(&hb->chain_lock); |
323 | read_unlock(&f->lock); | |
bc578a54 | 324 | qp_in->last_in |= INET_FRAG_COMPLETE; |
2588fe1d PE |
325 | inet_frag_put(qp_in, f); |
326 | return qp; | |
327 | } | |
328 | } | |
329 | #endif | |
330 | qp = qp_in; | |
b2fd5321 | 331 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
332 | atomic_inc(&qp->refcnt); |
333 | ||
334 | atomic_inc(&qp->refcnt); | |
19952cc4 | 335 | hlist_add_head(&qp->list, &hb->chain); |
24b9bf43 | 336 | inet_frag_lru_add(nf, qp); |
19952cc4 JDB |
337 | spin_unlock(&hb->chain_lock); |
338 | read_unlock(&f->lock); | |
24b9bf43 | 339 | |
2588fe1d PE |
340 | return qp; |
341 | } | |
e521db9d | 342 | |
ac18e750 PE |
343 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
344 | struct inet_frags *f, void *arg) | |
e521db9d PE |
345 | { |
346 | struct inet_frag_queue *q; | |
347 | ||
b13d3cbf FW |
348 | if (frag_mem_limit(nf) > nf->high_thresh) { |
349 | inet_frag_schedule_worker(f); | |
86e93e47 | 350 | return NULL; |
b13d3cbf | 351 | } |
86e93e47 | 352 | |
e521db9d PE |
353 | q = kzalloc(f->qsize, GFP_ATOMIC); |
354 | if (q == NULL) | |
355 | return NULL; | |
356 | ||
54db0cc2 | 357 | q->net = nf; |
c6fda282 | 358 | f->constructor(q, arg); |
d433673e JDB |
359 | add_frag_mem_limit(q, f->qsize); |
360 | ||
e521db9d PE |
361 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
362 | spin_lock_init(&q->lock); | |
363 | atomic_set(&q->refcnt, 1); | |
b56141ab | 364 | INIT_LIST_HEAD(&q->lru_list); |
e521db9d PE |
365 | |
366 | return q; | |
367 | } | |
c6fda282 | 368 | |
ac18e750 | 369 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
9a375803 | 370 | struct inet_frags *f, void *arg) |
c6fda282 PE |
371 | { |
372 | struct inet_frag_queue *q; | |
373 | ||
ac18e750 | 374 | q = inet_frag_alloc(nf, f, arg); |
c6fda282 PE |
375 | if (q == NULL) |
376 | return NULL; | |
377 | ||
9a375803 | 378 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 379 | } |
abd6523d | 380 | |
ac18e750 PE |
381 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
382 | struct inet_frags *f, void *key, unsigned int hash) | |
56bca31f | 383 | __releases(&f->lock) |
abd6523d | 384 | { |
19952cc4 | 385 | struct inet_frag_bucket *hb; |
abd6523d | 386 | struct inet_frag_queue *q; |
5a3da1fe | 387 | int depth = 0; |
abd6523d | 388 | |
b13d3cbf FW |
389 | if (frag_mem_limit(nf) > nf->low_thresh) |
390 | inet_frag_schedule_worker(f); | |
86e93e47 | 391 | |
fb3cfe6e | 392 | hash &= (INETFRAGS_HASHSZ - 1); |
19952cc4 JDB |
393 | hb = &f->hash[hash]; |
394 | ||
395 | spin_lock(&hb->chain_lock); | |
396 | hlist_for_each_entry(q, &hb->chain, list) { | |
ac18e750 | 397 | if (q->net == nf && f->match(q, key)) { |
abd6523d | 398 | atomic_inc(&q->refcnt); |
19952cc4 | 399 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
400 | read_unlock(&f->lock); |
401 | return q; | |
402 | } | |
5a3da1fe | 403 | depth++; |
abd6523d | 404 | } |
19952cc4 | 405 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
406 | read_unlock(&f->lock); |
407 | ||
5a3da1fe HFS |
408 | if (depth <= INETFRAGS_MAXDEPTH) |
409 | return inet_frag_create(nf, f, key); | |
410 | else | |
411 | return ERR_PTR(-ENOBUFS); | |
abd6523d PE |
412 | } |
413 | EXPORT_SYMBOL(inet_frag_find); | |
5a3da1fe HFS |
414 | |
415 | void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, | |
416 | const char *prefix) | |
417 | { | |
418 | static const char msg[] = "inet_frag_find: Fragment hash bucket" | |
419 | " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) | |
420 | ". Dropping fragment.\n"; | |
421 | ||
422 | if (PTR_ERR(q) == -ENOBUFS) | |
423 | LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg); | |
424 | } | |
425 | EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); |