xref: /openbmc/linux/net/ipv4/inet_fragment.c (revision 7b6d864b)
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  */
32 const 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 };
47 EXPORT_SYMBOL(ip_frag_ecn_table);
48 
49 static void inet_frag_secret_rebuild(unsigned long dummy)
50 {
51 	struct inet_frags *f = (struct inet_frags *)dummy;
52 	unsigned long now = jiffies;
53 	int i;
54 
55 	/* Per bucket lock NOT needed here, due to write lock protection */
56 	write_lock(&f->lock);
57 
58 	get_random_bytes(&f->rnd, sizeof(u32));
59 	for (i = 0; i < INETFRAGS_HASHSZ; i++) {
60 		struct inet_frag_bucket *hb;
61 		struct inet_frag_queue *q;
62 		struct hlist_node *n;
63 
64 		hb = &f->hash[i];
65 		hlist_for_each_entry_safe(q, n, &hb->chain, list) {
66 			unsigned int hval = f->hashfn(q);
67 
68 			if (hval != i) {
69 				struct inet_frag_bucket *hb_dest;
70 
71 				hlist_del(&q->list);
72 
73 				/* Relink to new hash chain. */
74 				hb_dest = &f->hash[hval];
75 				hlist_add_head(&q->list, &hb_dest->chain);
76 			}
77 		}
78 	}
79 	write_unlock(&f->lock);
80 
81 	mod_timer(&f->secret_timer, now + f->secret_interval);
82 }
83 
84 void inet_frags_init(struct inet_frags *f)
85 {
86 	int i;
87 
88 	for (i = 0; i < INETFRAGS_HASHSZ; i++) {
89 		struct inet_frag_bucket *hb = &f->hash[i];
90 
91 		spin_lock_init(&hb->chain_lock);
92 		INIT_HLIST_HEAD(&hb->chain);
93 	}
94 	rwlock_init(&f->lock);
95 
96 	f->rnd = (u32) ((totalram_pages ^ (totalram_pages >> 7)) ^
97 				   (jiffies ^ (jiffies >> 6)));
98 
99 	setup_timer(&f->secret_timer, inet_frag_secret_rebuild,
100 			(unsigned long)f);
101 	f->secret_timer.expires = jiffies + f->secret_interval;
102 	add_timer(&f->secret_timer);
103 }
104 EXPORT_SYMBOL(inet_frags_init);
105 
106 void inet_frags_init_net(struct netns_frags *nf)
107 {
108 	nf->nqueues = 0;
109 	init_frag_mem_limit(nf);
110 	INIT_LIST_HEAD(&nf->lru_list);
111 	spin_lock_init(&nf->lru_lock);
112 }
113 EXPORT_SYMBOL(inet_frags_init_net);
114 
115 void inet_frags_fini(struct inet_frags *f)
116 {
117 	del_timer(&f->secret_timer);
118 }
119 EXPORT_SYMBOL(inet_frags_fini);
120 
121 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
122 {
123 	nf->low_thresh = 0;
124 
125 	local_bh_disable();
126 	inet_frag_evictor(nf, f, true);
127 	local_bh_enable();
128 
129 	percpu_counter_destroy(&nf->mem);
130 }
131 EXPORT_SYMBOL(inet_frags_exit_net);
132 
133 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
134 {
135 	struct inet_frag_bucket *hb;
136 	unsigned int hash;
137 
138 	read_lock(&f->lock);
139 	hash = f->hashfn(fq);
140 	hb = &f->hash[hash];
141 
142 	spin_lock(&hb->chain_lock);
143 	hlist_del(&fq->list);
144 	spin_unlock(&hb->chain_lock);
145 
146 	read_unlock(&f->lock);
147 	inet_frag_lru_del(fq);
148 }
149 
150 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
151 {
152 	if (del_timer(&fq->timer))
153 		atomic_dec(&fq->refcnt);
154 
155 	if (!(fq->last_in & INET_FRAG_COMPLETE)) {
156 		fq_unlink(fq, f);
157 		atomic_dec(&fq->refcnt);
158 		fq->last_in |= INET_FRAG_COMPLETE;
159 	}
160 }
161 EXPORT_SYMBOL(inet_frag_kill);
162 
163 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
164 		struct sk_buff *skb)
165 {
166 	if (f->skb_free)
167 		f->skb_free(skb);
168 	kfree_skb(skb);
169 }
170 
171 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
172 					int *work)
173 {
174 	struct sk_buff *fp;
175 	struct netns_frags *nf;
176 	unsigned int sum, sum_truesize = 0;
177 
178 	WARN_ON(!(q->last_in & INET_FRAG_COMPLETE));
179 	WARN_ON(del_timer(&q->timer) != 0);
180 
181 	/* Release all fragment data. */
182 	fp = q->fragments;
183 	nf = q->net;
184 	while (fp) {
185 		struct sk_buff *xp = fp->next;
186 
187 		sum_truesize += fp->truesize;
188 		frag_kfree_skb(nf, f, fp);
189 		fp = xp;
190 	}
191 	sum = sum_truesize + f->qsize;
192 	if (work)
193 		*work -= sum;
194 	sub_frag_mem_limit(q, sum);
195 
196 	if (f->destructor)
197 		f->destructor(q);
198 	kfree(q);
199 
200 }
201 EXPORT_SYMBOL(inet_frag_destroy);
202 
203 int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force)
204 {
205 	struct inet_frag_queue *q;
206 	int work, evicted = 0;
207 
208 	if (!force) {
209 		if (frag_mem_limit(nf) <= nf->high_thresh)
210 			return 0;
