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#define INETFRAGS_EVICT_BUCKETS   128
29#define INETFRAGS_EVICT_MAX	  512
30
31/* don't rebuild inetfrag table with new secret more often than this */
32#define INETFRAGS_MIN_REBUILD_INTERVAL (5 * HZ)
33
34/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
35 * Value : 0xff if frame should be dropped.
36 *         0 or INET_ECN_CE value, to be ORed in to final iph->tos field
37 */
38const u8 ip_frag_ecn_table[16] = {
39	/* at least one fragment had CE, and others ECT_0 or ECT_1 */
40	[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0]			= INET_ECN_CE,
41	[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1]			= INET_ECN_CE,
42	[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1]	= INET_ECN_CE,
43
44	/* invalid combinations : drop frame */
45	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
46	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
47	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
48	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
49	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
50	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
51	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
52};
53EXPORT_SYMBOL(ip_frag_ecn_table);
54
55static unsigned int
56inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q)
57{
58	return f->hashfn(q) & (INETFRAGS_HASHSZ - 1);
59}
60
61static bool inet_frag_may_rebuild(struct inet_frags *f)
62{
63	return time_after(jiffies,
64	       f->last_rebuild_jiffies + INETFRAGS_MIN_REBUILD_INTERVAL);
65}
66
67static void inet_frag_secret_rebuild(struct inet_frags *f)
68{
69	int i;
70
71	write_seqlock_bh(&f->rnd_seqlock);
72
73	if (!inet_frag_may_rebuild(f))
74		goto out;
75
76	get_random_bytes(&f->rnd, sizeof(u32));
77
78	for (i = 0; i < INETFRAGS_HASHSZ; i++) {
79		struct inet_frag_bucket *hb;
80		struct inet_frag_queue *q;
81		struct hlist_node *n;
82
83		hb = &f->hash[i];
84		spin_lock(&hb->chain_lock);
85
86		hlist_for_each_entry_safe(q, n, &hb->chain, list) {
87			unsigned int hval = inet_frag_hashfn(f, q);
88
89			if (hval != i) {
90				struct inet_frag_bucket *hb_dest;
91
92				hlist_del(&q->list);
93
94				/* Relink to new hash chain. */
95				hb_dest = &f->hash[hval];
96
97				/* This is the only place where we take
98				 * another chain_lock while already holding
99				 * one.  As this will not run concurrently,
100				 * we cannot deadlock on hb_dest lock below, if its
101				 * already locked it will be released soon since
102				 * other caller cannot be waiting for hb lock
103				 * that we've taken above.
104				 */
105				spin_lock_nested(&hb_dest->chain_lock,
106						 SINGLE_DEPTH_NESTING);
107				hlist_add_head(&q->list, &hb_dest->chain);
108				spin_unlock(&hb_dest->chain_lock);
109			}
110		}
111		spin_unlock(&hb->chain_lock);
112	}
113
114	f->rebuild = false;
115	f->last_rebuild_jiffies = jiffies;
116out:
117	write_sequnlock_bh(&f->rnd_seqlock);
118}
119
120static bool inet_fragq_should_evict(const struct inet_frag_queue *q)
121{
122	return q->net->low_thresh == 0 ||
123	       frag_mem_limit(q->net) >= q->net->low_thresh;
124}
125
126static unsigned int
127inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb)
128{
129	struct inet_frag_queue *fq;
130	struct hlist_node *n;
131	unsigned int evicted = 0;
132	HLIST_HEAD(expired);
133
134evict_again:
135	spin_lock(&hb->chain_lock);
136
137	hlist_for_each_entry_safe(fq, n, &hb->chain, list) {
138		if (!inet_fragq_should_evict(fq))
139			continue;
140
141		if (!del_timer(&fq->timer)) {
142			/* q expiring right now thus increment its refcount so
143			 * it won't be freed under us and wait until the timer
144			 * has finished executing then destroy it
145			 */
146			atomic_inc(&fq->refcnt);
147			spin_unlock(&hb->chain_lock);
148			del_timer_sync(&fq->timer);
149			inet_frag_put(fq, f);
150			goto evict_again;
151		}
152
153		fq->flags |= INET_FRAG_EVICTED;
154		hlist_del(&fq->list);
155		hlist_add_head(&fq->list, &expired);
156		++evicted;
157	}
158
159	spin_unlock(&hb->chain_lock);
160
161	hlist_for_each_entry_safe(fq, n, &expired, list)
162		f->frag_expire((unsigned long) fq);
163
164	return evicted;
165}
166
167static void inet_frag_worker(struct work_struct *work)
168{
169	unsigned int budget = INETFRAGS_EVICT_BUCKETS;
170	unsigned int i, evicted = 0;
171	struct inet_frags *f;
172
