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 | ||
b13d3cbf FW |
134 | spin_lock(&hb->chain_lock); |
135 | ||
136 | hlist_for_each_entry_safe(fq, n, &hb->chain, list) { | |
137 | if (!inet_fragq_should_evict(fq)) | |
138 | continue; | |
139 | ||
5719b296 FW |
140 | if (!del_timer(&fq->timer)) |
141 | continue; | |
b13d3cbf | 142 | |
d1fe1944 | 143 | hlist_add_head(&fq->list_evictor, &expired); |
b13d3cbf FW |
144 | ++evicted; |
145 | } | |
146 | ||
147 | spin_unlock(&hb->chain_lock); | |
148 | ||
d1fe1944 | 149 | hlist_for_each_entry_safe(fq, n, &expired, list_evictor) |
b13d3cbf FW |
150 | f->frag_expire((unsigned long) fq); |
151 | ||
152 | return evicted; | |
153 | } | |
154 | ||
155 | static void inet_frag_worker(struct work_struct *work) | |
156 | { | |
157 | unsigned int budget = INETFRAGS_EVICT_BUCKETS; | |
158 | unsigned int i, evicted = 0; | |
159 | struct inet_frags *f; | |
160 | ||
161 | f = container_of(work, struct inet_frags, frags_work); | |
162 | ||
163 | BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ); | |
164 | ||
ab1c724f | 165 | local_bh_disable(); |
b13d3cbf FW |
166 | |
167 | for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) { | |
168 | evicted += inet_evict_bucket(f, &f->hash[i]); | |
169 | i = (i + 1) & (INETFRAGS_HASHSZ - 1); | |
170 | if (evicted > INETFRAGS_EVICT_MAX) | |
171 | break; | |
172 | } | |
173 | ||
174 | f->next_bucket = i; | |
175 | ||
ab1c724f FW |
176 | local_bh_enable(); |
177 | ||
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 | ||
d4ad4d22 | 188 | int inet_frags_init(struct inet_frags *f) |
7eb95156 PE |
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 | } | |
ab1c724f FW |
200 | |
201 | seqlock_init(&f->rnd_seqlock); | |
e3a57d18 | 202 | f->last_rebuild_jiffies = 0; |
d4ad4d22 NA |
203 | f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0, |
204 | NULL); | |
205 | if (!f->frags_cachep) | |
206 | return -ENOMEM; | |
207 | ||
208 | return 0; | |
7eb95156 PE |
209 | } |
210 | EXPORT_SYMBOL(inet_frags_init); | |
211 | ||
212 | void inet_frags_fini(struct inet_frags *f) | |
213 | { | |
b13d3cbf | 214 | cancel_work_sync(&f->frags_work); |
d4ad4d22 | 215 | kmem_cache_destroy(f->frags_cachep); |
7eb95156 PE |
216 | } |
217 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 218 | |
81566e83 PE |
219 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
220 | { | |
ab1c724f | 221 | unsigned int seq; |
b13d3cbf FW |
222 | int i; |
223 | ||
81566e83 | 224 | nf->low_thresh = 0; |
e8e16b70 | 225 | |
ab1c724f | 226 | evict_again: |
5719b296 | 227 | local_bh_disable(); |
ab1c724f | 228 | seq = read_seqbegin(&f->rnd_seqlock); |
b13d3cbf FW |
229 | |
230 | for (i = 0; i < INETFRAGS_HASHSZ ; i++) | |
231 | inet_evict_bucket(f, &f->hash[i]); | |
232 | ||
ab1c724f | 233 | local_bh_enable(); |
5719b296 FW |
234 | cond_resched(); |
235 | ||
236 | if (read_seqretry(&f->rnd_seqlock, seq) || | |
237 | percpu_counter_sum(&nf->mem)) | |
238 | goto evict_again; | |
6d7b857d JDB |
239 | |
240 | percpu_counter_destroy(&nf->mem); | |
81566e83 PE |
241 | } |
242 | EXPORT_SYMBOL(inet_frags_exit_net); | |
243 | ||
ab1c724f FW |
244 | static struct inet_frag_bucket * |
245 | get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f) | |
246 | __acquires(hb->chain_lock) | |
277e650d | 247 | { |
19952cc4 | 248 | struct inet_frag_bucket *hb; |
ab1c724f FW |
