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 | ||
ab1c724f | 71 | write_seqlock_bh(&f->rnd_seqlock); |
e3a57d18 FW |
72 | |
73 | if (!inet_frag_may_rebuild(f)) | |
74 | goto out; | |
19952cc4 | 75 | |
321a3a99 | 76 | get_random_bytes(&f->rnd, sizeof(u32)); |
e3a57d18 | 77 | |
321a3a99 | 78 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { |
19952cc4 | 79 | struct inet_frag_bucket *hb; |
321a3a99 | 80 | struct inet_frag_queue *q; |
b67bfe0d | 81 | struct hlist_node *n; |
321a3a99 | 82 | |
19952cc4 | 83 | hb = &f->hash[i]; |
ab1c724f FW |
84 | spin_lock(&hb->chain_lock); |
85 | ||
19952cc4 | 86 | hlist_for_each_entry_safe(q, n, &hb->chain, list) { |
fb3cfe6e | 87 | unsigned int hval = inet_frag_hashfn(f, q); |
321a3a99 PE |
88 | |
89 | if (hval != i) { | |
19952cc4 JDB |
90 | struct inet_frag_bucket *hb_dest; |
91 | ||
321a3a99 PE |
92 | hlist_del(&q->list); |
93 | ||
94 | /* Relink to new hash chain. */ | |
19952cc4 | 95 | hb_dest = &f->hash[hval]; |
ab1c724f FW |
96 | |
97 | /* This is the only place where we take | |
98 | * another chain_lock while already holding | |
99 | * one. As this will not run concurrently, | |
100 | * we cannot deadlock on hb_dest lock below, if its | |
101 | * already locked it will be released soon since | |
102 | * other caller cannot be waiting for hb lock | |
103 | * that we've taken above. | |
104 | */ | |
105 | spin_lock_nested(&hb_dest->chain_lock, | |
106 | SINGLE_DEPTH_NESTING); | |
19952cc4 | 107 | hlist_add_head(&q->list, &hb_dest->chain); |
ab1c724f | 108 | spin_unlock(&hb_dest->chain_lock); |
321a3a99 PE |
109 | } |
110 | } | |
ab1c724f | 111 | spin_unlock(&hb->chain_lock); |
321a3a99 | 112 | } |
321a3a99 | 113 | |
e3a57d18 FW |
114 | f->rebuild = false; |
115 | f->last_rebuild_jiffies = jiffies; | |
116 | out: | |
ab1c724f | 117 | write_sequnlock_bh(&f->rnd_seqlock); |
321a3a99 PE |
118 | } |
119 | ||
b13d3cbf FW |
120 | static bool inet_fragq_should_evict(const struct inet_frag_queue *q) |
121 | { | |
122 | return q->net->low_thresh == 0 || | |
123 | frag_mem_limit(q->net) >= q->net->low_thresh; | |
124 | } | |
125 | ||
126 | static unsigned int | |
127 | inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb) | |
128 | { | |
129 | struct inet_frag_queue *fq; | |
130 | struct hlist_node *n; | |
131 | unsigned int evicted = 0; | |
132 | HLIST_HEAD(expired); | |
133 | ||
134 | evict_again: | |
135 | spin_lock(&hb->chain_lock); | |
136 | ||
137 | hlist_for_each_entry_safe(fq, n, &hb->chain, list) { | |
138 | if (!inet_fragq_should_evict(fq)) | |
139 | continue; | |
140 | ||
141 | if (!del_timer(&fq->timer)) { | |
142 | /* q expiring right now thus increment its refcount so | |
143 | * it won't be freed under us and wait until the timer | |
144 | * has finished executing then destroy it | |
145 | */ | |
146 | atomic_inc(&fq->refcnt); | |
147 | spin_unlock(&hb->chain_lock); | |
148 | del_timer_sync(&fq->timer); | |
b13d3cbf FW |
149 | inet_frag_put(fq, f); |
150 | goto evict_again; | |
151 | } | |
152 | ||
06aa8b8a | 153 | fq->flags |= INET_FRAG_EVICTED; |
d1fe1944 | 154 | hlist_add_head(&fq->list_evictor, &expired); |
b13d3cbf FW |
155 | ++evicted; |
156 | } | |
157 | ||
158 | spin_unlock(&hb->chain_lock); | |
159 | ||
d1fe1944 | 160 | hlist_for_each_entry_safe(fq, n, &expired, list_evictor) |
b13d3cbf FW |
161 | f->frag_expire((unsigned long) fq); |
162 | ||
163 | return evicted; | |
164 | } | |
165 | ||
