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 #define INETFRAGS_EVICT_BUCKETS 128
29 #define INETFRAGS_EVICT_MAX 512
31 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
32 * Value : 0xff if frame should be dropped.
33 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
35 const u8 ip_frag_ecn_table
[16] = {
36 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
37 [IPFRAG_ECN_CE
| IPFRAG_ECN_ECT_0
] = INET_ECN_CE
,
38 [IPFRAG_ECN_CE
| IPFRAG_ECN_ECT_1
] = INET_ECN_CE
,
39 [IPFRAG_ECN_CE
| IPFRAG_ECN_ECT_0
| IPFRAG_ECN_ECT_1
] = INET_ECN_CE
,
41 /* invalid combinations : drop frame */
42 [IPFRAG_ECN_NOT_ECT
| IPFRAG_ECN_CE
] = 0xff,
43 [IPFRAG_ECN_NOT_ECT
| IPFRAG_ECN_ECT_0
] = 0xff,
44 [IPFRAG_ECN_NOT_ECT
| IPFRAG_ECN_ECT_1
] = 0xff,
45 [IPFRAG_ECN_NOT_ECT
| IPFRAG_ECN_ECT_0
| IPFRAG_ECN_ECT_1
] = 0xff,
46 [IPFRAG_ECN_NOT_ECT
| IPFRAG_ECN_CE
| IPFRAG_ECN_ECT_0
] = 0xff,
47 [IPFRAG_ECN_NOT_ECT
| IPFRAG_ECN_CE
| IPFRAG_ECN_ECT_1
] = 0xff,
48 [IPFRAG_ECN_NOT_ECT
| IPFRAG_ECN_CE
| IPFRAG_ECN_ECT_0
| IPFRAG_ECN_ECT_1
] = 0xff,
50 EXPORT_SYMBOL(ip_frag_ecn_table
);
53 inet_frag_hashfn(const struct inet_frags
*f
, const struct inet_frag_queue
*q
)
55 return f
->hashfn(q
) & (INETFRAGS_HASHSZ
- 1);
58 static void inet_frag_secret_rebuild(unsigned long dummy
)
60 struct inet_frags
*f
= (struct inet_frags
*)dummy
;
61 unsigned long now
= jiffies
;
64 /* Per bucket lock NOT needed here, due to write lock protection */
67 get_random_bytes(&f
->rnd
, sizeof(u32
));
68 for (i
= 0; i
< INETFRAGS_HASHSZ
; i
++) {
69 struct inet_frag_bucket
*hb
;
70 struct inet_frag_queue
*q
;
74 hlist_for_each_entry_safe(q
, n
, &hb
->chain
, list
) {
75 unsigned int hval
= inet_frag_hashfn(f
, q
);
78 struct inet_frag_bucket
*hb_dest
;
82 /* Relink to new hash chain. */
83 hb_dest
= &f
->hash
[hval
];
84 hlist_add_head(&q
->list
, &hb_dest
->chain
);
88 write_unlock(&f
->lock
);
90 mod_timer(&f
->secret_timer
, now
+ f
->secret_interval
);
93 static bool inet_fragq_should_evict(const struct inet_frag_queue
*q
)
95 return q
->net
->low_thresh
== 0 ||
96 frag_mem_limit(q
->net
) >= q
->net
->low_thresh
;
100 inet_evict_bucket(struct inet_frags
*f
, struct inet_frag_bucket
*hb
)
102 struct inet_frag_queue
*fq
;
103 struct hlist_node
*n
;
104 unsigned int evicted
= 0;
108 spin_lock(&hb
->chain_lock
);
110 hlist_for_each_entry_safe(fq
, n
, &hb
->chain
, list
) {
111 if (!inet_fragq_should_evict(fq
))
114 if (!del_timer(&fq
->timer
)) {
115 /* q expiring right now thus increment its refcount so
116 * it won't be freed under us and wait until the timer
117 * has finished executing then destroy it
119 atomic_inc(&fq
->refcnt
);
120 spin_unlock(&hb
->chain_lock
);
121 del_timer_sync(&fq
->timer
);
122 WARN_ON(atomic_read(&fq
->refcnt
) != 1);
123 inet_frag_put(fq
, f
);
127 /* suppress xmit of (icmp) error packet */
128 fq
->last_in
&= ~INET_FRAG_FIRST_IN
;
129 fq
->last_in
|= INET_FRAG_EVICTED
;
130 hlist_del(&fq
->list
);
131 hlist_add_head(&fq
->list
, &expired
);
135 spin_unlock(&hb
->chain_lock
);
137 hlist_for_each_entry_safe(fq
, n
, &expired
, list
)
138 f
->frag_expire((unsigned long) fq
);
143 static void inet_frag_worker(struct work_struct
*work
)
145 unsigned int budget
= INETFRAGS_EVICT_BUCKETS
;
146 unsigned int i
, evicted
= 0;
