2 * inet fragments management
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.
9 * Authors: Pavel Emelyanov <xemul@openvz.org>
10 * Started as consolidation of ipv4/ip_fragment.c,
11 * ipv6/reassembly. and ipv6 nf conntrack reassembly
14 #include <linux/list.h>
15 #include <linux/spinlock.h>
16 #include <linux/module.h>
17 #include <linux/timer.h>
19 #include <linux/random.h>
20 #include <linux/skbuff.h>
21 #include <linux/rtnetlink.h>
23 #include <net/inet_frag.h>
25 static void inet_frag_secret_rebuild(unsigned long dummy
)
27 struct inet_frags
*f
= (struct inet_frags
*)dummy
;
28 unsigned long now
= jiffies
;
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
;
37 hlist_for_each_entry_safe(q
, p
, n
, &f
->hash
[i
], list
) {
38 unsigned int hval
= f
->hashfn(q
);
43 /* Relink to new hash chain. */
44 hlist_add_head(&q
->list
, &f
->hash
[hval
]);
48 write_unlock(&f
->lock
);
50 mod_timer(&f
->secret_timer
, now
+ f
->secret_interval
);
53 void inet_frags_init(struct inet_frags
*f
)
57 for (i
= 0; i
< INETFRAGS_HASHSZ
; i
++)
58 INIT_HLIST_HEAD(&f
->hash
[i
]);
60 rwlock_init(&f
->lock
);
62 f
->rnd
= (u32
) ((num_physpages
^ (num_physpages
>>7)) ^
63 (jiffies
^ (jiffies
>> 6)));
65 setup_timer(&f
->secret_timer
, inet_frag_secret_rebuild
,
67 f
->secret_timer
.expires
= jiffies
+ f
->secret_interval
;
68 add_timer(&f
->secret_timer
);
70 EXPORT_SYMBOL(inet_frags_init
);
72 void inet_frags_init_net(struct netns_frags
*nf
)
75 atomic_set(&nf
->mem
, 0);
76 INIT_LIST_HEAD(&nf
->lru_list
);
78 EXPORT_SYMBOL(inet_frags_init_net
);
80 void inet_frags_fini(struct inet_frags
*f
)
82 del_timer(&f
->secret_timer
);
84 EXPORT_SYMBOL(inet_frags_fini
);
86 void inet_frags_exit_net(struct netns_frags
*nf
, struct inet_frags
*f
)
91 inet_frag_evictor(nf
, f
);
94 EXPORT_SYMBOL(inet_frags_exit_net
);
96 static inline void fq_unlink(struct inet_frag_queue
*fq
, struct inet_frags
*f
)
100 list_del(&fq
->lru_list
);
102 write_unlock(&f
->lock
);
105 void inet_frag_kill(struct inet_frag_queue
*fq
, struct inet_frags
*f
)
107 if (del_timer(&fq
->timer
))
108 atomic_dec(&fq
->refcnt
);
110 if (!(fq
->last_in
& INET_FRAG_COMPLETE
)) {
112 atomic_dec(&fq
->refcnt
);
113 fq
->last_in
|= INET_FRAG_COMPLETE
;
117 EXPORT_SYMBOL(inet_frag_kill
);
119 static inline void frag_kfree_skb(struct netns_frags
*nf
, struct inet_frags
*f
,
120 struct sk_buff
*skb
, int *work
)
123 *work
-= skb
->truesize
;
125 atomic_sub(skb
->truesize
, &nf
->mem
);
131 void inet_frag_destroy(struct inet_frag_queue
*q
, struct inet_frags
*f
,
135 struct netns_frags
*nf
;
137 WARN_ON(!(q
->last_in
& INET_FRAG_COMPLETE
));
138 WARN_ON(del_timer(&q
->timer
) != 0);
140 /* Release all fragment data. */
144 struct sk_buff
*xp
= fp
->next
;
146 frag_kfree_skb(nf
, f
, fp
, work
);
152 atomic_sub(f
->qsize
, &nf
->mem
);