166 | static void inet_frag_worker(struct work_struct *work) | |
167 | { | |
168 | unsigned int budget = INETFRAGS_EVICT_BUCKETS; | |
169 | unsigned int i, evicted = 0; | |
170 | struct inet_frags *f; | |
171 | ||
172 | f = container_of(work, struct inet_frags, frags_work); | |
173 | ||
174 | BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ); | |
175 | ||
ab1c724f | 176 | local_bh_disable(); |
b13d3cbf FW |
177 | |
178 | for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) { | |
179 | evicted += inet_evict_bucket(f, &f->hash[i]); | |
180 | i = (i + 1) & (INETFRAGS_HASHSZ - 1); | |
181 | if (evicted > INETFRAGS_EVICT_MAX) | |
182 | break; | |
183 | } | |
184 | ||
185 | f->next_bucket = i; | |
186 | ||
ab1c724f FW |
187 | local_bh_enable(); |
188 | ||
e3a57d18 FW |
189 | if (f->rebuild && inet_frag_may_rebuild(f)) |
190 | inet_frag_secret_rebuild(f); | |
b13d3cbf FW |
191 | } |
192 | ||
193 | static void inet_frag_schedule_worker(struct inet_frags *f) | |
194 | { | |
195 | if (unlikely(!work_pending(&f->frags_work))) | |
196 | schedule_work(&f->frags_work); | |
197 | } | |
198 | ||
d4ad4d22 | 199 | int inet_frags_init(struct inet_frags *f) |
7eb95156 PE |
200 | { |
201 | int i; | |
202 | ||
b13d3cbf FW |
203 | INIT_WORK(&f->frags_work, inet_frag_worker); |
204 | ||
19952cc4 JDB |
205 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { |
206 | struct inet_frag_bucket *hb = &f->hash[i]; | |
7eb95156 | 207 | |
19952cc4 JDB |
208 | spin_lock_init(&hb->chain_lock); |
209 | INIT_HLIST_HEAD(&hb->chain); | |
210 | } | |
ab1c724f FW |
211 | |
212 | seqlock_init(&f->rnd_seqlock); | |
e3a57d18 | 213 | f->last_rebuild_jiffies = 0; |
d4ad4d22 NA |
214 | f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0, |
215 | NULL); | |
216 | if (!f->frags_cachep) | |
217 | return -ENOMEM; | |
218 | ||
219 | return 0; | |
7eb95156 PE |
220 | } |
221 | EXPORT_SYMBOL(inet_frags_init); | |
222 | ||
e5a2bb84 PE |
223 | void inet_frags_init_net(struct netns_frags *nf) |
224 | { | |
d433673e | 225 | init_frag_mem_limit(nf); |
e5a2bb84 PE |
226 | } |
227 | EXPORT_SYMBOL(inet_frags_init_net); | |
228 | ||
7eb95156 PE |
229 | void inet_frags_fini(struct inet_frags *f) |
230 | { | |
b13d3cbf | 231 | cancel_work_sync(&f->frags_work); |
d4ad4d22 | 232 | kmem_cache_destroy(f->frags_cachep); |
7eb95156 PE |
233 | } |
234 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 235 | |
81566e83 PE |
236 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
237 | { | |
ab1c724f | 238 | unsigned int seq; |
b13d3cbf FW |
239 | int i; |
240 | ||
81566e83 | 241 | nf->low_thresh = 0; |
ab1c724f | 242 | local_bh_disable(); |
e8e16b70 | 243 | |
ab1c724f FW |
244 | evict_again: |
245 | seq = read_seqbegin(&f->rnd_seqlock); | |
b13d3cbf FW |
246 | |
247 | for (i = 0; i < INETFRAGS_HASHSZ ; i++) | |
248 | inet_evict_bucket(f, &f->hash[i]); | |
249 | ||
ab1c724f FW |
250 | if (read_seqretry(&f->rnd_seqlock, seq)) |
251 | goto evict_again; | |
252 | ||
253 | local_bh_enable(); | |
6d7b857d JDB |
254 | |
255 | percpu_counter_destroy(&nf->mem); | |
81566e83 PE |
256 | } |
257 | EXPORT_SYMBOL(inet_frags_exit_net); | |
258 | ||
ab1c724f FW |
259 | static struct inet_frag_bucket * |
260 | get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f) | |
261 | __acquires(hb->chain_lock) | |
277e650d | 262 | { |
19952cc4 | 263 | struct inet_frag_bucket *hb; |
ab1c724f FW |
264 | unsigned int seq, hash; |
265 | ||
266 | restart: | |
267 | seq = read_seqbegin(&f->rnd_seqlock); | |
