]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - net/ipv4/inet_fragment.c
Merge branch 'merge-fixes' into devel
[mirror_ubuntu-artful-kernel.git] / net / ipv4 / inet_fragment.c
CommitLineData
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>
7eb95156
PE
22
23#include <net/inet_frag.h>
24
321a3a99
PE
25static void inet_frag_secret_rebuild(unsigned long dummy)
26{
27 struct inet_frags *f = (struct inet_frags *)dummy;
28 unsigned long now = jiffies;
29 int i;
30
31 write_lock(&f->lock);
32 get_random_bytes(&f->rnd, sizeof(u32));
33 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
34 struct inet_frag_queue *q;
35 struct hlist_node *p, *n;
36
37 hlist_for_each_entry_safe(q, p, n, &f->hash[i], list) {
38 unsigned int hval = f->hashfn(q);
39
40 if (hval != i) {
41 hlist_del(&q->list);
42
43 /* Relink to new hash chain. */
44 hlist_add_head(&q->list, &f->hash[hval]);
45 }
46 }
47 }
48 write_unlock(&f->lock);
49
3b4bc4a2 50 mod_timer(&f->secret_timer, now + f->secret_interval);
321a3a99
PE
51}
52
7eb95156
PE
53void inet_frags_init(struct inet_frags *f)
54{
55 int i;
56
57 for (i = 0; i < INETFRAGS_HASHSZ; i++)
58 INIT_HLIST_HEAD(&f->hash[i]);
59
7eb95156
PE
60 rwlock_init(&f->lock);
61
62 f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
63 (jiffies ^ (jiffies >> 6)));
64
b24b8a24
PE
65 setup_timer(&f->secret_timer, inet_frag_secret_rebuild,
66 (unsigned long)f);
3b4bc4a2 67 f->secret_timer.expires = jiffies + f->secret_interval;
321a3a99 68 add_timer(&f->secret_timer);
7eb95156
PE
69}
70EXPORT_SYMBOL(inet_frags_init);
71
e5a2bb84
PE
72void inet_frags_init_net(struct netns_frags *nf)
73{
74 nf->nqueues = 0;
6ddc0822 75 atomic_set(&nf->mem, 0);
3140c25c 76 INIT_LIST_HEAD(&nf->lru_list);
e5a2bb84
PE
77}
78EXPORT_SYMBOL(inet_frags_init_net);
79
7eb95156
PE
80void inet_frags_fini(struct inet_frags *f)
81{
321a3a99 82 del_timer(&f->secret_timer);
7eb95156
PE
83}
84EXPORT_SYMBOL(inet_frags_fini);
277e650d 85
81566e83
PE
86void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
87{
88 nf->low_thresh = 0;
e8e16b70
DM
89
90 local_bh_disable();
81566e83 91 inet_frag_evictor(nf, f);
e8e16b70 92 local_bh_enable();
81566e83
PE
93}
94EXPORT_SYMBOL(inet_frags_exit_net);
95
277e650d
PE
96static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
97{
98 write_lock(&f->lock);
99 hlist_del(&fq->list);
100 list_del(&fq->lru_list);
e5a2bb84 101 fq->net->nqueues--;
277e650d
PE
102 write_unlock(&f->lock);
103}
104
105void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
106{
107 if (del_timer(&fq->timer))
108 atomic_dec(&fq->refcnt);
109
bc578a54 110 if (!(fq->last_in & INET_FRAG_COMPLETE)) {
277e650d
PE
111 fq_unlink(fq, f);
112 atomic_dec(&fq->refcnt);
bc578a54 113 fq->last_in |= INET_FRAG_COMPLETE;
277e650d
PE
114 }
115}
116
117EXPORT_SYMBOL(inet_frag_kill);
1e4b8287 118
6ddc0822
PE
119static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
120 struct sk_buff *skb, int *work)
1e4b8287
PE
121{
122 if (work)
123 *work -= skb->truesize;
124
6ddc0822 125 atomic_sub(skb->truesize, &nf->mem);
1e4b8287
PE
126 if (f->skb_free)
127 f->skb_free(skb);
128 kfree_skb(skb);
129}
130
131void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
132 int *work)
133{
134 struct sk_buff *fp;
6ddc0822 135 struct netns_frags *nf;
1e4b8287 136
bc578a54 137 BUG_TRAP(q->last_in & INET_FRAG_COMPLETE);
1e4b8287
PE
138 BUG_TRAP(del_timer(&q->timer) == 0);
139
140 /* Release all fragment data. */
141 fp = q->fragments;
6ddc0822 142 nf = q->net;
1e4b8287
PE
143 while (fp) {
144 struct sk_buff *xp = fp->next;
145
6ddc0822 146 frag_kfree_skb(nf, f, fp, work);
1e4b8287
PE
147 fp = xp;
148 }
149
150 if (work)
151 *work -= f->qsize;
