1/*
  2 * inet fragments management
  3 *
  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.
  8 *
  9 * 		Authors:	Pavel Emelyanov <xemul@openvz.org>
 10 *				Started as consolidation of ipv4/ip_fragment.c,
 11 *				ipv6/reassembly. and ipv6 nf conntrack reassembly
 12 */
 13
 14#include <linux/list.h>
 15#include <linux/spinlock.h>
 16#include <linux/module.h>
 17#include <linux/timer.h>
 18#include <linux/mm.h>
 19#include <linux/random.h>
 20#include <linux/skbuff.h>
 21#include <linux/rtnetlink.h>
 22#include <linux/slab.h>
 23
 24#include <net/sock.h>
 25#include <net/inet_frag.h>
 26#include <net/inet_ecn.h>
 27
 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
 31 */
 32const 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,
 37
 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,
 46};
 47EXPORT_SYMBOL(ip_frag_ecn_table);
 48
 49int inet_frags_init(struct inet_frags *f)
 50{
 51	f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0,
 52					    NULL);
 53	if (!f->frags_cachep)
 54		return -ENOMEM;
 55
 56	return 0;
 57}
 58EXPORT_SYMBOL(inet_frags_init);
 59
 60void inet_frags_fini(struct inet_frags *f)
 61{
 62	/* We must wait that all inet_frag_destroy_rcu() have completed. */
 63	rcu_barrier();
 64
 65	kmem_cache_destroy(f->frags_cachep);
 66	f->frags_cachep = NULL;
 67}
 68EXPORT_SYMBOL(inet_frags_fini);
 69
 70static void inet_frags_free_cb(void *ptr, void *arg)
 71{
 72	struct inet_frag_queue *fq = ptr;
 73
 74	/* If we can not cancel the timer, it means this frag_queue
 75	 * is already disappearing, we have nothing to do.
 76	 * Otherwise, we own a refcount until the end of this function.
 77	 */
 78	if (!del_timer(&fq->timer))
 79		return;
 80
 81	spin_lock_bh(&fq->lock);
 82	if (!(fq->flags & INET_FRAG_COMPLETE)) {
 83		fq->flags |= INET_FRAG_COMPLETE;
 84		refcount_dec(&fq->refcnt);
 85	}
 86	spin_unlock_bh(&fq->lock);
 87
 88	inet_frag_put(fq);
 
 
 
 
 89}
 
 90
 91void inet_frags_exit_net(struct netns_frags *nf)
 92{
 93	nf->low_thresh = 0; /* prevent creation of new frags */
 
 
 
 
 
 
 94
 95	rhashtable_free_and_destroy(&nf->rhashtable, inet_frags_free_cb, NULL);
 
 
 96}
 97EXPORT_SYMBOL(inet_frags_exit_net);
 98
 99void inet_frag_kill(struct inet_frag_queue *fq)
100{
101	if (del_timer(&fq->timer))
102		refcount_dec(&fq->refcnt);
 
 
 
 
103
104	if (!(fq->flags & INET_FRAG_COMPLETE)) {
105		struct netns_frags *nf = fq->net;
 
 
106
107		fq->flags |= INET_FRAG_COMPLETE;
108		rhashtable_remove_fast(&nf->rhashtable, &fq->node, nf->f->rhash_params);
109		refcount_dec(&fq->refcnt);
 
110	}
111}
112EXPORT_SYMBOL(inet_frag_kill);
113
114static void inet_frag_destroy_rcu(struct rcu_head *head)
 
115{
116	struct inet_frag_queue *q = container_of(head, struct inet_frag_queue,
117						 rcu);
118	struct inet_frags *f = q->net->f;
119
120	if (f->destructor)
121		f->destructor(q);
122	kmem_cache_free(f->frags_cachep, q);
 
123}
124
125void inet_frag_destroy(struct inet_frag_queue *q)
 
126{
127	struct sk_buff *fp;
128	struct netns_frags *nf;
129	unsigned int sum, sum_truesize = 0;
130	struct inet_frags *f;
131
132	WARN_ON(!(q->flags & INET_FRAG_COMPLETE));
133	WARN_ON(del_timer(&q->timer) != 0);
134
135	/* Release all fragment data. */
136	fp = q->fragments;
137	nf = q->net;
138	f = nf->f;
139	while (fp) {
140		struct sk_buff *xp = fp->next;
141
142		sum_truesize += fp->truesize;
143		kfree_skb(fp);
144		fp = xp;
145	}
146	sum = sum_truesize + f->qsize;
147
148	call_rcu(&q->rcu, inet_frag_destroy_rcu);
 
 
 
 
 
 
149
150	sub_frag_mem_limit(nf, sum);
151}
152EXPORT_SYMBOL(inet_frag_destroy);
153
154static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
155					       struct inet_frags *f,
156					       void *arg)
157{
158	struct inet_frag_queue *q;
 