19952cc4 | 268 | |
fb3cfe6e | 269 | hash = inet_frag_hashfn(f, fq); |
19952cc4 JDB |
270 | hb = &f->hash[hash]; |
271 | ||
272 | spin_lock(&hb->chain_lock); | |
ab1c724f FW |
273 | if (read_seqretry(&f->rnd_seqlock, seq)) { |
274 | spin_unlock(&hb->chain_lock); | |
275 | goto restart; | |
276 | } | |
277 | ||
278 | return hb; | |
279 | } | |
280 | ||
281 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) | |
282 | { | |
283 | struct inet_frag_bucket *hb; | |
284 | ||
285 | hb = get_frag_bucket_locked(fq, f); | |
d1fe1944 | 286 | hlist_del(&fq->list); |
19952cc4 | 287 | spin_unlock(&hb->chain_lock); |
277e650d PE |
288 | } |
289 | ||
290 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
291 | { | |
292 | if (del_timer(&fq->timer)) | |
293 | atomic_dec(&fq->refcnt); | |
294 | ||
06aa8b8a | 295 | if (!(fq->flags & INET_FRAG_COMPLETE)) { |
277e650d PE |
296 | fq_unlink(fq, f); |
297 | atomic_dec(&fq->refcnt); | |
06aa8b8a | 298 | fq->flags |= INET_FRAG_COMPLETE; |
277e650d PE |
299 | } |
300 | } | |
277e650d | 301 | EXPORT_SYMBOL(inet_frag_kill); |
1e4b8287 | 302 | |
6ddc0822 | 303 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
06aa8b8a | 304 | struct sk_buff *skb) |
1e4b8287 | 305 | { |
1e4b8287 PE |
306 | if (f->skb_free) |
307 | f->skb_free(skb); | |
308 | kfree_skb(skb); | |
309 | } | |
310 | ||
3fd588eb | 311 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f) |
1e4b8287 PE |
312 | { |
313 | struct sk_buff *fp; | |
6ddc0822 | 314 | struct netns_frags *nf; |
d433673e | 315 | unsigned int sum, sum_truesize = 0; |
1e4b8287 | 316 | |
06aa8b8a | 317 | WARN_ON(!(q->flags & INET_FRAG_COMPLETE)); |
547b792c | 318 | WARN_ON(del_timer(&q->timer) != 0); |
1e4b8287 PE |
319 | |
320 | /* Release all fragment data. */ | |
321 | fp = q->fragments; | |
6ddc0822 | 322 | nf = q->net; |
1e4b8287 PE |
323 | while (fp) { |
324 | struct sk_buff *xp = fp->next; | |
325 | ||
d433673e JDB |
326 | sum_truesize += fp->truesize; |
327 | frag_kfree_skb(nf, f, fp); | |
1e4b8287 PE |
328 | fp = xp; |
329 | } | |
d433673e | 330 | sum = sum_truesize + f->qsize; |
d433673e | 331 | sub_frag_mem_limit(q, sum); |
1e4b8287 | 332 | |
c9547709 PE |
333 | if (f->destructor) |
334 | f->destructor(q); | |
d4ad4d22 | 335 | kmem_cache_free(f->frags_cachep, q); |
1e4b8287 PE |
336 | } |
337 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 338 | |
ac18e750 | 339 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
f926e236 NA |
340 | struct inet_frag_queue *qp_in, |
341 | struct inet_frags *f, | |
342 | void *arg) | |
2588fe1d | 343 | { |
ab1c724f | 344 | struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f); |
2588fe1d | 345 | struct inet_frag_queue *qp; |
19952cc4 | 346 | |
2588fe1d PE |
347 | #ifdef CONFIG_SMP |
348 | /* With SMP race we have to recheck hash table, because | |
ab1c724f FW |
349 | * such entry could have been created on other cpu before |
350 | * we acquired hash bucket lock. | |
2588fe1d | 351 | */ |
19952cc4 | 352 | hlist_for_each_entry(qp, &hb->chain, list) { |
ac18e750 | 353 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d | 354 | atomic_inc(&qp->refcnt); |
19952cc4 | 355 | spin_unlock(&hb->chain_lock); |
06aa8b8a | 356 | qp_in->flags |= INET_FRAG_COMPLETE; |
2588fe1d PE |
357 | inet_frag_put(qp_in, f); |
358 | return qp; | |
359 | } | |
360 | } | |
361 | #endif | |
362 | qp = qp_in; | |
b2fd5321 | 363 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
