]>
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 | ||
28 | /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements | |
29 | * Value : 0xff if frame should be dropped. | |
30 | * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field | |
31 | */ | |
32 | const u8 ip_frag_ecn_table[16] = { | |
33 | /* at least one fragment had CE, and others ECT_0 or ECT_1 */ | |
34 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE, | |
35 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
36 | [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE, | |
37 | ||
38 | /* invalid combinations : drop frame */ | |
39 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff, | |
40 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff, | |
41 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff, | |
42 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
43 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff, | |
44 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff, | |
45 | [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, | |
46 | }; | |
47 | EXPORT_SYMBOL(ip_frag_ecn_table); | |
7eb95156 | 48 | |
321a3a99 PE |
49 | static void inet_frag_secret_rebuild(unsigned long dummy) |
50 | { | |
51 | struct inet_frags *f = (struct inet_frags *)dummy; | |
52 | unsigned long now = jiffies; | |
53 | int i; | |
54 | ||
19952cc4 | 55 | /* Per bucket lock NOT needed here, due to write lock protection */ |
321a3a99 | 56 | write_lock(&f->lock); |
19952cc4 | 57 | |
321a3a99 PE |
58 | get_random_bytes(&f->rnd, sizeof(u32)); |
59 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { | |
19952cc4 | 60 | struct inet_frag_bucket *hb; |
321a3a99 | 61 | struct inet_frag_queue *q; |
b67bfe0d | 62 | struct hlist_node *n; |
321a3a99 | 63 | |
19952cc4 JDB |
64 | hb = &f->hash[i]; |
65 | hlist_for_each_entry_safe(q, n, &hb->chain, list) { | |
321a3a99 PE |
66 | unsigned int hval = f->hashfn(q); |
67 | ||
68 | if (hval != i) { | |
19952cc4 JDB |
69 | struct inet_frag_bucket *hb_dest; |
70 | ||
321a3a99 PE |
71 | hlist_del(&q->list); |
72 | ||
73 | /* Relink to new hash chain. */ | |
19952cc4 JDB |
74 | hb_dest = &f->hash[hval]; |
75 | hlist_add_head(&q->list, &hb_dest->chain); | |
321a3a99 PE |
76 | } |
77 | } | |
78 | } | |
79 | write_unlock(&f->lock); | |
80 | ||
3b4bc4a2 | 81 | mod_timer(&f->secret_timer, now + f->secret_interval); |
321a3a99 PE |
82 | } |
83 | ||
7eb95156 PE |
84 | void inet_frags_init(struct inet_frags *f) |
85 | { | |
86 | int i; | |
87 | ||
19952cc4 JDB |
88 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { |
89 | struct inet_frag_bucket *hb = &f->hash[i]; | |
7eb95156 | 90 | |
19952cc4 JDB |
91 | spin_lock_init(&hb->chain_lock); |
92 | INIT_HLIST_HEAD(&hb->chain); | |
93 | } | |
7eb95156 PE |
94 | rwlock_init(&f->lock); |
95 | ||
b24b8a24 PE |
96 | setup_timer(&f->secret_timer, inet_frag_secret_rebuild, |
97 | (unsigned long)f); | |
3b4bc4a2 | 98 | f->secret_timer.expires = jiffies + f->secret_interval; |
321a3a99 | 99 | add_timer(&f->secret_timer); |
7eb95156 PE |
100 | } |
101 | EXPORT_SYMBOL(inet_frags_init); | |
102 | ||
e5a2bb84 PE |
103 | void inet_frags_init_net(struct netns_frags *nf) |
104 | { | |
105 | nf->nqueues = 0; | |
d433673e | 106 | init_frag_mem_limit(nf); |
3140c25c | 107 | INIT_LIST_HEAD(&nf->lru_list); |
3ef0eb0d | 108 | spin_lock_init(&nf->lru_lock); |
e5a2bb84 PE |
109 | } |
110 | EXPORT_SYMBOL(inet_frags_init_net); | |
111 | ||
7eb95156 PE |
112 | void inet_frags_fini(struct inet_frags *f) |
113 | { | |
321a3a99 | 114 | del_timer(&f->secret_timer); |
7eb95156 PE |
115 | } |
116 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 117 | |
81566e83 PE |
118 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
119 | { | |
120 | nf->low_thresh = 0; | |
e8e16b70 DM |
121 | |
122 | local_bh_disable(); | |
6b102865 | 123 | inet_frag_evictor(nf, f, true); |
e8e16b70 | 124 | local_bh_enable(); |
6d7b857d JDB |
125 | |
126 | percpu_counter_destroy(&nf->mem); | |
81566e83 PE |
127 | } |
128 | EXPORT_SYMBOL(inet_frags_exit_net); | |
129 | ||
277e650d PE |