147 struct inet_frags
*f
;
149 f
= container_of(work
, struct inet_frags
, frags_work
);
151 BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS
>= INETFRAGS_HASHSZ
);
153 read_lock_bh(&f
->lock
);
155 for (i
= ACCESS_ONCE(f
->next_bucket
); budget
; --budget
) {
156 evicted
+= inet_evict_bucket(f
, &f
->hash
[i
]);
157 i
= (i
+ 1) & (INETFRAGS_HASHSZ
- 1);
158 if (evicted
> INETFRAGS_EVICT_MAX
)
164 read_unlock_bh(&f
->lock
);
167 static void inet_frag_schedule_worker(struct inet_frags
*f
)
169 if (unlikely(!work_pending(&f
->frags_work
)))
170 schedule_work(&f
->frags_work
);
173 void inet_frags_init(struct inet_frags
*f
)
177 INIT_WORK(&f
->frags_work
, inet_frag_worker
);
179 for (i
= 0; i
< INETFRAGS_HASHSZ
; i
++) {
180 struct inet_frag_bucket
*hb
= &f
->hash
[i
];
182 spin_lock_init(&hb
->chain_lock
);
183 INIT_HLIST_HEAD(&hb
->chain
);
185 rwlock_init(&f
->lock
);
187 setup_timer(&f
->secret_timer
, inet_frag_secret_rebuild
,
189 f
->secret_timer
.expires
= jiffies
+ f
->secret_interval
;
190 add_timer(&f
->secret_timer
);
192 EXPORT_SYMBOL(inet_frags_init
);
194 void inet_frags_init_net(struct netns_frags
*nf
)
196 init_frag_mem_limit(nf
);
198 EXPORT_SYMBOL(inet_frags_init_net
);
200 void inet_frags_fini(struct inet_frags
*f
)
202 del_timer(&f
->secret_timer
);
203 cancel_work_sync(&f
->frags_work
);
205 EXPORT_SYMBOL(inet_frags_fini
);
207 void inet_frags_exit_net(struct netns_frags
*nf
, struct inet_frags
*f
)
213 read_lock_bh(&f
->lock
);
215 for (i
= 0; i
< INETFRAGS_HASHSZ
; i
++)
216 inet_evict_bucket(f
, &f
->hash
[i
]);
218 read_unlock_bh(&f
->lock
);
220 percpu_counter_destroy(&nf
->mem
);
222 EXPORT_SYMBOL(inet_frags_exit_net
);
224 static inline void fq_unlink(struct inet_frag_queue
*fq
, struct inet_frags
*f
)
226 struct inet_frag_bucket
*hb
;
230 hash
= inet_frag_hashfn(f
, fq
);
233 spin_lock(&hb
->chain_lock
);
234 hlist_del(&fq
->list
);
235 spin_unlock(&hb
->chain_lock
);
237 read_unlock(&f
->lock
);
240 void inet_frag_kill(struct inet_frag_queue
*fq
, struct inet_frags
*f
)
242 if (del_timer(&fq
->timer
))
243 atomic_dec(&fq
->refcnt
);
245 if (!(fq
->last_in
& INET_FRAG_COMPLETE
)) {
247 atomic_dec(&fq
->refcnt
);
248 fq
->last_in
|= INET_FRAG_COMPLETE
;
251 EXPORT_SYMBOL(inet_frag_kill
);
253 static inline void frag_kfree_skb(struct netns_frags
*nf
, struct inet_frags
*f
,
261 void inet_frag_destroy(struct inet_frag_queue
*q
, struct inet_frags
*f
)
264 struct netns_frags
*nf
;
265 unsigned int sum
, sum_truesize
= 0;
267 WARN_ON(!(q
->last_in
& INET_FRAG_COMPLETE
));
268 WARN_ON(del_timer(&q
->timer
) != 0);
270 /* Release all fragment data. */
274 struct sk_buff
*xp
= fp
->next
;
276 sum_truesize
+= fp
->truesize
;
277 frag_kfree_skb(nf
, f
, fp
);
280 sum
= sum_truesize
+ f
->qsize
;
281 sub_frag_mem_limit(q
, sum
);
287 EXPORT_SYMBOL(inet_frag_destroy
);
289 static struct inet_frag_queue
*inet_frag_intern(struct netns_frags
*nf
,
290 struct inet_frag_queue
*qp_in
, struct inet_frags
*f
,
293 struct inet_frag_bucket
*hb
;
294 struct inet_frag_queue
*qp
;
297 read_lock(&f
->lock
); /* Protects against hash rebuild */
299 * While we stayed w/o the lock other CPU could update
300 * the rnd seed, so we need to re-calculate the hash
301 * chain. Fortunatelly the qp_in can be used to get one.
