xref: /linux/net/ipv4/inet_fragment.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
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  */
38 const 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 };
53 EXPORT_SYMBOL(ip_frag_ecn_table);
54 
55 static unsigned int
56 inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q)
57 {
58 	return f->hashfn(q) & (INETFRAGS_HASHSZ - 1);
59 }
60 
61 static 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 
67 static 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;
116 out:
117 	write_sequnlock_bh(&f->rnd_seqlock);
118 }
119 
120 static 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 
126 static unsigned int
127 inet_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 
134 	spin_lock(&hb->chain_lock);
135 
136 	hlist_for_each_entry_safe(fq, n, &hb->chain, list) {
137 		if (!inet_fragq_should_evict(fq))
138 			continue;
139 
140 		if (!del_timer(&fq->timer))
141 			continue;
142 
143 		hlist_add_head(&fq->list_evictor, &expired);
144 		++evicted;
145 	}
146 
147 	spin_unlock(&hb->chain_lock);
148 
149 	hlist_for_each_entry_safe(fq, n, &expired, list_evictor)
150 		f->frag_expire((unsigned long) fq);
151 
152 	return evicted;
153 }
154 
155 static void inet_frag_worker(struct work_struct *work)
156 {
157 	unsigned int budget = INETFRAGS_EVICT_BUCKETS;
158 	unsigned int i, evicted = 0;
159 	struct inet_frags *f;
160 
161 	f = container_of(work, struct inet_frags, frags_work);
162 
163 	BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ);
164 
165 	local_bh_disable();
166 
167 	for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) {
168 		evicted += inet_evict_bucket(f, &f->hash[i]);
169 		i = (i + 1) & (INETFRAGS_HASHSZ - 1);
170 		if (evicted > INETFRAGS_EVICT_MAX)
171 			break;
172 	}
173 
174 	f->next_bucket = i;
175 
176 	local_bh_enable();
177 
178 	if (f->rebuild && inet_frag_may_rebuild(f))
179 		inet_frag_secret_rebuild(f);
180 }
181 
182 static void inet_frag_schedule_worker(struct inet_frags *f)
183 {
184 	if (unlikely(!work_pending(&f->frags_work)))
185 		schedule_work(&f->frags_work);
186 }
187 
188 int inet_frags_init(struct inet_frags *f)
189 {
190 	int i;
191 
192 	INIT_WORK(&f->frags_work, inet_frag_worker);
193 
194 	for (i = 0; i < INETFRAGS_HASHSZ; i++) {
195 		struct inet_frag_bucket *hb = &f->hash[i];
196 
197 		spin_lock_init(&hb->chain_lock);
198 		INIT_HLIST_HEAD(&hb->chain);
199 	}
200 
201 	seqlock_init(&f->rnd_seqlock);
202 	f->last_rebuild_jiffies = 0;
203 	f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0,
204 					    NULL);
205 	if (!f->frags_cachep)
206 		return -ENOMEM;
207 
208 	return 0;
209 }
210 EXPORT_SYMBOL(inet_frags_init);
211 
212 void inet_frags_fini(struct inet_frags *f)
213 {
214 	cancel_work_sync(&f->frags_work);
215 	kmem_cache_destroy(f->frags_cachep);
216 }
217 EXPORT_SYMBOL(inet_frags_fini);
218 
219 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
220 {
221 	unsigned int seq;
222 	int i;
223 
224 	nf->low_thresh = 0;
225 
226 evict_again:
227 	local_bh_disable();
228 	seq = read_seqbegin(&f->rnd_seqlock);
229 
230 	for (i = 0; i < INETFRAGS_HASHSZ ; i++)
231 		inet_evict_bucket(f, &f->hash[i]);
232 
233 	local_bh_enable();
234 	cond_resched();
235 
236 	if (read_seqretry(&f->rnd_seqlock, seq) ||
237 	    percpu_counter_sum(&nf->mem))
238 		goto evict_again;
239 
240 	percpu_counter_destroy(&nf->mem);
241 }
242 EXPORT_SYMBOL(inet_frags_exit_net);
243 
244 static struct inet_frag_bucket *
245 get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f)
246 __acquires(hb->chain_lock)
247 {
248 	struct inet_frag_bucket *hb;
249 	unsigned int seq, hash;
250 
251  restart:
252 	seq = read_seqbegin(&f->rnd_seqlock);
253 
254 	hash = inet_frag_hashfn(f, fq);
255 	hb = &f->hash[hash];
256 
257 	spin_lock(&hb->chain_lock);
258 	if (read_seqretry(&f->rnd_seqlock, seq)) {
259 		spin_unlock(&hb->chain_lock);
260 		goto restart;
261 	}
262 
263 	return hb;
264 }
265 
266 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
267 {
268 	struct inet_frag_bucket *hb;
269 
270 	hb = get_frag_bucket_locked(fq, f);
271 	hlist_del(&fq->list);
272 	fq->flags |= INET_FRAG_COMPLETE;
273 	spin_unlock(&hb->chain_lock);
274 }
275 
276 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
