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); |
e5a2bb84 PE |
197 | } |
198 | EXPORT_SYMBOL(inet_frags_init_net); | |
199 | ||
7eb95156 PE |
200 | void inet_frags_fini(struct inet_frags *f) |
201 | { | |
321a3a99 | 202 | del_timer(&f->secret_timer); |
b13d3cbf | 203 | cancel_work_sync(&f->frags_work); |
7eb95156 PE |
204 | } |
205 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 206 | |
81566e83 PE |
207 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
208 | { | |
b13d3cbf FW |
209 | int i; |
210 | ||
81566e83 | 211 | nf->low_thresh = 0; |
e8e16b70 | 212 | |
b13d3cbf FW |
213 | read_lock_bh(&f->lock); |
214 | ||
215 | for (i = 0; i < INETFRAGS_HASHSZ ; i++) | |
216 | inet_evict_bucket(f, &f->hash[i]); | |
217 | ||
218 | read_unlock_bh(&f->lock); | |
6d7b857d JDB |
219 | |
220 | percpu_counter_destroy(&nf->mem); | |
81566e83 PE |
221 | } |
222 | EXPORT_SYMBOL(inet_frags_exit_net); | |
223 | ||
277e650d PE |
224 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) |
225 | { | |
19952cc4 JDB |
226 | struct inet_frag_bucket *hb; |
227 | unsigned int hash; | |
228 | ||
229 | read_lock(&f->lock); | |
fb3cfe6e | 230 | hash = inet_frag_hashfn(f, fq); |
19952cc4 JDB |
231 | hb = &f->hash[hash]; |
232 | ||
233 | spin_lock(&hb->chain_lock); | |
277e650d | 234 | hlist_del(&fq->list); |
19952cc4 JDB |
235 | spin_unlock(&hb->chain_lock); |
236 | ||
237 | read_unlock(&f->lock); | |
277e650d PE |
238 | } |
239 | ||
240 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
241 | { | |
242 | if (del_timer(&fq->timer)) | |
243 | atomic_dec(&fq->refcnt); | |
244 | ||
bc578a54 | 245 | if (!(fq->last_in & INET_FRAG_COMPLETE)) { |
277e650d PE |
246 | fq_unlink(fq, f); |
247 | atomic_dec(&fq->refcnt); | |
bc578a54 | 248 | fq->last_in |= INET_FRAG_COMPLETE; |
277e650d PE |
249 | } |
250 | } | |
277e650d | 251 | EXPORT_SYMBOL(inet_frag_kill); |
1e4b8287 | 252 | |
6ddc0822 | 253 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
d433673e | 254 | struct sk_buff *skb) |
1e4b8287 | 255 | { |
1e4b8287 PE |
256 | if (f->skb_free) |
257 | f->skb_free(skb); | |
258 | kfree_skb(skb); | |
259 | } | |
260 | ||
3fd588eb | 261 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f) |
1e4b8287 PE |
262 | { |
263 | struct sk_buff *fp; | |
6ddc0822 | 264 | struct netns_frags *nf; |
d433673e | 265 | unsigned int sum, sum_truesize = 0; |
1e4b8287 | 266 | |
547b792c IJ |
267 | WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); |
268 | WARN_ON(del_timer(&q->timer) != 0); | |
1e4b8287 PE |
269 | |
270 | /* Release all fragment data. */ | |
271 | fp = q->fragments; | |
6ddc0822 | 272 | nf = q->net; |
1e4b8287 PE |
273 | while (fp) { |
274 | struct sk_buff *xp = fp->next; | |
275 | ||
d433673e JDB |
276 | sum_truesize += fp->truesize; |
277 | frag_kfree_skb(nf, f, fp); | |
1e4b8287 PE |
278 | fp = xp; |
279 | } | |
d433673e | 280 | sum = sum_truesize + f->qsize; |
d433673e | 281 | sub_frag_mem_limit(q, sum); |
1e4b8287 | 282 | |
c9547709 PE |
283 | if (f->destructor) |
284 | f->destructor(q); | |
285 | kfree(q); | |
1e4b8287 PE |
286 | } |
287 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 288 | |
ac18e750 PE |
289 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
290 | struct inet_frag_queue *qp_in, struct inet_frags *f, | |
9a375803 | 291 | void *arg) |
2588fe1d | 292 | { |
19952cc4 | 293 | struct inet_frag_bucket *hb; |
2588fe1d | 294 | struct inet_frag_queue *qp; |
9a375803 | 295 | unsigned int hash; |
2588fe1d | 296 | |
19952cc4 | 297 | read_lock(&f->lock); /* Protects against hash rebuild */ |
9a375803 PE |
298 | /* |
299 | * While we stayed w/o the lock other CPU could update | |
300 | * the rnd seed, so we need to re-calculate the hash | |
301 | * chain. Fortunatelly the qp_in can be used to get one. | |
302 | */ | |
fb3cfe6e | 303 | hash = inet_frag_hashfn(f, qp_in); |
19952cc4 JDB |
304 | hb = &f->hash[hash]; |
305 | spin_lock(&hb->chain_lock); | |
306 | ||
2588fe1d PE |
307 | #ifdef CONFIG_SMP |
308 | /* With SMP race we have to recheck hash table, because | |