364 | atomic_inc(&qp->refcnt); |
365 | ||
366 | atomic_inc(&qp->refcnt); | |
19952cc4 | 367 | hlist_add_head(&qp->list, &hb->chain); |
3fd588eb | 368 | |
19952cc4 | 369 | spin_unlock(&hb->chain_lock); |
24b9bf43 | 370 | |
2588fe1d PE |
371 | return qp; |
372 | } | |
e521db9d | 373 | |
ac18e750 | 374 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
f926e236 NA |
375 | struct inet_frags *f, |
376 | void *arg) | |
e521db9d PE |
377 | { |
378 | struct inet_frag_queue *q; | |
379 | ||
b13d3cbf FW |
380 | if (frag_mem_limit(nf) > nf->high_thresh) { |
381 | inet_frag_schedule_worker(f); | |
86e93e47 | 382 | return NULL; |
b13d3cbf | 383 | } |
86e93e47 | 384 | |
d4ad4d22 | 385 | q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC); |
51456b29 | 386 | if (!q) |
e521db9d PE |
387 | return NULL; |
388 | ||
54db0cc2 | 389 | q->net = nf; |
c6fda282 | 390 | f->constructor(q, arg); |
d433673e JDB |
391 | add_frag_mem_limit(q, f->qsize); |
392 | ||
e521db9d PE |
393 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
394 | spin_lock_init(&q->lock); | |
395 | atomic_set(&q->refcnt, 1); | |
396 | ||
397 | return q; | |
398 | } | |
c6fda282 | 399 | |
ac18e750 | 400 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
f926e236 NA |
401 | struct inet_frags *f, |
402 | void *arg) | |
c6fda282 PE |
403 | { |
404 | struct inet_frag_queue *q; | |
405 | ||
ac18e750 | 406 | q = inet_frag_alloc(nf, f, arg); |
51456b29 | 407 | if (!q) |
c6fda282 PE |
408 | return NULL; |
409 | ||
9a375803 | 410 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 411 | } |
abd6523d | 412 | |
ac18e750 | 413 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
f926e236 NA |
414 | struct inet_frags *f, void *key, |
415 | unsigned int hash) | |
abd6523d | 416 | { |
19952cc4 | 417 | struct inet_frag_bucket *hb; |
abd6523d | 418 | struct inet_frag_queue *q; |
5a3da1fe | 419 | int depth = 0; |
abd6523d | 420 | |
b13d3cbf FW |
421 | if (frag_mem_limit(nf) > nf->low_thresh) |
422 | inet_frag_schedule_worker(f); | |
86e93e47 | 423 | |
fb3cfe6e | 424 | hash &= (INETFRAGS_HASHSZ - 1); |
19952cc4 JDB |
425 | hb = &f->hash[hash]; |
426 | ||
427 | spin_lock(&hb->chain_lock); | |
428 | hlist_for_each_entry(q, &hb->chain, list) { | |
ac18e750 | 429 | if (q->net == nf && f->match(q, key)) { |
abd6523d | 430 | atomic_inc(&q->refcnt); |
19952cc4 | 431 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
432 | return q; |
433 | } | |
5a3da1fe | 434 | depth++; |
abd6523d | 435 | } |
19952cc4 | 436 | spin_unlock(&hb->chain_lock); |
abd6523d | 437 | |
5a3da1fe HFS |
438 | if (depth <= INETFRAGS_MAXDEPTH) |
439 | return inet_frag_create(nf, f, key); | |
e3a57d18 FW |
440 | |
441 | if (inet_frag_may_rebuild(f)) { | |
ab1c724f FW |
442 | if (!f->rebuild) |
443 | f->rebuild = true; | |
e3a57d18 FW |
444 | inet_frag_schedule_worker(f); |
445 | } | |
446 | ||
447 | return ERR_PTR(-ENOBUFS); | |
abd6523d PE |
448 | } |
449 | EXPORT_SYMBOL(inet_frag_find); | |
5a3da1fe HFS |
450 | |
451 | void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, | |
452 | const char *prefix) | |
453 | { | |
454 | static const char msg[] = "inet_frag_find: Fragment hash bucket" | |
455 | " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) | |
456 | ". Dropping fragment.\n"; | |
457 | ||
458 | if (PTR_ERR(q) == -ENOBUFS) | |
ba7a46f1 | 459 | net_dbg_ratelimited("%s%s", prefix, msg); |
5a3da1fe HFS |
460 | } |
461 | EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); |