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>
24 #include <net/inet_frag.h>
26 static void inet_frag_secret_rebuild(unsigned long dummy)
28 struct inet_frags *f = (struct inet_frags *)dummy;
29 unsigned long now = jiffies;
33 get_random_bytes(&f->rnd, sizeof(u32));
34 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
35 struct inet_frag_queue *q;
38 hlist_for_each_entry_safe(q, n, &f->hash[i], list) {
39 unsigned int hval = f->hashfn(q);
44 /* Relink to new hash chain. */
45 hlist_add_head(&q->list, &f->hash[hval]);
49 write_unlock(&f->lock);
51 mod_timer(&f->secret_timer, now + f->secret_interval);
54 void inet_frags_init(struct inet_frags *f)
58 for (i = 0; i < INETFRAGS_HASHSZ; i++)
59 INIT_HLIST_HEAD(&f->hash[i]);
61 rwlock_init(&f->lock);
63 f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
64 (jiffies ^ (jiffies >> 6)));
66 setup_timer(&f->secret_timer, inet_frag_secret_rebuild,
68 f->secret_timer.expires = jiffies + f->secret_interval;
69 add_timer(&f->secret_timer);
71 EXPORT_SYMBOL(inet_frags_init);
73 void inet_frags_init_net(struct netns_frags *nf)
76 init_frag_mem_limit(nf);
77 INIT_LIST_HEAD(&nf->lru_list);
78 spin_lock_init(&nf->lru_lock);
80 EXPORT_SYMBOL(inet_frags_init_net);
82 void inet_frags_fini(struct inet_frags *f)
84 del_timer(&f->secret_timer);
86 EXPORT_SYMBOL(inet_frags_fini);
88 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
93 inet_frag_evictor(nf, f, true);
96 percpu_counter_destroy(&nf->mem);
98 EXPORT_SYMBOL(inet_frags_exit_net);
100 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
102 write_lock(&f->lock);
103 hlist_del(&fq->list);
105 write_unlock(&f->lock);
106 inet_frag_lru_del(fq);
109 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
111 if (del_timer(&fq->timer))
112 atomic_dec(&fq->refcnt);
114 if (!(fq->last_in & INET_FRAG_COMPLETE)) {
116 atomic_dec(&fq->refcnt);
117 fq->last_in |= INET_FRAG_COMPLETE;
120 EXPORT_SYMBOL(inet_frag_kill);
122 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
130 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
134 struct netns_frags *nf;
135 unsigned int sum, sum_truesize = 0;
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 sum_truesize += fp->truesize;
147 frag_kfree_skb(nf, f, fp);
150 sum = sum_truesize + f->qsize;
153 sub_frag_mem_limit(q, sum);
160 EXPORT_SYMBOL(inet_frag_destroy);
162 int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force)
164 struct inet_frag_queue *q;
165 int work, evicted = 0;
168 if (frag_mem_limit(nf) <= nf->high_thresh)
172 work = frag_mem_limit(nf) - nf->low_thresh;
174 spin_lock(&nf->lru_lock);
176 if (list_empty(&nf->lru_list)) {
177 spin_unlock(&nf->lru_lock);
181 q = list_first_entry(&nf->lru_list,
182 struct inet_frag_queue, lru_list);
183 atomic_inc(&q->refcnt);
184 spin_unlock(&nf->lru_lock);
187 if (!(q->last_in & INET_FRAG_COMPLETE))
188 inet_frag_kill(q, f);
189 spin_unlock(&q->lock);
191 if (atomic_dec_and_test(&q->refcnt))
192 inet_frag_destroy(q, f, &work);
198 EXPORT_SYMBOL(inet_frag_evictor);
200 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
201 struct inet_frag_queue *qp_in, struct inet_frags *f,
204 struct inet_frag_queue *qp;
209 write_lock(&f->lock);
211 * While we stayed w/o the lock other CPU could update
212 * the rnd seed, so we need to re-calculate the hash
213 * chain. Fortunatelly the qp_in can be used to get one.
215 hash = f->hashfn(qp_in);
217 /* With SMP race we have to recheck hash table, because
218 * such entry could be created on other cpu, while we
219 * promoted read lock to write lock.
221 hlist_for_each_entry(qp, &f->hash[hash], list) {
222 if (qp->net == nf && f->match(qp, arg)) {
223 atomic_inc(&qp->refcnt);
224 write_unlock(&f->lock);
225 qp_in->last_in |= INET_FRAG_COMPLETE;
226 inet_frag_put(qp_in, f);
232 if (!mod_timer(&qp->timer, jiffies + nf->timeout))
233 atomic_inc(&qp->refcnt);
235 atomic_inc(&qp->refcnt);
236 hlist_add_head(&qp->list, &f->hash[hash]);
238 write_unlock(&f->lock);
239 inet_frag_lru_add(nf, qp);
243 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
244 struct inet_frags *f, void *arg)
246 struct inet_frag_queue *q;
248 q = kzalloc(f->qsize, GFP_ATOMIC);
253 f->constructor(q, arg);
254 add_frag_mem_limit(q, f->qsize);
256 setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
257 spin_lock_init(&q->lock);
258 atomic_set(&q->refcnt, 1);
263 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
264 struct inet_frags *f, void *arg)
266 struct inet_frag_queue *q;
268 q = inet_frag_alloc(nf, f, arg);
272 return inet_frag_intern(nf, q, f, arg);
275 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
276 struct inet_frags *f, void *key, unsigned int hash)
279 struct inet_frag_queue *q;
281 hlist_for_each_entry(q, &f->hash[hash], list) {
282 if (q->net == nf && f->match(q, key)) {
283 atomic_inc(&q->refcnt);
284 read_unlock(&f->lock);
288 read_unlock(&f->lock);
290 return inet_frag_create(nf, f, key);
292 EXPORT_SYMBOL(inet_frag_find);