6ddc0822 152 atomic_sub(f->qsize, &nf->mem);
1e4b8287 153
c9547709
PE
154 if (f->destructor)
155 f->destructor(q);
156 kfree(q);
1e4b8287
PE
157
158}
159EXPORT_SYMBOL(inet_frag_destroy);
8e7999c4 160
6ddc0822 161int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f)
8e7999c4
PE
162{
163 struct inet_frag_queue *q;
164 int work, evicted = 0;
165
e31e0bdc 166 work = atomic_read(&nf->mem) - nf->low_thresh;
8e7999c4
PE
167 while (work > 0) {
168 read_lock(&f->lock);
3140c25c 169 if (list_empty(&nf->lru_list)) {
8e7999c4
PE
170 read_unlock(&f->lock);
171 break;
172 }
173
3140c25c 174 q = list_first_entry(&nf->lru_list,
8e7999c4
PE
175 struct inet_frag_queue, lru_list);
176 atomic_inc(&q->refcnt);
177 read_unlock(&f->lock);
178
179 spin_lock(&q->lock);
bc578a54 180 if (!(q->last_in & INET_FRAG_COMPLETE))
8e7999c4
PE
181 inet_frag_kill(q, f);
182 spin_unlock(&q->lock);
183
184 if (atomic_dec_and_test(&q->refcnt))
185 inet_frag_destroy(q, f, &work);
186 evicted++;
187 }
188
189 return evicted;
190}
191EXPORT_SYMBOL(inet_frag_evictor);
2588fe1d 192
ac18e750
PE
193static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
194 struct inet_frag_queue *qp_in, struct inet_frags *f,
195 unsigned int hash, void *arg)
2588fe1d
PE
196{
197 struct inet_frag_queue *qp;
198#ifdef CONFIG_SMP
199 struct hlist_node *n;
200#endif
201
202 write_lock(&f->lock);
203#ifdef CONFIG_SMP
204 /* With SMP race we have to recheck hash table, because
205 * such entry could be created on other cpu, while we
206 * promoted read lock to write lock.
207 */
208 hlist_for_each_entry(qp, n, &f->hash[hash], list) {
ac18e750 209 if (qp->net == nf && f->match(qp, arg)) {
2588fe1d
PE
210 atomic_inc(&qp->refcnt);
211 write_unlock(&f->lock);
bc578a54 212 qp_in->last_in |= INET_FRAG_COMPLETE;
2588fe1d
PE
213 inet_frag_put(qp_in, f);
214 return qp;
215 }
216 }
217#endif
218 qp = qp_in;
b2fd5321 219 if (!mod_timer(&qp->timer, jiffies + nf->timeout))
2588fe1d
PE
220 atomic_inc(&qp->refcnt);
221
222 atomic_inc(&qp->refcnt);
223 hlist_add_head(&qp->list, &f->hash[hash]);
3140c25c 224 list_add_tail(&qp->lru_list, &nf->lru_list);
e5a2bb84 225 nf->nqueues++;
2588fe1d
PE
226 write_unlock(&f->lock);
227 return qp;
228}
e521db9d 229
ac18e750
PE
230static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
231 struct inet_frags *f, void *arg)
e521db9d
PE
232{
233 struct inet_frag_queue *q;
234
235 q = kzalloc(f->qsize, GFP_ATOMIC);
236 if (q == NULL)
237 return NULL;
238
c6fda282 239 f->constructor(q, arg);
6ddc0822 240 atomic_add(f->qsize, &nf->mem);
e521db9d
PE
241 setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
242 spin_lock_init(&q->lock);
243 atomic_set(&q->refcnt, 1);
ac18e750 244 q->net = nf;
e521db9d
PE
245
246 return q;
247}
c6fda282 248
ac18e750
PE
249static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
250 struct inet_frags *f, void *arg, unsigned int hash)
c6fda282
PE
251{
252 struct inet_frag_queue *q;
253
ac18e750 254 q = inet_frag_alloc(nf, f, arg);
c6fda282
PE
255 if (q == NULL)
256 return NULL;
257
ac18e750 258 return inet_frag_intern(nf, q, f, hash, arg);
c6fda282 259}
abd6523d 260
ac18e750
PE
261struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
262 struct inet_frags *f, void *key, unsigned int hash)
abd6523d
PE
263{
264 struct inet_frag_queue *q;
265 struct hlist_node *n;
266
267 read_lock(&f->lock);
268 hlist_for_each_entry(q, n, &f->hash[hash], list) {
ac18e750 269 if (q->net == nf && f->match(q, key)) {
abd6523d
PE
270 atomic_inc(&q->refcnt);
271 read_unlock(&f->lock);
272 return q;
273 }
274 }
275 read_unlock(&f->lock);
276
ac18e750 277 return inet_frag_create(nf, f, key, hash);
abd6523d
PE
278}
279EXPORT_SYMBOL(inet_frag_find);