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 | ||
55 | write_lock(&f->lock); | |
56 | get_random_bytes(&f->rnd, sizeof(u32)); | |
57 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { | |
58 | struct inet_frag_queue *q; | |
b67bfe0d | 59 | struct hlist_node *n; |
321a3a99 | 60 | |
b67bfe0d | 61 | hlist_for_each_entry_safe(q, n, &f->hash[i], list) { |
321a3a99 PE |
62 | unsigned int hval = f->hashfn(q); |
63 | ||
64 | if (hval != i) { | |
65 | hlist_del(&q->list); | |
66 | ||
67 | /* Relink to new hash chain. */ | |
68 | hlist_add_head(&q->list, &f->hash[hval]); | |
69 | } | |
70 | } | |
71 | } | |
72 | write_unlock(&f->lock); | |
73 | ||
3b4bc4a2 | 74 | mod_timer(&f->secret_timer, now + f->secret_interval); |
321a3a99 PE |
75 | } |
76 | ||
7eb95156 PE |
77 | void inet_frags_init(struct inet_frags *f) |
78 | { | |
79 | int i; | |
80 | ||
81 | for (i = 0; i < INETFRAGS_HASHSZ; i++) | |
82 | INIT_HLIST_HEAD(&f->hash[i]); | |
83 | ||
7eb95156 PE |
84 | rwlock_init(&f->lock); |
85 | ||
86 | f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ | |
87 | (jiffies ^ (jiffies >> 6))); | |
88 | ||
b24b8a24 PE |
89 | setup_timer(&f->secret_timer, inet_frag_secret_rebuild, |
90 | (unsigned long)f); | |
3b4bc4a2 | 91 | f->secret_timer.expires = jiffies + f->secret_interval; |
321a3a99 | 92 | add_timer(&f->secret_timer); |
7eb95156 PE |
93 | } |
94 | EXPORT_SYMBOL(inet_frags_init); | |
95 | ||
e5a2bb84 PE |
96 | void inet_frags_init_net(struct netns_frags *nf) |
97 | { | |
98 | nf->nqueues = 0; | |
d433673e | 99 | init_frag_mem_limit(nf); |
3140c25c | 100 | INIT_LIST_HEAD(&nf->lru_list); |
3ef0eb0d | 101 | spin_lock_init(&nf->lru_lock); |
e5a2bb84 PE |
102 | } |
103 | EXPORT_SYMBOL(inet_frags_init_net); | |
104 | ||
7eb95156 PE |
105 | void inet_frags_fini(struct inet_frags *f) |
106 | { | |
321a3a99 | 107 | del_timer(&f->secret_timer); |
7eb95156 PE |
108 | } |
109 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 110 | |
81566e83 PE |
111 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
112 | { | |
113 | nf->low_thresh = 0; | |
e8e16b70 DM |
114 | |
115 | local_bh_disable(); | |
6b102865 | 116 | inet_frag_evictor(nf, f, true); |
e8e16b70 | 117 | local_bh_enable(); |
6d7b857d JDB |
118 | |
119 | percpu_counter_destroy(&nf->mem); | |
81566e83 PE |
120 | } |
121 | EXPORT_SYMBOL(inet_frags_exit_net); | |
122 | ||
277e650d PE |
123 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) |
124 | { | |
125 | write_lock(&f->lock); | |
126 | hlist_del(&fq->list); | |
e5a2bb84 | 127 | fq->net->nqueues--; |
277e650d | 128 | write_unlock(&f->lock); |
3ef0eb0d | 129 | inet_frag_lru_del(fq); |
277e650d PE |
130 | } |
131 | ||
132 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
133 | { | |
134 | if (del_timer(&fq->timer)) | |
135 | atomic_dec(&fq->refcnt); | |
136 | ||
bc578a54 | 137 | if (!(fq->last_in & INET_FRAG_COMPLETE)) { |
277e650d PE |
138 | fq_unlink(fq, f); |
139 | atomic_dec(&fq->refcnt); | |
bc578a54 | 140 | fq->last_in |= INET_FRAG_COMPLETE; |
277e650d PE |
141 | } |
142 | } | |
277e650d | 143 | EXPORT_SYMBOL(inet_frag_kill); |
1e4b8287 | 144 | |
6ddc0822 | 145 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
d433673e | 146 | struct sk_buff *skb) |
1e4b8287 | 147 | { |