303 hash
= inet_frag_hashfn(f
, qp_in
);
305 spin_lock(&hb
->chain_lock
);
308 /* With SMP race we have to recheck hash table, because
309 * such entry could be created on other cpu, while we
310 * released the hash bucket lock.
312 hlist_for_each_entry(qp
, &hb
->chain
, list
) {
313 if (qp
->net
== nf
&& f
->match(qp
, arg
)) {
314 atomic_inc(&qp
->refcnt
);
315 spin_unlock(&hb
->chain_lock
);
316 read_unlock(&f
->lock
);
317 qp_in
->last_in
|= INET_FRAG_COMPLETE
;
318 inet_frag_put(qp_in
, f
);
324 if (!mod_timer(&qp
->timer
, jiffies
+ nf
->timeout
))
325 atomic_inc(&qp
->refcnt
);
327 atomic_inc(&qp
->refcnt
);
328 hlist_add_head(&qp
->list
, &hb
->chain
);
330 spin_unlock(&hb
->chain_lock
);
331 read_unlock(&f
->lock
);
336 static struct inet_frag_queue
*inet_frag_alloc(struct netns_frags
*nf
,
337 struct inet_frags
*f
, void *arg
)
339 struct inet_frag_queue
*q
;
341 if (frag_mem_limit(nf
) > nf
->high_thresh
) {
342 inet_frag_schedule_worker(f
);
346 q
= kzalloc(f
->qsize
, GFP_ATOMIC
);
351 f
->constructor(q
, arg
);
352 add_frag_mem_limit(q
, f
->qsize
);
354 setup_timer(&q
->timer
, f
->frag_expire
, (unsigned long)q
);
355 spin_lock_init(&q
->lock
);
356 atomic_set(&q
->refcnt
, 1);
361 static struct inet_frag_queue
*inet_frag_create(struct netns_frags
*nf
,
362 struct inet_frags
*f
, void *arg
)
364 struct inet_frag_queue
*q
;
366 q
= inet_frag_alloc(nf
, f
, arg
);
370 return inet_frag_intern(nf
, q
, f
, arg
);
373 struct inet_frag_queue
*inet_frag_find(struct netns_frags
*nf
,
374 struct inet_frags
*f
, void *key
, unsigned int hash
)
377 struct inet_frag_bucket
*hb
;
378 struct inet_frag_queue
*q
;
381 if (frag_mem_limit(nf
) > nf
->low_thresh
)
382 inet_frag_schedule_worker(f
);
384 hash
&= (INETFRAGS_HASHSZ
- 1);
387 spin_lock(&hb
->chain_lock
);
388 hlist_for_each_entry(q
, &hb
->chain
, list
) {
389 if (q
->net
== nf
&& f
->match(q
, key
)) {
390 atomic_inc(&q
->refcnt
);
391 spin_unlock(&hb
->chain_lock
);
392 read_unlock(&f
->lock
);
397 spin_unlock(&hb
->chain_lock
);
398 read_unlock(&f
->lock
);
400 if (depth
<= INETFRAGS_MAXDEPTH
)
401 return inet_frag_create(nf
, f
, key
);
403 return ERR_PTR(-ENOBUFS
);
405 EXPORT_SYMBOL(inet_frag_find
);
407 void inet_frag_maybe_warn_overflow(struct inet_frag_queue
*q
,
410 static const char msg
[] = "inet_frag_find: Fragment hash bucket"
411 " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH
)
412 ". Dropping fragment.\n";
414 if (PTR_ERR(q
) == -ENOBUFS
)
415 LIMIT_NETDEBUG(KERN_WARNING
"%s%s", prefix
, msg
);
417 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow
);