249 | unsigned int seq, hash; |
250 | ||
251 | restart: | |
252 | seq = read_seqbegin(&f->rnd_seqlock); | |
19952cc4 | 253 | |
fb3cfe6e | 254 | hash = inet_frag_hashfn(f, fq); |
19952cc4 JDB |
255 | hb = &f->hash[hash]; |
256 | ||
257 | spin_lock(&hb->chain_lock); | |
ab1c724f FW |
258 | if (read_seqretry(&f->rnd_seqlock, seq)) { |
259 | spin_unlock(&hb->chain_lock); | |
260 | goto restart; | |
261 | } | |
262 | ||
263 | return hb; | |
264 | } | |
265 | ||
266 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) | |
267 | { | |
268 | struct inet_frag_bucket *hb; | |
269 | ||
270 | hb = get_frag_bucket_locked(fq, f); | |
d1fe1944 | 271 | hlist_del(&fq->list); |
5719b296 | 272 | fq->flags |= INET_FRAG_COMPLETE; |
19952cc4 | 273 | spin_unlock(&hb->chain_lock); |
277e650d PE |
274 | } |
275 | ||
276 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
277 | { | |
278 | if (del_timer(&fq->timer)) | |
279 | atomic_dec(&fq->refcnt); | |
280 | ||
06aa8b8a | 281 | if (!(fq->flags & INET_FRAG_COMPLETE)) { |
277e650d PE |
282 | fq_unlink(fq, f); |
283 | atomic_dec(&fq->refcnt); | |
277e650d PE |
284 | } |
285 | } | |
277e650d | 286 | EXPORT_SYMBOL(inet_frag_kill); |
1e4b8287 | 287 | |
3fd588eb | 288 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f) |
1e4b8287 PE |
289 | { |
290 | struct sk_buff *fp; | |
6ddc0822 | 291 | struct netns_frags *nf; |
d433673e | 292 | unsigned int sum, sum_truesize = 0; |
1e4b8287 | 293 | |
06aa8b8a | 294 | WARN_ON(!(q->flags & INET_FRAG_COMPLETE)); |
547b792c | 295 | WARN_ON(del_timer(&q->timer) != 0); |
1e4b8287 PE |
296 | |
297 | /* Release all fragment data. */ | |
298 | fp = q->fragments; | |
6ddc0822 | 299 | nf = q->net; |
1e4b8287 PE |
300 | while (fp) { |
301 | struct sk_buff *xp = fp->next; | |
302 | ||
d433673e | 303 | sum_truesize += fp->truesize; |
a72a5e2d | 304 | kfree_skb(fp); |
1e4b8287 PE |
305 | fp = xp; |
306 | } | |
d433673e | 307 | sum = sum_truesize + f->qsize; |
1e4b8287 | 308 | |
c9547709 PE |
309 | if (f->destructor) |
310 | f->destructor(q); | |
d4ad4d22 | 311 | kmem_cache_free(f->frags_cachep, q); |
5719b296 FW |
312 | |
313 | sub_frag_mem_limit(nf, sum); | |
1e4b8287 PE |
314 | } |
315 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 316 | |
ac18e750 | 317 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
f926e236 NA |
318 | struct inet_frag_queue *qp_in, |
319 | struct inet_frags *f, | |
320 | void *arg) | |
2588fe1d | 321 | { |
ab1c724f | 322 | struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f); |
2588fe1d | 323 | struct inet_frag_queue *qp; |
19952cc4 | 324 | |
2588fe1d PE |
325 | #ifdef CONFIG_SMP |
326 | /* With SMP race we have to recheck hash table, because | |
ab1c724f FW |
327 | * such entry could have been created on other cpu before |
328 | * we acquired hash bucket lock. | |
2588fe1d | 329 | */ |
19952cc4 | 330 | hlist_for_each_entry(qp, &hb->chain, list) { |
ac18e750 | 331 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d | 332 | atomic_inc(&qp->refcnt); |
19952cc4 | 333 | spin_unlock(&hb->chain_lock); |
06aa8b8a | 334 | qp_in->flags |= INET_FRAG_COMPLETE; |
2588fe1d PE |
335 | inet_frag_put(qp_in, f); |
336 | return qp; | |
337 | } | |
338 | } | |
339 | #endif | |
340 | qp = qp_in; | |
b2fd5321 | 341 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
342 | atomic_inc(&qp->refcnt); |
343 | ||
344 | atomic_inc(&qp->refcnt); | |