1e4b8287 PE |
148 | if (f->skb_free) |
149 | f->skb_free(skb); | |
150 | kfree_skb(skb); | |
151 | } | |
152 | ||
153 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, | |
154 | int *work) | |
155 | { | |
156 | struct sk_buff *fp; | |
6ddc0822 | 157 | struct netns_frags *nf; |
d433673e | 158 | unsigned int sum, sum_truesize = 0; |
1e4b8287 | 159 | |
547b792c IJ |
160 | WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); |
161 | WARN_ON(del_timer(&q->timer) != 0); | |
1e4b8287 PE |
162 | |
163 | /* Release all fragment data. */ | |
164 | fp = q->fragments; | |
6ddc0822 | 165 | nf = q->net; |
1e4b8287 PE |
166 | while (fp) { |
167 | struct sk_buff *xp = fp->next; | |
168 | ||
d433673e JDB |
169 | sum_truesize += fp->truesize; |
170 | frag_kfree_skb(nf, f, fp); | |
1e4b8287 PE |
171 | fp = xp; |
172 | } | |
d433673e | 173 | sum = sum_truesize + f->qsize; |
1e4b8287 | 174 | if (work) |
d433673e JDB |
175 | *work -= sum; |
176 | sub_frag_mem_limit(q, sum); | |
1e4b8287 | 177 | |
c9547709 PE |
178 | if (f->destructor) |
179 | f->destructor(q); | |
180 | kfree(q); | |
1e4b8287 PE |
181 | |
182 | } | |
183 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 184 | |
6b102865 | 185 | int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force) |
8e7999c4 PE |
186 | { |
187 | struct inet_frag_queue *q; | |
188 | int work, evicted = 0; | |
189 | ||
6b102865 | 190 | if (!force) { |
d433673e | 191 | if (frag_mem_limit(nf) <= nf->high_thresh) |
6b102865 AW |
192 | return 0; |
193 | } | |
194 | ||
d433673e | 195 | work = frag_mem_limit(nf) - nf->low_thresh; |
8e7999c4 | 196 | while (work > 0) { |
3ef0eb0d JDB |
197 | spin_lock(&nf->lru_lock); |
198 | ||
3140c25c | 199 | if (list_empty(&nf->lru_list)) { |
3ef0eb0d | 200 | spin_unlock(&nf->lru_lock); |
8e7999c4 PE |
201 | break; |
202 | } | |
203 | ||
3140c25c | 204 | q = list_first_entry(&nf->lru_list, |
8e7999c4 PE |
205 | struct inet_frag_queue, lru_list); |
206 | atomic_inc(&q->refcnt); | |
3ef0eb0d | 207 | spin_unlock(&nf->lru_lock); |
8e7999c4 PE |
208 | |
209 | spin_lock(&q->lock); | |
bc578a54 | 210 | if (!(q->last_in & INET_FRAG_COMPLETE)) |
8e7999c4 PE |
211 | inet_frag_kill(q, f); |
212 | spin_unlock(&q->lock); | |
213 | ||
214 | if (atomic_dec_and_test(&q->refcnt)) | |
215 | inet_frag_destroy(q, f, &work); | |
216 | evicted++; | |
217 | } | |
218 | ||
219 | return evicted; | |
220 | } | |
221 | EXPORT_SYMBOL(inet_frag_evictor); | |
2588fe1d | 222 | |
ac18e750 PE |
223 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
224 | struct inet_frag_queue *qp_in, struct inet_frags *f, | |
9a375803 | 225 | void *arg) |
2588fe1d PE |
226 | { |
227 | struct inet_frag_queue *qp; | |
228 | #ifdef CONFIG_SMP | |
2588fe1d | 229 | #endif |
9a375803 | 230 | unsigned int hash; |
2588fe1d PE |
231 | |
232 | write_lock(&f->lock); | |
9a375803 PE |
233 | /* |
234 | * While we stayed w/o the lock other CPU could update | |
235 | * the rnd seed, so we need to re-calculate the hash | |
236 | * chain. Fortunatelly the qp_in can be used to get one. | |
237 | */ | |
238 | hash = f->hashfn(qp_in); | |
2588fe1d PE |
239 | #ifdef CONFIG_SMP |
240 | /* With SMP race we have to recheck hash table, because | |
241 | * such entry could be created on other cpu, while we | |
