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>
22 #include <linux/slab.h>
25 #include <net/inet_frag.h>
26 #include <net/inet_ecn.h>
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
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
,
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,
47 EXPORT_SYMBOL(ip_frag_ecn_table
);
49 static void inet_frag_secret_rebuild(unsigned long dummy
)
51 struct inet_frags
*f
= (struct inet_frags
*)dummy
;
52 unsigned long now
= jiffies
;
56 get_random_bytes(&f
->rnd
, sizeof(u32
));
57 for (i
= 0; i
< INETFRAGS_HASHSZ
; i
++) {
58 struct inet_frag_queue
*q
;
61 hlist_for_each_entry_safe(q
, n
, &f
->hash
[i
], list
) {
62 unsigned int hval
= f
->hashfn(q
);
67 /* Relink to new hash chain. */
68 hlist_add_head(&q
->list
, &f
->hash
[hval
]);
72 write_unlock(&f
->lock
);
74 mod_timer(&f
->secret_timer
, now
+ f
->secret_interval
);
77 void inet_frags_init(struct inet_frags
*f
)
81 for (i
= 0; i
< INETFRAGS_HASHSZ
; i
++)
82 INIT_HLIST_HEAD(&f
->hash
[i
]);
84 rwlock_init(&f
->lock
);
86 f
->rnd
= (u32
) ((num_physpages
^ (num_physpages
>>7)) ^
87 (jiffies
^ (jiffies
>> 6)));
89 setup_timer(&f
->secret_timer
, inet_frag_secret_rebuild
,
91 f
->secret_timer
.expires
= jiffies
+ f
->secret_interval
;
92 add_timer(&f
->secret_timer
);
94 EXPORT_SYMBOL(inet_frags_init
);
96 void inet_frags_init_net(struct netns_frags
*nf
)
99 init_frag_mem_limit(nf
);
100 INIT_LIST_HEAD(&nf
->lru_list
);
101 spin_lock_init(&nf
->lru_lock
);
103 EXPORT_SYMBOL(inet_frags_init_net
);
105 void inet_frags_fini(struct inet_frags
*f
)
107 del_timer(&f
->secret_timer
);
109 EXPORT_SYMBOL(inet_frags_fini
);
111 void inet_frags_exit_net(struct netns_frags
*nf
, struct inet_frags
*f
)
116 inet_frag_evictor(nf
, f
, true);
119 percpu_counter_destroy(&nf
->mem
);
121 EXPORT_SYMBOL(inet_frags_exit_net
);
123 static inline void fq_unlink(struct inet_frag_queue
*fq
, struct inet_frags
*f
)
125 write_lock(&f
->lock
);
126 hlist_del(&fq
->list
);
128 write_unlock(&f
->lock
);
129 inet_frag_lru_del(fq
);
132 void inet_frag_kill(struct inet_frag_queue
*fq
, struct inet_frags
*f
)
134 if (del_timer(&fq
->timer
))
135 atomic_dec(&fq
->refcnt
);
137 if (!(fq
->last_in
& INET_FRAG_COMPLETE
)) {
139 atomic_dec(&fq
->refcnt
);
140 fq
->last_in
|= INET_FRAG_COMPLETE
;
143 EXPORT_SYMBOL(inet_frag_kill
);
145 static inline void frag_kfree_skb(struct netns_frags
*nf
, struct inet_frags
*f
,
153 void inet_frag_destroy(struct inet_frag_queue
*q
, struct inet_frags
*f
,
157 struct netns_frags
*nf
;
158 unsigned int sum
, sum_truesize
= 0;
160 WARN_ON(!(q
->last_in
& INET_FRAG_COMPLETE
));
161 WARN_ON(del_timer(&q
->timer
) != 0);
163 /* Release all fragment data. */
167 struct sk_buff
*xp
= fp
->next
;
169 sum_truesize
+= fp
->truesize
;
170 frag_kfree_skb(nf
, f
, fp
);
173 sum
= sum_truesize
+ f
->qsize
;
176 sub_frag_mem_limit(q
, sum
);
183 EXPORT_SYMBOL(inet_frag_destroy
);
185 int inet_frag_evictor(struct netns_frags
*nf
, struct inet_frags
