]>
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 PE |
23 | |
24 | #include <net/inet_frag.h> | |
25 | ||
321a3a99 PE |
26 | static void inet_frag_secret_rebuild(unsigned long dummy) |
27 | { | |
28 | struct inet_frags *f = (struct inet_frags *)dummy; | |
29 | unsigned long now = jiffies; | |
30 | int i; | |
31 | ||
32 | write_lock(&f->lock); | |
33 | get_random_bytes(&f->rnd, sizeof(u32)); | |
34 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { | |
35 | struct inet_frag_queue *q; | |
36 | struct hlist_node *p, *n; | |
37 | ||
38 | hlist_for_each_entry_safe(q, p, n, &f->hash[i], list) { | |
39 | unsigned int hval = f->hashfn(q); | |
40 | ||
41 | if (hval != i) { | |
42 | hlist_del(&q->list); | |
43 | ||
44 | /* Relink to new hash chain. */ | |
45 | hlist_add_head(&q->list, &f->hash[hval]); | |
46 | } | |
47 | } | |
48 | } | |
49 | write_unlock(&f->lock); | |
50 | ||
3b4bc4a2 | 51 | mod_timer(&f->secret_timer, now + f->secret_interval); |
321a3a99 PE |
52 | } |
53 | ||
7eb95156 PE |
54 | void inet_frags_init(struct inet_frags *f) |
55 | { | |
56 | int i; | |
57 | ||
58 | for (i = 0; i < INETFRAGS_HASHSZ; i++) | |
59 | INIT_HLIST_HEAD(&f->hash[i]); | |
60 | ||
7eb95156 PE |
61 | rwlock_init(&f->lock); |
62 | ||
63 | f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ | |
64 | (jiffies ^ (jiffies >> 6))); | |
65 | ||
b24b8a24 PE |
66 | setup_timer(&f->secret_timer, inet_frag_secret_rebuild, |
67 | (unsigned long)f); | |
3b4bc4a2 | 68 | f->secret_timer.expires = jiffies + f->secret_interval; |
321a3a99 | 69 | add_timer(&f->secret_timer); |
7eb95156 PE |
70 | } |
71 | EXPORT_SYMBOL(inet_frags_init); | |
72 | ||
e5a2bb84 PE |
73 | void inet_frags_init_net(struct netns_frags *nf) |
74 | { | |
75 | nf->nqueues = 0; | |
6ddc0822 | 76 | atomic_set(&nf->mem, 0); |
3140c25c | 77 | INIT_LIST_HEAD(&nf->lru_list); |
e5a2bb84 PE |
78 | } |
79 | EXPORT_SYMBOL(inet_frags_init_net); | |
80 | ||
7eb95156 PE |
81 | void inet_frags_fini(struct inet_frags *f) |
82 | { | |
321a3a99 | 83 | del_timer(&f->secret_timer); |
7eb95156 PE |
84 | } |
85 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d | 86 | |
81566e83 PE |
87 | void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) |
88 | { | |
89 | nf->low_thresh = 0; | |
e8e16b70 DM |
90 | |
91 | local_bh_disable(); | |
81566e83 | 92 | inet_frag_evictor(nf, f); |
e8e16b70 | 93 | local_bh_enable(); |
81566e83 PE |
94 | } |
95 | EXPORT_SYMBOL(inet_frags_exit_net); | |
96 | ||
277e650d PE |
97 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) |
98 | { | |
99 | write_lock(&f->lock); | |
100 | hlist_del(&fq->list); | |
101 | list_del(&fq->lru_list); | |
e5a2bb84 | 102 | fq->net->nqueues--; |
277e650d PE |
103 | write_unlock(&f->lock); |
104 | } | |
105 | ||
106 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
107 | { | |
108 | if (del_timer(&fq->timer)) | |
109 | atomic_dec(&fq->refcnt); | |
110 | ||
bc578a54 | 111 | if (!(fq->last_in & INET_FRAG_COMPLETE)) { |
277e650d PE |