277 {
278 	if (del_timer(&fq->timer))
279 		atomic_dec(&fq->refcnt);
280 
281 	if (!(fq->flags & INET_FRAG_COMPLETE)) {
282 		fq_unlink(fq, f);
283 		atomic_dec(&fq->refcnt);
284 	}
285 }
286 EXPORT_SYMBOL(inet_frag_kill);
287 
288 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f)
289 {
290 	struct sk_buff *fp;
291 	struct netns_frags *nf;
292 	unsigned int sum, sum_truesize = 0;
293 
294 	WARN_ON(!(q->flags & INET_FRAG_COMPLETE));
295 	WARN_ON(del_timer(&q->timer) != 0);
296 
297 	/* Release all fragment data. */
298 	fp = q->fragments;
299 	nf = q->net;
300 	while (fp) {
301 		struct sk_buff *xp = fp->next;
302 
303 		sum_truesize += fp->truesize;
304 		kfree_skb(fp);
305 		fp = xp;
306 	}
307 	sum = sum_truesize + f->qsize;
308 
309 	if (f->destructor)
310 		f->destructor(q);
311 	kmem_cache_free(f->frags_cachep, q);
312 
313 	sub_frag_mem_limit(nf, sum);
314 }
315 EXPORT_SYMBOL(inet_frag_destroy);
316 
317 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
318 						struct inet_frag_queue *qp_in,
319 						struct inet_frags *f,
320 						void *arg)
321 {
322 	struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f);
323 	struct inet_frag_queue *qp;
324 
325 #ifdef CONFIG_SMP
326 	/* With SMP race we have to recheck hash table, because
327 	 * such entry could have been created on other cpu before
328 	 * we acquired hash bucket lock.
329 	 */
330 	hlist_for_each_entry(qp, &hb->chain, list) {
331 		if (qp->net == nf && f->match(qp, arg)) {
332 			atomic_inc(&qp->refcnt);
333 			spin_unlock(&hb->chain_lock);
334 			qp_in->flags |= INET_FRAG_COMPLETE;
335 			inet_frag_put(qp_in, f);
336 			return qp;
337 		}
338 	}
339 #endif
340 	qp = qp_in;
341 	if (!mod_timer(&qp->timer, jiffies + nf->timeout))
342 		atomic_inc(&qp->refcnt);
343 
344 	atomic_inc(&qp->refcnt);
345 	hlist_add_head(&qp->list, &hb->chain);
346 
347 	spin_unlock(&hb->chain_lock);
348 
349 	return qp;
350 }
351 
352 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
353 					       struct inet_frags *f,
354 					       void *arg)
355 {
356 	struct inet_frag_queue *q;
357 
358 	if (!nf->high_thresh || frag_mem_limit(nf) > nf->high_thresh) {
359 		inet_frag_schedule_worker(f);
360 		return NULL;
361 	}
362 
363 	q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
364 	if (!q)
365 		return NULL;
366 
367 	q->net = nf;
368 	f->constructor(q, arg);
369 	add_frag_mem_limit(nf, f->qsize);
370 
371 	setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
372 	spin_lock_init(&q->lock);
373 	atomic_set(&q->refcnt, 1);
374 
375 	return q;
376 }
377 
378 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
379 						struct inet_frags *f,
380 						void *arg)
381 {
382 	struct inet_frag_queue *q;
383 
384 	q = inet_frag_alloc(nf, f, arg);
385 	if (!q)
386 		return NULL;
387 
388 	return inet_frag_intern(nf, q, f, arg);
389 }
390 
391 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
392 				       struct inet_frags *f, void *key,
393 				       unsigned int hash)
394 {
395 	struct inet_frag_bucket *hb;
396 	struct inet_frag_queue *q;
397 	int depth = 0;
398 
399 	if (frag_mem_limit(nf) > nf->low_thresh)
400 		inet_frag_schedule_worker(f);
401 
402 	hash &= (INETFRAGS_HASHSZ - 1);
403 	hb = &f->hash[hash];
404 
405 	spin_lock(&hb->chain_lock);
406 	hlist_for_each_entry(q, &hb->chain, list) {
407 		if (q->net == nf && f->match(q, key)) {
408 			atomic_inc(&q->refcnt);
409 			spin_unlock(&hb->chain_lock);
410 			return q;
411 		}
412 		depth++;
413 	}
414 	spin_unlock(&hb->chain_lock);
415 
416 	if (depth <= INETFRAGS_MAXDEPTH)
417 		return inet_frag_create(nf, f, key);
418 
419 	if (inet_frag_may_rebuild(f)) {
420 		if (!f->rebuild)
421 			f->rebuild = true;
422 		inet_frag_schedule_worker(f);
423 	}
424 
425 	return ERR_PTR(-ENOBUFS);
426 }
427 EXPORT_SYMBOL(inet_frag_find);
428 
429 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
430 				   const char *prefix)
431 {
432 	static const char msg[] = "inet_frag_find: Fragment hash bucket"
433 		" list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
434 		". Dropping fragment.\n";
435 
436 	if (PTR_ERR(q) == -ENOBUFS)
437 		net_dbg_ratelimited("%s%s", prefix, msg);
438 }
439 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);
440