309 | * such entry could be created on other cpu, while we | |
19952cc4 | 310 | * released the hash bucket lock. |
2588fe1d | 311 | */ |
19952cc4 | 312 | hlist_for_each_entry(qp, &hb->chain, list) { |
ac18e750 | 313 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d | 314 | atomic_inc(&qp->refcnt); |
19952cc4 JDB |
315 | spin_unlock(&hb->chain_lock); |
316 | read_unlock(&f->lock); | |
bc578a54 | 317 | qp_in->last_in |= INET_FRAG_COMPLETE; |
2588fe1d PE |
318 | inet_frag_put(qp_in, f); |
319 | return qp; | |
320 | } | |
321 | } | |
322 | #endif | |
323 | qp = qp_in; | |
b2fd5321 | 324 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
325 | atomic_inc(&qp->refcnt); |
326 | ||
327 | atomic_inc(&qp->refcnt); | |
19952cc4 | 328 | hlist_add_head(&qp->list, &hb->chain); |
3fd588eb | 329 | |
19952cc4 JDB |
330 | spin_unlock(&hb->chain_lock); |
331 | read_unlock(&f->lock); | |
24b9bf43 | 332 | |
2588fe1d PE |
333 | return qp; |
334 | } | |
e521db9d | 335 | |
ac18e750 PE |
336 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
337 | struct inet_frags *f, void *arg) | |
e521db9d PE |
338 | { |
339 | struct inet_frag_queue *q; | |
340 | ||
b13d3cbf FW |
341 | if (frag_mem_limit(nf) > nf->high_thresh) { |
342 | inet_frag_schedule_worker(f); | |
86e93e47 | 343 | return NULL; |
b13d3cbf | 344 | } |
86e93e47 | 345 | |
e521db9d PE |
346 | q = kzalloc(f->qsize, GFP_ATOMIC); |
347 | if (q == NULL) | |
348 | return NULL; | |
349 | ||
54db0cc2 | 350 | q->net = nf; |
c6fda282 | 351 | f->constructor(q, arg); |
d433673e JDB |
352 | add_frag_mem_limit(q, f->qsize); |
353 | ||
e521db9d PE |
354 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
355 | spin_lock_init(&q->lock); | |
356 | atomic_set(&q->refcnt, 1); | |
357 | ||
358 | return q; | |
359 | } | |
c6fda282 | 360 | |
ac18e750 | 361 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
9a375803 | 362 | struct inet_frags *f, void *arg) |
c6fda282 PE |
363 | { |
364 | struct inet_frag_queue *q; | |
365 | ||
ac18e750 | 366 | q = inet_frag_alloc(nf, f, arg); |
c6fda282 PE |
367 | if (q == NULL) |
368 | return NULL; | |
369 | ||
9a375803 | 370 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 371 | } |
abd6523d | 372 | |
ac18e750 PE |
373 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
374 | struct inet_frags *f, void *key, unsigned int hash) | |
56bca31f | 375 | __releases(&f->lock) |
abd6523d | 376 | { |
19952cc4 | 377 | struct inet_frag_bucket *hb; |
abd6523d | 378 | struct inet_frag_queue *q; |
5a3da1fe | 379 | int depth = 0; |
abd6523d | 380 | |
b13d3cbf FW |
381 | if (frag_mem_limit(nf) > nf->low_thresh) |
382 | inet_frag_schedule_worker(f); | |
86e93e47 | 383 | |
fb3cfe6e | 384 | hash &= (INETFRAGS_HASHSZ - 1); |
19952cc4 JDB |
385 | hb = &f->hash[hash]; |
386 | ||
387 | spin_lock(&hb->chain_lock); | |
388 | hlist_for_each_entry(q, &hb->chain, list) { | |
ac18e750 | 389 | if (q->net == nf && f->match(q, key)) { |
abd6523d | 390 | atomic_inc(&q->refcnt); |
19952cc4 | 391 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
392 | read_unlock(&f->lock); |
393 | return q; | |
394 | } | |
5a3da1fe | 395 | depth++; |
abd6523d | 396 | } |
19952cc4 | 397 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
398 | read_unlock(&f->lock); |
399 | ||
5a3da1fe HFS |
400 | if (depth <= INETFRAGS_MAXDEPTH) |
401 | return inet_frag_create(nf, f, key); | |
402 | else | |
403 | return ERR_PTR(-ENOBUFS); | |
abd6523d PE |
404 | } |
405 | EXPORT_SYMBOL(inet_frag_find); | |
5a3da1fe HFS |
406 | |
407 | void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, | |
408 | const char *prefix) | |
409 | { | |
410 | static const char msg[] = "inet_frag_find: Fragment hash bucket" | |
411 | " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) | |
412 | ". Dropping fragment.\n"; | |
413 | ||
414 | if (PTR_ERR(q) == -ENOBUFS) | |
415 | LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg); | |
416 | } | |
417 | EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); |