159
160	if (!nf->high_thresh || frag_mem_limit(nf) > nf->high_thresh)
161		return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
162
163	q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
164	if (!q)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
165		return NULL;
166
167	q->net = nf;
168	f->constructor(q, arg);
169	add_frag_mem_limit(nf, f->qsize);
170
171	timer_setup(&q->timer, f->frag_expire, 0);
172	spin_lock_init(&q->lock);
173	refcount_set(&q->refcnt, 3);
 
174
175	return q;
176}
177
178static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
179						void *arg)
180{
181	struct inet_frags *f = nf->f;
182	struct inet_frag_queue *q;
183	int err;
184
185	q = inet_frag_alloc(nf, f, arg);
186	if (!q)
187		return NULL;
188
189	mod_timer(&q->timer, jiffies + nf->timeout);
190
191	err = rhashtable_insert_fast(&nf->rhashtable, &q->node,
192				     f->rhash_params);
193	if (err < 0) {
194		q->flags |= INET_FRAG_COMPLETE;
195		inet_frag_kill(q);
196		inet_frag_destroy(q);
197		return NULL;
198	}
199	return q;
200}
201
202/* TODO : call from rcu_read_lock() and no longer use refcount_inc_not_zero() */
203struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, void *key)
 
204{
205	struct inet_frag_queue *fq;
206
207	rcu_read_lock();
208
209	fq = rhashtable_lookup(&nf->rhashtable, key, nf->f->rhash_params);
210	if (fq) {
211		if (!refcount_inc_not_zero(&fq->refcnt))
212			fq = NULL;
213		rcu_read_unlock();
214		return fq;
215	}
216	rcu_read_unlock();
217
218	return inet_frag_create(nf, key);
219}
220EXPORT_SYMBOL(inet_frag_find);

  1/*
  2 * inet fragments management
  3 *
  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.
  8 *
  9 * 		Authors:	Pavel Emelyanov <xemul@openvz.org>
 10 *				Started as consolidation of ipv4/ip_fragment.c,
 11 *				ipv6/reassembly. and ipv6 nf conntrack reassembly
 12 */
 13
 14#include <linux/list.h>
 15#include <linux/spinlock.h>
 16#include <linux/module.h>
 17#include <linux/timer.h>
 18#include <linux/mm.h>
 19#include <linux/random.h>
 20#include <linux/skbuff.h>
 21#include <linux/rtnetlink.h>
 22#include <linux/slab.h>
 23
 
 24#include <net/inet_frag.h>
 
 25
 26static void inet_frag_secret_rebuild(unsigned long dummy)
 27{
 28	struct inet_frags *f = (struct inet_frags *)dummy;
 29	unsigned long now = jiffies;
 30	int i;
 31
 32	write_lock(&f->lock);
 33	get_random_bytes(&f->rnd, sizeof(u32));
 34	for (i = 0; i < INETFRAGS_HASHSZ; i++) {
 35		struct inet_frag_queue *q;
 36		struct hlist_node *p, *n;
 37
 38		hlist_for_each_entry_safe(q, p, n, &f->hash[i], list) {
 39			unsigned int hval = f->hashfn(q);
 40
 41			if (hval != i) {
 42				hlist_del(&q->list);
 43
 44				/* Relink to new hash chain. */
 45				hlist_add_head(&q->list, &f->hash[hval]);
 46			}
 47		}
 48	}
 49	write_unlock(&f->lock);
 
 
 
 50
 51	mod_timer(&f->secret_timer, now + f->secret_interval);
 52}
 
 53
 54void inet_frags_init(struct inet_frags *f)
 55{
 56	int i;
 
 57
 58	for (i = 0; i < INETFRAGS_HASHSZ; i++)
 59		INIT_HLIST_HEAD(&f->hash[i]);
 
 
 60
 61	rwlock_init(&f->lock);
 
 
 62
 63	f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
 64				   (jiffies ^ (jiffies >> 6)));
 
 
 