112 | fq_unlink(fq, f); |
113 | atomic_dec(&fq->refcnt); | |
bc578a54 | 114 | fq->last_in |= INET_FRAG_COMPLETE; |
277e650d PE |
115 | } |
116 | } | |
117 | ||
118 | EXPORT_SYMBOL(inet_frag_kill); | |
1e4b8287 | 119 | |
6ddc0822 PE |
120 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
121 | struct sk_buff *skb, int *work) | |
1e4b8287 PE |
122 | { |
123 | if (work) | |
124 | *work -= skb->truesize; | |
125 | ||
6ddc0822 | 126 | atomic_sub(skb->truesize, &nf->mem); |
1e4b8287 PE |
127 | if (f->skb_free) |
128 | f->skb_free(skb); | |
129 | kfree_skb(skb); | |
130 | } | |
131 | ||
132 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, | |
133 | int *work) | |
134 | { | |
135 | struct sk_buff *fp; | |
6ddc0822 | 136 | struct netns_frags *nf; |
1e4b8287 | 137 | |
547b792c IJ |
138 | WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); |
139 | WARN_ON(del_timer(&q->timer) != 0); | |
1e4b8287 PE |
140 | |
141 | /* Release all fragment data. */ | |
142 | fp = q->fragments; | |
6ddc0822 | 143 | nf = q->net; |
1e4b8287 PE |
144 | while (fp) { |
145 | struct sk_buff *xp = fp->next; | |
146 | ||
6ddc0822 | 147 | frag_kfree_skb(nf, f, fp, work); |
1e4b8287 PE |
148 | fp = xp; |
149 | } | |
150 | ||
151 | if (work) | |
152 | *work -= f->qsize; | |
6ddc0822 | 153 | atomic_sub(f->qsize, &nf->mem); |
1e4b8287 | 154 | |
c9547709 PE |
155 | if (f->destructor) |
156 | f->destructor(q); | |
157 | kfree(q); | |
1e4b8287 PE |
158 | |
159 | } | |
160 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 161 | |
6ddc0822 | 162 | int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f) |
8e7999c4 PE |
163 | { |
164 | struct inet_frag_queue *q; | |
165 | int work, evicted = 0; | |
166 | ||
e31e0bdc | 167 | work = atomic_read(&nf->mem) - nf->low_thresh; |
8e7999c4 PE |
168 | while (work > 0) { |
169 | read_lock(&f->lock); | |
3140c25c | 170 | if (list_empty(&nf->lru_list)) { |
8e7999c4 PE |
171 | read_unlock(&f->lock); |
172 | break; | |
173 | } | |
174 | ||
3140c25c | 175 | q = list_first_entry(&nf->lru_list, |
8e7999c4 PE |
176 | struct inet_frag_queue, lru_list); |
177 | atomic_inc(&q->refcnt); | |
178 | read_unlock(&f->lock); | |
179 | ||
180 | spin_lock(&q->lock); | |
bc578a54 | 181 | if (!(q->last_in & INET_FRAG_COMPLETE)) |
8e7999c4 PE |
182 | inet_frag_kill(q, f); |
183 | spin_unlock(&q->lock); | |
184 | ||
185 | if (atomic_dec_and_test(&q->refcnt)) | |
186 | inet_frag_destroy(q, f, &work); | |
187 | evicted++; | |
188 | } | |
189 | ||
190 | return evicted; | |
191 | } | |
192 | EXPORT_SYMBOL(inet_frag_evictor); | |
2588fe1d | 193 | |
ac18e750 PE |
194 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
195 | struct inet_frag_queue *qp_in, struct inet_frags *f, | |
9a375803 | 196 | void *arg) |
2588fe1d PE |
197 | { |
198 | struct inet_frag_queue *qp; | |
199 | #ifdef CONFIG_SMP | |
200 | struct hlist_node *n; | |
201 | #endif | |
9a375803 | 202 | unsigned int hash; |
2588fe1d PE |
203 | |
204 | write_lock(&f->lock); | |
9a375803 PE |
205 | /* |