*f
, bool force
)
187 struct inet_frag_queue
*q
;
188 int work
, evicted
= 0;
191 if (frag_mem_limit(nf
) <= nf
->high_thresh
)
195 work
= frag_mem_limit(nf
) - nf
->low_thresh
;
197 spin_lock(&nf
->lru_lock
);
199 if (list_empty(&nf
->lru_list
)) {
200 spin_unlock(&nf
->lru_lock
);
204 q
= list_first_entry(&nf
->lru_list
,
205 struct inet_frag_queue
, lru_list
);
206 atomic_inc(&q
->refcnt
);
207 spin_unlock(&nf
->lru_lock
);
210 if (!(q
->last_in
& INET_FRAG_COMPLETE
))
211 inet_frag_kill(q
, f
);
212 spin_unlock(&q
->lock
);
214 if (atomic_dec_and_test(&q
->refcnt
))
215 inet_frag_destroy(q
, f
, &work
);
221 EXPORT_SYMBOL(inet_frag_evictor
);
223 static struct inet_frag_queue
*inet_frag_intern(struct netns_frags
*nf
,
224 struct inet_frag_queue
*qp_in
, struct inet_frags
*f
,
227 struct inet_frag_queue
*qp
;
232 write_lock(&f
->lock
);
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.
238 hash
= f
->hashfn(qp_in
);
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.
244 hlist_for_each_entry(qp
, &f
->hash
[hash
], list
) {
245 if (qp
->net
== nf
&& f
->match(qp
, arg
)) {
246 atomic_inc(&qp
->refcnt
);
247 write_unlock(&f
->lock
);
248 qp_in
->last_in
|= INET_FRAG_COMPLETE
;
249 inet_frag_put(qp_in
, f
);
255 if (!mod_timer(&qp
->timer
, jiffies
+ nf
->timeout
))
256 atomic_inc(&qp
->refcnt
);
258 atomic_inc(&qp
->refcnt
);
259 hlist_add_head(&qp
->list
, &f
->hash
[hash
]);
261 write_unlock(&f
->lock
);
262 inet_frag_lru_add(nf
, qp
);
266 static struct inet_frag_queue
*inet_frag_alloc(struct netns_frags
*nf
,
267 struct inet_frags
*f
, void *arg
)
269 struct inet_frag_queue
*q
;
271 q
= kzalloc(f
->qsize
, GFP_ATOMIC
);
276 f
->constructor(q
, arg
);
277 add_frag_mem_limit(q
, f
->qsize
);
279 setup_timer(&q
->timer
, f
->frag_expire
, (unsigned long)q
);
280 spin_lock_init(&q
->lock
);
281 atomic_set(&q
->refcnt
, 1);
286 static struct inet_frag_queue
*inet_frag_create(struct netns_frags
*nf
,
287 struct inet_frags
*f
, void *arg
)
289 struct inet_frag_queue
*q
;
291 q
= inet_frag_alloc(nf
, f
, arg
);
295 return inet_frag_intern(nf
, q
, f
, arg
);
298 struct inet_frag_queue
*inet_frag_find(struct netns_frags
*nf
,
299 struct inet_frags
*f
, void *key
, unsigned int hash
)
302 struct inet_frag_queue
*q
;
305 hlist_for_each_entry(q
, &f
->hash
[hash
], list
) {
306 if (q
->net
== nf
&& f
->match(q
, key
)) {
307 atomic_inc(&q
->refcnt
);
308 read_unlock(&f
->lock
);
313 read_unlock(&f
->lock
);
315 if (depth
<= INETFRAGS_MAXDEPTH
)
316 return inet_frag_create(nf
, f
, key
);
318 return ERR_PTR(-ENOBUFS
);
320 EXPORT_SYMBOL(inet_frag_find
);
322 void inet_frag_maybe_warn_overflow(struct inet_frag_queue
*q
,
325 static const char msg
[] = "inet_frag_find: Fragment hash bucket"
326 " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH
)
327 ". Dropping fragment.\n";
329 if (PTR_ERR(q
) == -ENOBUFS
)
330 LIMIT_NETDEBUG(KERN_WARNING
"%s%s", prefix
, msg
);
332 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow
);