211 	}
212 
213 	work = frag_mem_limit(nf) - nf->low_thresh;
214 	while (work > 0) {
215 		spin_lock(&nf->lru_lock);
216 
217 		if (list_empty(&nf->lru_list)) {
218 			spin_unlock(&nf->lru_lock);
219 			break;
220 		}
221 
222 		q = list_first_entry(&nf->lru_list,
223 				struct inet_frag_queue, lru_list);
224 		atomic_inc(&q->refcnt);
225 		/* Remove q from list to avoid several CPUs grabbing it */
226 		list_del_init(&q->lru_list);
227 
228 		spin_unlock(&nf->lru_lock);
229 
230 		spin_lock(&q->lock);
231 		if (!(q->last_in & INET_FRAG_COMPLETE))
232 			inet_frag_kill(q, f);
233 		spin_unlock(&q->lock);
234 
235 		if (atomic_dec_and_test(&q->refcnt))
236 			inet_frag_destroy(q, f, &work);
237 		evicted++;
238 	}
239 
240 	return evicted;
241 }
242 EXPORT_SYMBOL(inet_frag_evictor);
243 
244 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
245 		struct inet_frag_queue *qp_in, struct inet_frags *f,
246 		void *arg)
247 {
248 	struct inet_frag_bucket *hb;
249 	struct inet_frag_queue *qp;
250 #ifdef CONFIG_SMP
251 #endif
252 	unsigned int hash;
253 
254 	read_lock(&f->lock); /* Protects against hash rebuild */
255 	/*
256 	 * While we stayed w/o the lock other CPU could update
257 	 * the rnd seed, so we need to re-calculate the hash
258 	 * chain. Fortunatelly the qp_in can be used to get one.
259 	 */
260 	hash = f->hashfn(qp_in);
261 	hb = &f->hash[hash];
262 	spin_lock(&hb->chain_lock);
263 
264 #ifdef CONFIG_SMP
265 	/* With SMP race we have to recheck hash table, because
266 	 * such entry could be created on other cpu, while we
267 	 * released the hash bucket lock.
268 	 */
269 	hlist_for_each_entry(qp, &hb->chain, list) {
270 		if (qp->net == nf && f->match(qp, arg)) {
271 			atomic_inc(&qp->refcnt);
272 			spin_unlock(&hb->chain_lock);
273 			read_unlock(&f->lock);
274 			qp_in->last_in |= INET_FRAG_COMPLETE;
275 			inet_frag_put(qp_in, f);
276 			return qp;
277 		}
278 	}
279 #endif
280 	qp = qp_in;
281 	if (!mod_timer(&qp->timer, jiffies + nf->timeout))
282 		atomic_inc(&qp->refcnt);
283 
284 	atomic_inc(&qp->refcnt);
285 	hlist_add_head(&qp->list, &hb->chain);
286 	spin_unlock(&hb->chain_lock);
287 	read_unlock(&f->lock);
288 	inet_frag_lru_add(nf, qp);
289 	return qp;
290 }
291 
292 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
293 		struct inet_frags *f, void *arg)
294 {
295 	struct inet_frag_queue *q;
296 
297 	q = kzalloc(f->qsize, GFP_ATOMIC);
298 	if (q == NULL)
299 		return NULL;
300 
301 	q->net = nf;
302 	f->constructor(q, arg);
303 	add_frag_mem_limit(q, f->qsize);
304 
305 	setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
306 	spin_lock_init(&q->lock);
307 	atomic_set(&q->refcnt, 1);
308 	INIT_LIST_HEAD(&q->lru_list);
309 
310 	return q;
311 }
312 
313 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
314 		struct inet_frags *f, void *arg)
315 {
316 	struct inet_frag_queue *q;
317 
318 	q = inet_frag_alloc(nf, f, arg);
319 	if (q == NULL)
320 		return NULL;
321 
322 	return inet_frag_intern(nf, q, f, arg);
323 }
324 
325 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
326 		struct inet_frags *f, void *key, unsigned int hash)
327 	__releases(&f->lock)
328 {
329 	struct inet_frag_bucket *hb;
330 	struct inet_frag_queue *q;
331 	int depth = 0;
332 
333 	hb = &f->hash[hash];
334 
335 	spin_lock(&hb->chain_lock);
336 	hlist_for_each_entry(q, &hb->chain, list) {
337 		if (q->net == nf && f->match(q, key)) {
338 			atomic_inc(&q->refcnt);
339 			spin_unlock(&hb->chain_lock);
340 			read_unlock(&f->lock);
341 			return q;
342 		}
343 		depth++;
344 	}
345 	spin_unlock(&hb->chain_lock);
346 	read_unlock(&f->lock);
347 
348 	if (depth <= INETFRAGS_MAXDEPTH)
349 		return inet_frag_create(nf, f, key);
350 	else
351 		return ERR_PTR(-ENOBUFS);
352 }
353 EXPORT_SYMBOL(inet_frag_find);
354 
355 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
356 				   const char *prefix)
357 {
358 	static const char msg[] = "inet_frag_find: Fragment hash bucket"
359 		" list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
360 		". Dropping fragment.\n";
361 
362 	if (PTR_ERR(q) == -ENOBUFS)
363 		LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg);
364 }
365 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);
366