173	f = container_of(work, struct inet_frags, frags_work);
174
175	BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ);
176
177	local_bh_disable();
178
179	for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) {
180		evicted += inet_evict_bucket(f, &f->hash[i]);
181		i = (i + 1) & (INETFRAGS_HASHSZ - 1);
182		if (evicted > INETFRAGS_EVICT_MAX)
183			break;
184	}
185
186	f->next_bucket = i;
187
188	local_bh_enable();
189
190	if (f->rebuild && inet_frag_may_rebuild(f))
191		inet_frag_secret_rebuild(f);
192}
193
194static void inet_frag_schedule_worker(struct inet_frags *f)
195{
196	if (unlikely(!work_pending(&f->frags_work)))
197		schedule_work(&f->frags_work);
198}
199
200int inet_frags_init(struct inet_frags *f)
201{
202	int i;
203
204	INIT_WORK(&f->frags_work, inet_frag_worker);
205
206	for (i = 0; i < INETFRAGS_HASHSZ; i++) {
207		struct inet_frag_bucket *hb = &f->hash[i];
208
209		spin_lock_init(&hb->chain_lock);
210		INIT_HLIST_HEAD(&hb->chain);
211	}
212
213	seqlock_init(&f->rnd_seqlock);
214	f->last_rebuild_jiffies = 0;
215	f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0,
216					    NULL);
217	if (!f->frags_cachep)
218		return -ENOMEM;
219
220	return 0;
221}
222EXPORT_SYMBOL(inet_frags_init);
223
224void inet_frags_init_net(struct netns_frags *nf)
225{
226	init_frag_mem_limit(nf);
227}
228EXPORT_SYMBOL(inet_frags_init_net);
229
230void inet_frags_fini(struct inet_frags *f)
231{
232	cancel_work_sync(&f->frags_work);
233	kmem_cache_destroy(f->frags_cachep);
234}
235EXPORT_SYMBOL(inet_frags_fini);
236
237void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
238{
239	unsigned int seq;
240	int i;
241
242	nf->low_thresh = 0;
243	local_bh_disable();
244
245evict_again:
246	seq = read_seqbegin(&f->rnd_seqlock);
247
248	for (i = 0; i < INETFRAGS_HASHSZ ; i++)
249		inet_evict_bucket(f, &f->hash[i]);
250
251	if (read_seqretry(&f->rnd_seqlock, seq))
252		goto evict_again;
253
254	local_bh_enable();
255
256	percpu_counter_destroy(&nf->mem);
257}
258EXPORT_SYMBOL(inet_frags_exit_net);
259
260static struct inet_frag_bucket *
261get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f)
262__acquires(hb->chain_lock)
263{
264	struct inet_frag_bucket *hb;
265	unsigned int seq, hash;
266
267 restart:
268	seq = read_seqbegin(&f->rnd_seqlock);
269
270	hash = inet_frag_hashfn(f, fq);
271	hb = &f->hash[hash];
272
273	spin_lock(&hb->chain_lock);
274	if (read_seqretry(&f->rnd_seqlock, seq)) {
275		spin_unlock(&hb->chain_lock);
276		goto restart;
277	}
278
279	return hb;
280}
281
282static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
283{
284	struct inet_frag_bucket *hb;
285
286	hb = get_frag_bucket_locked(fq, f);
287	if (!(fq->flags & INET_FRAG_EVICTED))
288		hlist_del(&fq->list);
289	spin_unlock(&hb->chain_lock);
290}
291
292void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
293{
294	if (del_timer(&fq->timer))
295		atomic_dec(&fq->refcnt);
296
297	if (!(fq->flags & INET_FRAG_COMPLETE)) {
298		fq_unlink(fq, f);
299		atomic_dec(&fq->refcnt);
300		fq->flags |= INET_FRAG_COMPLETE;
301	}
302}
303EXPORT_SYMBOL(inet_frag_kill);
304
305static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
306				  struct sk_buff *skb)
307{
308	if (f->skb_free)
309		f->skb_free(skb);
310	kfree_skb(skb);
311}
312
313void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f)
314{
315	struct sk_buff *fp;
316	struct netns_frags *nf;
317	unsigned int sum, sum_truesize = 0;
318
319	WARN_ON(!(q->flags & INET_FRAG_COMPLETE));
320	WARN_ON(del_timer(&q->timer) != 0);
321
322	/* Release all fragment data. */
323	fp = q->fragments;
324	nf = q->net;
325	while (fp) {
326		struct sk_buff *xp = fp->next;
327
328		sum_truesize += fp->truesize;
329		frag_kfree_skb(nf, f, fp);
330		fp = xp;
331	}
332	sum = sum_truesize + f->qsize;
333	sub_frag_mem_limit(q, sum);
334
335	if (f->destructor)
336		f->destructor(q);
337	kmem_cache_free(f->frags_cachep, q);
338}
339EXPORT_SYMBOL(inet_frag_destroy);
340
341static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
342						struct inet_frag_queue *qp_in,
343						struct inet_frags *f,
344						void *arg)
345{
346	struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f);
347	struct inet_frag_queue *qp;
348
349#ifdef CONFIG_SMP
350	/* With SMP race we have to recheck hash table, because
351	 * such entry could have been created on other cpu before
352	 * we acquired hash bucket lock.