159 EXPORT_SYMBOL(inet_frag_destroy
);
161 int inet_frag_evictor(struct netns_frags
*nf
, struct inet_frags
*f
)
163 struct inet_frag_queue
*q
;
164 int work
, evicted
= 0;
166 work
= atomic_read(&nf
->mem
) - nf
->low_thresh
;
169 if (list_empty(&nf
->lru_list
)) {
170 read_unlock(&f
->lock
);
174 q
= list_first_entry(&nf
->lru_list
,
175 struct inet_frag_queue
, lru_list
);
176 atomic_inc(&q
->refcnt
);
177 read_unlock(&f
->lock
);
180 if (!(q
->last_in
& INET_FRAG_COMPLETE
))
181 inet_frag_kill(q
, f
);
182 spin_unlock(&q
->lock
);
184 if (atomic_dec_and_test(&q
->refcnt
))
185 inet_frag_destroy(q
, f
, &work
);
191 EXPORT_SYMBOL(inet_frag_evictor
);
193 static struct inet_frag_queue
*inet_frag_intern(struct netns_frags
*nf
,
194 struct inet_frag_queue
*qp_in
, struct inet_frags
*f
,
197 struct inet_frag_queue
*qp
;
199 struct hlist_node
*n
;
203 write_lock(&f
->lock
);
205 * While we stayed w/o the lock other CPU could update
206 * the rnd seed, so we need to re-calculate the hash
207 * chain. Fortunatelly the qp_in can be used to get one.
209 hash
= f
->hashfn(qp_in
);
211 /* With SMP race we have to recheck hash table, because
212 * such entry could be created on other cpu, while we
213 * promoted read lock to write lock.
215 hlist_for_each_entry(qp
, n
, &f
->hash
[hash
], list
) {
216 if (qp
->net
== nf
&& f
->match(qp
, arg
)) {
217 atomic_inc(&qp
->refcnt
);
218 write_unlock(&f
->lock
);
219 qp_in
->last_in
|= INET_FRAG_COMPLETE
;
220 inet_frag_put(qp_in
, f
);
226 if (!mod_timer(&qp
->timer
, jiffies
+ nf
->timeout
))
227 atomic_inc(&qp
->refcnt
);
229 atomic_inc(&qp
->refcnt
);
230 hlist_add_head(&qp
->list
, &f
->hash
[hash
]);
231 list_add_tail(&qp
->lru_list
, &nf
->lru_list
);
233 write_unlock(&f
->lock
);
237 static struct inet_frag_queue
*inet_frag_alloc(struct netns_frags
*nf
,
238 struct inet_frags
*f
, void *arg
)
240 struct inet_frag_queue
*q
;
242 q
= kzalloc(f
->qsize
, GFP_ATOMIC
);
246 f
->constructor(q
, arg
);
247 atomic_add(f
->qsize
, &nf
->mem
);
248 setup_timer(&q
->timer
, f
->frag_expire
, (unsigned long)q
);
249 spin_lock_init(&q
->lock
);
250 atomic_set(&q
->refcnt
, 1);
256 static struct inet_frag_queue
*inet_frag_create(struct netns_frags
*nf
,
257 struct inet_frags
*f
, void *arg
)
259 struct inet_frag_queue
*q
;
261 q
= inet_frag_alloc(nf
, f
, arg
);
265 return inet_frag_intern(nf
, q
, f
, arg
);
268 struct inet_frag_queue
*inet_frag_find(struct netns_frags
*nf
,
269 struct inet_frags
*f
, void *key
, unsigned int hash
)
272 struct inet_frag_queue
*q
;
273 struct hlist_node
*n
;
275 hlist_for_each_entry(q
, n
, &f
->hash
[hash
], list
) {
276 if (q
->net
== nf
&& f
->match(q
, key
)) {
277 atomic_inc(&q
->refcnt
);
278 read_unlock(&f
->lock
);
282 read_unlock(&f
->lock
);
284 return inet_frag_create(nf
, f
, key
);
286 EXPORT_SYMBOL(inet_frag_find
);