 
 65
 66	setup_timer(&f->secret_timer, inet_frag_secret_rebuild,
 67			(unsigned long)f);
 68	f->secret_timer.expires = jiffies + f->secret_interval;
 69	add_timer(&f->secret_timer);
 70}
 71EXPORT_SYMBOL(inet_frags_init);
 72
 73void inet_frags_init_net(struct netns_frags *nf)
 74{
 75	nf->nqueues = 0;
 76	atomic_set(&nf->mem, 0);
 77	INIT_LIST_HEAD(&nf->lru_list);
 78}
 79EXPORT_SYMBOL(inet_frags_init_net);
 80
 81void inet_frags_fini(struct inet_frags *f)
 82{
 83	del_timer(&f->secret_timer);
 84}
 85EXPORT_SYMBOL(inet_frags_fini);
 86
 87void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
 88{
 89	nf->low_thresh = 0;
 90
 91	local_bh_disable();
 92	inet_frag_evictor(nf, f);
 93	local_bh_enable();
 94}
 95EXPORT_SYMBOL(inet_frags_exit_net);
 96
 97static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
 98{
 99	write_lock(&f->lock);
100	hlist_del(&fq->list);
101	list_del(&fq->lru_list);
102	fq->net->nqueues--;
103	write_unlock(&f->lock);
104}
105
106void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
107{
108	if (del_timer(&fq->timer))
109		atomic_dec(&fq->refcnt);
110
111	if (!(fq->last_in & INET_FRAG_COMPLETE)) {
112		fq_unlink(fq, f);
113		atomic_dec(&fq->refcnt);
114		fq->last_in |= INET_FRAG_COMPLETE;
115	}
116}
117EXPORT_SYMBOL(inet_frag_kill);
118
119static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
120		struct sk_buff *skb, int *work)
121{
122	if (work)
123		*work -= skb->truesize;
 
124
125	atomic_sub(skb->truesize, &nf->mem);
126	if (f->skb_free)
127		f->skb_free(skb);
128	kfree_skb(skb);
129}
130
131void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
132					int *work)
133{
134	struct sk_buff *fp;
135	struct netns_frags *nf;
 
 
136
137	WARN_ON(!(q->last_in & INET_FRAG_COMPLETE));
138	WARN_ON(del_timer(&q->timer) != 0);
139
140	/* Release all fragment data. */
141	fp = q->fragments;
142	nf = q->net;
 
143	while (fp) {
144		struct sk_buff *xp = fp->next;
145
146		frag_kfree_skb(nf, f, fp, work);
 
147		fp = xp;
148	}
 
149
150	if (work)
151		*work -= f->qsize;
152	atomic_sub(f->qsize, &nf->mem);
153
154	if (f->destructor)
155		f->destructor(q);
156	kfree(q);
157
 
158}
159EXPORT_SYMBOL(inet_frag_destroy);
160
161int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f)
 
 
162{
163	struct inet_frag_queue *q;
164	int work, evicted = 0;
165
166	work = atomic_read(&nf->mem) - nf->low_thresh;
167	while (work > 0) {
168		read_lock(&f->lock);
169		if (list_empty(&nf->lru_list)) {
170			read_unlock(&f->lock);
171			break;
172		}
173
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);
178
179		spin_lock(&q->lock);
180		if (!(q->last_in & INET_FRAG_COMPLETE))
181			inet_frag_kill(q, f);
182		spin_unlock(&q->lock);
183
184		if (atomic_dec_and_test(&q->refcnt))
185			inet_frag_destroy(q, f, &work);
186		evicted++;
187	}
188
189	return evicted;
190}
191EXPORT_SYMBOL(inet_frag_evictor);
192
193static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
194		struct inet_frag_queue *qp_in, struct inet_frags *f,
195		void *arg)
196{
197	struct inet_frag_queue *qp;
198#ifdef CONFIG_SMP
199	struct hlist_node *n;
200#endif
201	unsigned int hash;
202
203	write_lock(&f->lock);
204	/*
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.
208	 */
209	hash = f->hashfn(qp_in);
210#ifdef CONFIG_SMP
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.
214	 */
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);
221			return qp;
222		}
223	}
224#endif
225	qp = qp_in;
226	if (!mod_timer(&qp->timer, jiffies + nf->timeout))
227		atomic_inc(&qp->refcnt);
228
229	atomic_inc(&qp->refcnt);
230	hlist_add_head(&qp->list, &f->hash[hash]);
231	list_add_tail(&qp->lru_list, &nf->lru_list);
232	nf->nqueues++;
233	write_unlock(&f->lock);
234	return qp;
235}
236
237static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
238		struct inet_frags *f, void *arg)
239{
240	struct inet_frag_queue *q;
241
242	q = kzalloc(f->qsize, GFP_ATOMIC);
243	if (q == NULL)
244		return NULL;
245
 
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);
251	q->net = nf;
252
253	return q;
254}
255
256static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
257		struct inet_frags *f, void *arg)
258{
 
259	struct inet_frag_queue *q;
 
260
261	q = inet_frag_alloc(nf, f, arg);
262	if (q == NULL)
263		return NULL;
264
265	return inet_frag_intern(nf, q, f, arg);
 
 
 
 
 
 
 
 
 
 
266}
267
268struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
269		struct inet_frags *f, void *key, unsigned int hash)
270	__releases(&f->lock)
271{
272	struct inet_frag_queue *q;
273	struct hlist_node *n;
 
274
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);
279			return q;
280		}
281	}
282	read_unlock(&f->lock);
283
284	return inet_frag_create(nf, f, key);
285}
286EXPORT_SYMBOL(inet_frag_find);