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 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
90 list_del(&fq->lru_list);
92 write_unlock(&f->lock);
95 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
97 if (del_timer(&fq->timer))
98 atomic_dec(&fq->refcnt);
100 if (!(fq->last_in & COMPLETE)) {
102 atomic_dec(&fq->refcnt);
103 fq->last_in |= COMPLETE;
107 EXPORT_SYMBOL(inet_frag_kill);
109 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
110 struct sk_buff *skb, int *work)
113 *work -= skb->truesize;
115 atomic_sub(skb->truesize, &nf->mem);
121 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
125 struct netns_frags *nf;
127 BUG_TRAP(q->last_in & COMPLETE);
128 BUG_TRAP(del_timer(&q->timer) == 0);
130 /* Release all fragment data. */
134 struct sk_buff *xp = fp->next;
136 frag_kfree_skb(nf, f, fp, work);
142 atomic_sub(f->qsize, &nf->mem);
149 EXPORT_SYMBOL(inet_frag_destroy);
151 int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f)
153 struct inet_frag_queue *q;
154 int work, evicted = 0;
156 work = atomic_read(&nf->mem) - nf->low_thresh;
159 if (list_empty(&nf->lru_list)) {
160 read_unlock(&f->lock);
164 q = list_first_entry(&nf->lru_list,
165 struct inet_frag_queue, lru_list);
166 atomic_inc(&q->refcnt);
167 read_unlock(&f->lock);
170 if (!(q->last_in & COMPLETE))
171 inet_frag_kill(q, f);
172 spin_unlock(&q->lock);
174 if (atomic_dec_and_test(&q->refcnt))
175 inet_frag_destroy(q, f, &work);
181 EXPORT_SYMBOL(inet_frag_evictor);
183 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
184 struct inet_frag_queue *qp_in, struct inet_frags *f,
185 unsigned int hash, void *arg)
187 struct inet_frag_queue *qp;
189 struct hlist_node *n;
192 write_lock(&f->lock);
194 /* With SMP race we have to recheck hash table, because
195 * such entry could be created on other cpu, while we
196 * promoted read lock to write lock.
198 hlist_for_each_entry(qp, n, &f->hash[hash], list) {
199 if (qp->net == nf && f->match(qp, arg)) {
200 atomic_inc(&qp->refcnt);
201 write_unlock(&f->lock);
202 qp_in->last_in |= COMPLETE;
203 inet_frag_put(qp_in, f);
209 if (!mod_timer(&qp->timer, jiffies + nf->timeout))
210 atomic_inc(&qp->refcnt);
212 atomic_inc(&qp->refcnt);
213 hlist_add_head(&qp->list, &f->hash[hash]);
214 list_add_tail(&qp->lru_list, &nf->lru_list);
216 write_unlock(&f->lock);
220 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
221 struct inet_frags *f, void *arg)
223 struct inet_frag_queue *q;
225 q = kzalloc(f->qsize, GFP_ATOMIC);
229 f->constructor(q, arg);
230 atomic_add(f->qsize, &nf->mem);
231 setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
232 spin_lock_init(&q->lock);
233 atomic_set(&q->refcnt, 1);
239 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
240 struct inet_frags *f, void *arg, unsigned int hash)
242 struct inet_frag_queue *q;
244 q = inet_frag_alloc(nf, f, arg);
248 return inet_frag_intern(nf, q, f, hash, arg);
251 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
252 struct inet_frags *f, void *key, unsigned int hash)
254 struct inet_frag_queue *q;
255 struct hlist_node *n;
258 hlist_for_each_entry(q, n, &f->hash[hash], list) {
259 if (q->net == nf && f->match(q, key)) {
260 atomic_inc(&q->refcnt);
261 read_unlock(&f->lock);
265 read_unlock(&f->lock);
267 return inet_frag_create(nf, f, key, hash);
269 EXPORT_SYMBOL(inet_frag_find);