130 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) |
131 | { | |
19952cc4 JDB |
132 | struct inet_frag_bucket *hb; |
133 | unsigned int hash; | |
134 | ||
135 | read_lock(&f->lock); | |
136 | hash = f->hashfn(fq); | |
137 | hb = &f->hash[hash]; | |
138 | ||
139 | spin_lock(&hb->chain_lock); | |
277e650d | 140 | hlist_del(&fq->list); |
19952cc4 JDB |
141 | spin_unlock(&hb->chain_lock); |
142 | ||
143 | read_unlock(&f->lock); | |
3ef0eb0d | 144 | inet_frag_lru_del(fq); |
277e650d PE |
145 | } |
146 | ||
147 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
148 | { | |
149 | if (del_timer(&fq->timer)) | |
150 | atomic_dec(&fq->refcnt); | |
151 | ||
bc578a54 | 152 | if (!(fq->last_in & INET_FRAG_COMPLETE)) { |
277e650d PE |
153 | fq_unlink(fq, f); |
154 | atomic_dec(&fq->refcnt); | |
bc578a54 | 155 | fq->last_in |= INET_FRAG_COMPLETE; |
277e650d PE |
156 | } |
157 | } | |
277e650d | 158 | EXPORT_SYMBOL(inet_frag_kill); |
1e4b8287 | 159 | |
6ddc0822 | 160 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
d433673e | 161 | struct sk_buff *skb) |
1e4b8287 | 162 | { |
1e4b8287 PE |
163 | if (f->skb_free) |
164 | f->skb_free(skb); | |
165 | kfree_skb(skb); | |
166 | } | |
167 | ||
168 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, | |
169 | int *work) | |
170 | { | |
171 | struct sk_buff *fp; | |
6ddc0822 | 172 | struct netns_frags *nf; |
d433673e | 173 | unsigned int sum, sum_truesize = 0; |
1e4b8287 | 174 | |
547b792c IJ |
175 | WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); |
176 | WARN_ON(del_timer(&q->timer) != 0); | |
1e4b8287 PE |
177 | |
178 | /* Release all fragment data. */ | |
179 | fp = q->fragments; | |
6ddc0822 | 180 | nf = q->net; |
1e4b8287 PE |
181 | while (fp) { |
182 | struct sk_buff *xp = fp->next; | |
183 | ||
d433673e JDB |
184 | sum_truesize += fp->truesize; |
185 | frag_kfree_skb(nf, f, fp); | |
1e4b8287 PE |
186 | fp = xp; |
187 | } | |
d433673e | 188 | sum = sum_truesize + f->qsize; |
1e4b8287 | 189 | if (work) |
d433673e JDB |
190 | *work -= sum; |
191 | sub_frag_mem_limit(q, sum); | |
1e4b8287 | 192 | |
c9547709 PE |
193 | if (f->destructor) |
194 | f->destructor(q); | |
195 | kfree(q); | |
1e4b8287 PE |
196 | |
197 | } | |
198 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 199 | |
6b102865 | 200 | int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force) |
8e7999c4 PE |
201 | { |
202 | struct inet_frag_queue *q; | |
203 | int work, evicted = 0; | |
204 | ||
6b102865 | 205 | if (!force) { |
d433673e | 206 | if (frag_mem_limit(nf) <= nf->high_thresh) |
6b102865 AW |
207 | return 0; |
208 | } | |
209 | ||
d433673e | 210 | work = frag_mem_limit(nf) - nf->low_thresh; |
8e7999c4 | 211 | while (work > 0) { |
3ef0eb0d JDB |
212 | spin_lock(&nf->lru_lock); |
213 | ||
3140c25c | 214 | if (list_empty(&nf->lru_list)) { |
3ef0eb0d | 215 | spin_unlock(&nf->lru_lock); |
8e7999c4 PE |
216 | break; |
217 | } | |
218 | ||
3140c25c | 219 | q = list_first_entry(&nf->lru_list, |
8e7999c4 PE |
220 | struct inet_frag_queue, lru_list); |
221 | atomic_inc(&q->refcnt); | |
68399ac3 JDB |
222 | /* Remove q from list to avoid several CPUs grabbing it */ |
223 | list_del_init(&q->lru_list); | |
224 | ||
3ef0eb0d | 225 | spin_unlock(&nf->lru_lock); |
8e7999c4 PE |
226 | |
227 | spin_lock(&q->lock); | |
bc578a54 | 228 | if (!(q->last_in & INET_FRAG_COMPLETE)) |
8e7999c4 PE |
229 | inet_frag_kill(q, f); |
230 | spin_unlock(&q->lock); | |
231 | ||
232 | if (atomic_dec_and_test(&q->refcnt)) | |
233 | inet_frag_destroy(q, f, &work); | |
234 | evicted++; | |
235 | } | |
236 | ||
237 | return evicted; | |
238 | } | |
239 | EXPORT_SYMBOL(inet_frag_evictor); | |
2588fe1d | 240 | |
ac18e750 PE |
241 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
242 | struct inet_frag_queue *qp_in, struct inet_frags *f, | |
9a375803 | 243 | void *arg) |
2588fe1d | 244 | { |
19952cc4 | 245 | struct inet_frag_bucket *hb; |
2588fe1d | 246 | struct inet_frag_queue *qp; |
9a375803 | 247 | unsigned int hash; |
2588fe1d | 248 | |
19952cc4 | 249 | read_lock(&f->lock); /* Protects against hash rebuild */ |
9a375803 PE |
250 | /* |