242 | * promoted read lock to write lock. | |
243 | */ | |
b67bfe0d | 244 | hlist_for_each_entry(qp, &f->hash[hash], list) { |
ac18e750 | 245 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d PE |
246 | atomic_inc(&qp->refcnt); |
247 | write_unlock(&f->lock); | |
bc578a54 | 248 | qp_in->last_in |= INET_FRAG_COMPLETE; |
2588fe1d PE |
249 | inet_frag_put(qp_in, f); |
250 | return qp; | |
251 | } | |
252 | } | |
253 | #endif | |
254 | qp = qp_in; | |
b2fd5321 | 255 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
256 | atomic_inc(&qp->refcnt); |
257 | ||
258 | atomic_inc(&qp->refcnt); | |
259 | hlist_add_head(&qp->list, &f->hash[hash]); | |
e5a2bb84 | 260 | nf->nqueues++; |
2588fe1d | 261 | write_unlock(&f->lock); |
3ef0eb0d | 262 | inet_frag_lru_add(nf, qp); |
2588fe1d PE |
263 | return qp; |
264 | } | |
e521db9d | 265 | |
ac18e750 PE |
266 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
267 | struct inet_frags *f, void *arg) | |
e521db9d PE |
268 | { |
269 | struct inet_frag_queue *q; | |
270 | ||
271 | q = kzalloc(f->qsize, GFP_ATOMIC); | |
272 | if (q == NULL) | |
273 | return NULL; | |
274 | ||
54db0cc2 | 275 | q->net = nf; |
c6fda282 | 276 | f->constructor(q, arg); |
d433673e JDB |
277 | add_frag_mem_limit(q, f->qsize); |
278 | ||
e521db9d PE |
279 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
280 | spin_lock_init(&q->lock); | |
281 | atomic_set(&q->refcnt, 1); | |
282 | ||
283 | return q; | |
284 | } | |
c6fda282 | 285 | |
ac18e750 | 286 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
9a375803 | 287 | struct inet_frags *f, void *arg) |
c6fda282 PE |
288 | { |
289 | struct inet_frag_queue *q; | |
290 | ||
ac18e750 | 291 | q = inet_frag_alloc(nf, f, arg); |
c6fda282 PE |
292 | if (q == NULL) |
293 | return NULL; | |
294 | ||
9a375803 | 295 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 296 | } |
abd6523d | 297 | |
ac18e750 PE |
298 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
299 | struct inet_frags *f, void *key, unsigned int hash) | |
56bca31f | 300 | __releases(&f->lock) |
abd6523d PE |
301 | { |
302 | struct inet_frag_queue *q; | |
5a3da1fe | 303 | int depth = 0; |
abd6523d | 304 | |
b67bfe0d | 305 | hlist_for_each_entry(q, &f->hash[hash], list) { |
ac18e750 | 306 | if (q->net == nf && f->match(q, key)) { |
abd6523d PE |
307 | atomic_inc(&q->refcnt); |
308 | read_unlock(&f->lock); | |
309 | return q; | |
310 | } | |
5a3da1fe | 311 | depth++; |
abd6523d PE |
312 | } |
313 | read_unlock(&f->lock); | |
314 | ||
5a3da1fe HFS |
315 | if (depth <= INETFRAGS_MAXDEPTH) |
316 | return inet_frag_create(nf, f, key); | |
317 | else | |
318 | return ERR_PTR(-ENOBUFS); | |
abd6523d PE |
319 | } |
320 | EXPORT_SYMBOL(inet_frag_find); | |
5a3da1fe HFS |
321 | |
322 | void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, | |
323 | const char *prefix) | |
324 | { | |
325 | static const char msg[] = "inet_frag_find: Fragment hash bucket" | |
326 | " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH) | |
327 | ". Dropping fragment.\n"; | |
328 | ||
329 | if (PTR_ERR(q) == -ENOBUFS) | |
330 | LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg); | |
331 | } | |
332 | EXPORT_SYMBOL(inet_frag_maybe_warn_overflow); |