206 | * While we stayed w/o the lock other CPU could update | |
207 | * the rnd seed, so we need to re-calculate the hash | |
208 | * chain. Fortunatelly the qp_in can be used to get one. | |
209 | */ | |
210 | hash = f->hashfn(qp_in); | |
2588fe1d PE |
211 | #ifdef CONFIG_SMP |
212 | /* With SMP race we have to recheck hash table, because | |
213 | * such entry could be created on other cpu, while we | |
214 | * promoted read lock to write lock. | |
215 | */ | |
216 | hlist_for_each_entry(qp, n, &f->hash[hash], list) { | |
ac18e750 | 217 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d PE |
218 | atomic_inc(&qp->refcnt); |
219 | write_unlock(&f->lock); | |
bc578a54 | 220 | qp_in->last_in |= INET_FRAG_COMPLETE; |
2588fe1d PE |
221 | inet_frag_put(qp_in, f); |
222 | return qp; | |
223 | } | |
224 | } | |
225 | #endif | |
226 | qp = qp_in; | |
b2fd5321 | 227 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
228 | atomic_inc(&qp->refcnt); |
229 | ||
230 | atomic_inc(&qp->refcnt); | |
231 | hlist_add_head(&qp->list, &f->hash[hash]); | |
3140c25c | 232 | list_add_tail(&qp->lru_list, &nf->lru_list); |
e5a2bb84 | 233 | nf->nqueues++; |
2588fe1d PE |
234 | write_unlock(&f->lock); |
235 | return qp; | |
236 | } | |
e521db9d | 237 | |
ac18e750 PE |
238 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
239 | struct inet_frags *f, void *arg) | |
e521db9d PE |
240 | { |
241 | struct inet_frag_queue *q; | |
242 | ||
243 | q = kzalloc(f->qsize, GFP_ATOMIC); | |
244 | if (q == NULL) | |
245 | return NULL; | |
246 | ||
c6fda282 | 247 | f->constructor(q, arg); |
6ddc0822 | 248 | atomic_add(f->qsize, &nf->mem); |
e521db9d PE |
249 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
250 | spin_lock_init(&q->lock); | |
251 | atomic_set(&q->refcnt, 1); | |
ac18e750 | 252 | q->net = nf; |
e521db9d PE |
253 | |
254 | return q; | |
255 | } | |
c6fda282 | 256 | |
ac18e750 | 257 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
9a375803 | 258 | struct inet_frags *f, void *arg) |
c6fda282 PE |
259 | { |
260 | struct inet_frag_queue *q; | |
261 | ||
ac18e750 | 262 | q = inet_frag_alloc(nf, f, arg); |
c6fda282 PE |
263 | if (q == NULL) |
264 | return NULL; | |
265 | ||
9a375803 | 266 | return inet_frag_intern(nf, q, f, arg); |
c6fda282 | 267 | } |
abd6523d | 268 | |
ac18e750 PE |
269 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
270 | struct inet_frags *f, void *key, unsigned int hash) | |
56bca31f | 271 | __releases(&f->lock) |
abd6523d PE |
272 | { |
273 | struct inet_frag_queue *q; | |
274 | struct hlist_node *n; | |
275 | ||
abd6523d | 276 | hlist_for_each_entry(q, n, &f->hash[hash], list) { |
ac18e750 | 277 | if (q->net == nf && f->match(q, key)) { |
abd6523d PE |
278 | atomic_inc(&q->refcnt); |
279 | read_unlock(&f->lock); | |
280 | return q; | |
281 | } | |
282 | } | |
283 | read_unlock(&f->lock); | |
284 | ||
9a375803 | 285 | return inet_frag_create(nf, f, key); |
abd6523d PE |
286 | } |
287 | EXPORT_SYMBOL(inet_frag_find); |