353	 */
354	hlist_for_each_entry(qp, &hb->chain, list) {
355		if (qp->net == nf && f->match(qp, arg)) {
356			atomic_inc(&qp->refcnt);
357			spin_unlock(&hb->chain_lock);
358			qp_in->flags |= INET_FRAG_COMPLETE;
359			inet_frag_put(qp_in, f);
360			return qp;
361		}
362	}
363#endif
364	qp = qp_in;
365	if (!mod_timer(&qp->timer, jiffies + nf->timeout))
366		atomic_inc(&qp->refcnt);
367
368	atomic_inc(&qp->refcnt);
369	hlist_add_head(&qp->list, &hb->chain);
370
371	spin_unlock(&hb->chain_lock);
372
373	return qp;
374}
375
376static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
377					       struct inet_frags *f,
378					       void *arg)
379{
380	struct inet_frag_queue *q;
381
382	if (frag_mem_limit(nf) > nf->high_thresh) {
383		inet_frag_schedule_worker(f);
384		return NULL;
385	}
386
387	q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
388	if (!q)
389		return NULL;
390
391	q->net = nf;
392	f->constructor(q, arg);
393	add_frag_mem_limit(q, f->qsize);
394
395	setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
396	spin_lock_init(&q->lock);
397	atomic_set(&q->refcnt, 1);
398
399	return q;
400}
401
402static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
403						struct inet_frags *f,
404						void *arg)
405{
406	struct inet_frag_queue *q;
407
408	q = inet_frag_alloc(nf, f, arg);
409	if (!q)
410		return NULL;
411
412	return inet_frag_intern(nf, q, f, arg);
413}
414
415struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
416				       struct inet_frags *f, void *key,
417				       unsigned int hash)
418{
419	struct inet_frag_bucket *hb;
420	struct inet_frag_queue *q;
421	int depth = 0;
422
423	if (frag_mem_limit(nf) > nf->low_thresh)
424		inet_frag_schedule_worker(f);
425
426	hash &= (INETFRAGS_HASHSZ - 1);
427	hb = &f->hash[hash];
428
429	spin_lock(&hb->chain_lock);
430	hlist_for_each_entry(q, &hb->chain, list) {
431		if (q->net == nf && f->match(q, key)) {
432			atomic_inc(&q->refcnt);
433			spin_unlock(&hb->chain_lock);
434			return q;
435		}
436		depth++;
437	}
438	spin_unlock(&hb->chain_lock);
439
440	if (depth <= INETFRAGS_MAXDEPTH)
441		return inet_frag_create(nf, f, key);
442
443	if (inet_frag_may_rebuild(f)) {
444		if (!f->rebuild)
445			f->rebuild = true;
446		inet_frag_schedule_worker(f);
447	}
448
449	return ERR_PTR(-ENOBUFS);
450}
451EXPORT_SYMBOL(inet_frag_find);
452
453void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
454				   const char *prefix)
455{
456	static const char msg[] = "inet_frag_find: Fragment hash bucket"
457		" list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
458		". Dropping fragment.\n";
459
460	if (PTR_ERR(q) == -ENOBUFS)
461		net_dbg_ratelimited("%s%s", prefix, msg);
462}
463EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);
464