251 | * While we stayed w/o the lock other CPU could update | |
252 | * the rnd seed, so we need to re-calculate the hash | |
253 | * chain. Fortunatelly the qp_in can be used to get one. | |
254 | */ | |
255 | hash = f->hashfn(qp_in); | |
19952cc4 JDB |
256 | hb = &f->hash[hash]; |
257 | spin_lock(&hb->chain_lock); | |
258 | ||
2588fe1d PE |
259 | #ifdef CONFIG_SMP |
260 | /* With SMP race we have to recheck hash table, because | |
261 | * such entry could be created on other cpu, while we | |
19952cc4 | 262 | * released the hash bucket lock. |
2588fe1d | 263 | */ |
19952cc4 | 264 | hlist_for_each_entry(qp, &hb->chain, list) { |
ac18e750 | 265 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d | 266 | atomic_inc(&qp->refcnt); |
19952cc4 JDB |
267 | spin_unlock(&hb->chain_lock); |
268 | read_unlock(&f->lock); | |
bc578a54 | 269 | qp_in->last_in |= INET_FRAG_COMPLETE; |
2588fe1d PE |
270 | inet_frag_put(qp_in, f); |
271 | return qp; | |
272 | } | |
273 | } | |
274 | #endif | |
275 | qp = qp_in; | |
b2fd5321 | 276 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
277 | atomic_inc(&qp->refcnt); |
278 | ||
279 | atomic_inc(&qp->refcnt); | |
19952cc4 JDB |
280 | hlist_add_head(&qp->list, &hb->chain); |
281 | spin_unlock(&hb->chain_lock); | |
282 | read_unlock(&f->lock); | |
3ef0eb0d | 283 | inet_frag_lru_add(nf, qp); |
2588fe1d PE |
284 | return qp; |
285 | } | |
e521db9d | 286 | |
ac18e750 PE |
287 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
288 | struct inet_frags *f, void *arg) | |
e521db9d PE |
289 | { |
290 | struct inet_frag_queue *q; | |
291 | ||
292 | q = kzalloc(f->qsize, GFP_ATOMIC); | |
293 | if (q == NULL) | |
294 | return NULL; | |
295 | ||
54db0cc2 | 296 | q->net = nf; |
c6fda282 | 297 | f->constructor(q, arg); |
d433673e JDB |
298 | add_frag_mem_limit(q, f->qsize); |
299 | ||
e521db9d PE |
300 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
301 | spin_lock_init(&q->lock); | |
302 | atomic_set(&q->refcnt, 1); | |
b56141ab | 303 | INIT_LIST_HEAD(&q->lru_list); |
e521db9d PE |
304 | |
305 | return q; | |
306 | } | |
c6fda282 | 307 | |
ac18e750 | 308 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
9a375803 | 309 | struct inet_frags *f, void *arg) |
c6fda282 PE |
310 | { |
311 | struct inet_frag_queue *q; | |
312 | ||
ac18e750 | 313 | q = inet_frag_alloc(nf, f, arg); |
c6fda282 PE |
314 | if (q == NULL) |
315 | return NULL; | |
316 | ||
9a375803 | 317 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 318 | } |
abd6523d | 319 | |
ac18e750 PE |
320 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
321 | struct inet_frags *f, void *key, unsigned int hash) | |
56bca31f | 322 | __releases(&f->lock) |
abd6523d | 323 | { |
19952cc4 | 324 | struct inet_frag_bucket *hb; |
abd6523d | 325 | struct inet_frag_queue *q; |
5a3da1fe | 326 | int depth = 0; |
abd6523d | 327 | |
19952cc4 JDB |
328 | hb = &f->hash[hash]; |
329 | ||
330 | spin_lock(&hb->chain_lock); | |
331 | hlist_for_each_entry(q, &hb->chain, list) { | |
ac18e750 | 332 | if (q->net == nf && f->match(q, key)) { |
abd6523d | 333 | atomic_inc(&q->refcnt); |
19952cc4 | 334 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
335 | read_unlock(&f->lock); |
336 | return q; | |
337 | } | |
5a3da1fe | 338 | depth++; |
abd6523d | 339 | } |
19952cc4 | 340 | spin_unlock(&hb->chain_lock); |
abd6523d PE |
341 | read_unlock(&f->lock); |
342 | ||
5a3da1fe HFS |
343 | if (depth <= INETFRAGS_MAXDEPTH) |
344 | return inet_frag_create(nf, f, key); | |
345 | else | |
346 | return ERR_PTR(-ENOBUFS); | |
abd6523d PE |
347 | } |
348 | EXPORT_SYMBOL(inet_frag_find); | |
5a3da1fe HFS |
349 | |
350 | void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, | |
351 | const char *prefix) | |
352 | { | |
353 | static const char msg[] = "inet_frag_find: Fragment hash bucket" | |
354 | " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) | |
355 | ". Dropping fragment.\n"; | |
356 | ||
357 | if (PTR_ERR(q) == -ENOBUFS) | |
358 | LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg); | |
359 | } | |
360 | EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); |