19952cc4 | 345 | hlist_add_head(&qp->list, &hb->chain); |
3fd588eb | 346 | |
19952cc4 | 347 | spin_unlock(&hb->chain_lock); |
24b9bf43 | 348 | |
2588fe1d PE |
349 | return qp; |
350 | } | |
e521db9d | 351 | |
ac18e750 | 352 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
f926e236 NA |
353 | struct inet_frags *f, |
354 | void *arg) | |
e521db9d PE |
355 | { |
356 | struct inet_frag_queue *q; | |
357 | ||
b13d3cbf FW |
358 | if (frag_mem_limit(nf) > nf->high_thresh) { |
359 | inet_frag_schedule_worker(f); | |
86e93e47 | 360 | return NULL; |
b13d3cbf | 361 | } |
86e93e47 | 362 | |
d4ad4d22 | 363 | q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC); |
51456b29 | 364 | if (!q) |
e521db9d PE |
365 | return NULL; |
366 | ||
54db0cc2 | 367 | q->net = nf; |
c6fda282 | 368 | f->constructor(q, arg); |
0e60d245 | 369 | add_frag_mem_limit(nf, f->qsize); |
d433673e | 370 | |
e521db9d PE |
371 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
372 | spin_lock_init(&q->lock); | |
373 | atomic_set(&q->refcnt, 1); | |
374 | ||
375 | return q; | |
376 | } | |
c6fda282 | 377 | |
ac18e750 | 378 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
f926e236 NA |
379 | struct inet_frags *f, |
380 | void *arg) | |
c6fda282 PE |
381 | { |
382 | struct inet_frag_queue *q; | |
383 | ||
ac18e750 | 384 | q = inet_frag_alloc(nf, f, arg); |
51456b29 | 385 | if (!q) |
c6fda282 PE |
386 | return NULL; |
387 | ||
9a375803 | 388 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 389 | } |
abd6523d | 390 | |
ac18e750 | 391 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
f926e236 NA |
392 | struct inet_frags *f, void *key, |
393 | unsigned int hash) | |
abd6523d | 394 | { |
19952cc4 | 395 | struct inet_frag_bucket *hb; |
abd6523d | 396 | struct inet_frag_queue *q; |
5a3da1fe | 397 | int depth = 0; |
abd6523d | 398 | |
b13d3cbf FW |
399 | if (frag_mem_limit(nf) > nf->low_thresh) |
400 | inet_frag_schedule_worker(f); | |
86e93e47 | 401 | |
fb3cfe6e | 402 | hash &= (INETFRAGS_HASHSZ - 1); |
19952cc4 JDB |
403 | hb = &f->hash[hash]; |
404 | ||
405 | spin_lock(&hb->chain_lock); | |
406 | hlist_for_each_entry(q, &hb->chain, list) { | |
ac18e750 | 407 | if (q->net == nf && f->match(q, key)) { |
abd6523d | 408 | atomic_inc(&q->refcnt); |
19952cc4 | 409 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
410 | return q; |
411 | } | |
5a3da1fe | 412 | depth++; |
abd6523d | 413 | } |
19952cc4 | 414 | spin_unlock(&hb->chain_lock); |
abd6523d | 415 | |
5a3da1fe HFS |
416 | if (depth <= INETFRAGS_MAXDEPTH) |
417 | return inet_frag_create(nf, f, key); | |
e3a57d18 FW |
418 | |
419 | if (inet_frag_may_rebuild(f)) { | |
ab1c724f FW |
420 | if (!f->rebuild) |
421 | f->rebuild = true; | |
e3a57d18 FW |
422 | inet_frag_schedule_worker(f); |
423 | } | |
424 | ||
425 | return ERR_PTR(-ENOBUFS); | |
abd6523d PE |
426 | } |
427 | EXPORT_SYMBOL(inet_frag_find); | |
5a3da1fe HFS |
428 | |
429 | void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, | |
430 | const char *prefix) | |
431 | { | |
432 | static const char msg[] = "inet_frag_find: Fragment hash bucket" | |
433 | " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) | |
434 | ". Dropping fragment.\n"; | |
435 | ||
436 | if (PTR_ERR(q) == -ENOBUFS) | |
ba7a46f1 | 437 | net_dbg_ratelimited("%s%s", prefix, msg); |
5a3da1fe HFS |
438 | } |
439 | EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); |