xref: /openbmc/linux/net/ipv4/inet_fragment.c (revision 82ced6fd)
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 
23 #include <net/inet_frag.h>
24 
25 static void inet_frag_secret_rebuild(unsigned long dummy)
26 {
27 	struct inet_frags *f = (struct inet_frags *)dummy;
28 	unsigned long now = jiffies;
29 	int i;
30 
31 	write_lock(&f->lock);
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;
36 
37 		hlist_for_each_entry_safe(q, p, n, &f->hash[i], list) {
38 			unsigned int hval = f->hashfn(q);
39 
40 			if (hval != i) {
41 				hlist_del(&q->list);
42 
43 				/* Relink to new hash chain. */
44 				hlist_add_head(&q->list, &f->hash[hval]);
45 			}
46 		}
47 	}
48 	write_unlock(&f->lock);
49 
50 	mod_timer(&f->secret_timer, now + f->secret_interval);
51 }
52 
53 void inet_frags_init(struct inet_frags *f)
54 {
55 	int i;
56 
57 	for (i = 0; i < INETFRAGS_HASHSZ; i++)
58 		INIT_HLIST_HEAD(&f->hash[i]);
59 
60 	rwlock_init(&f->lock);
61 
62 	f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
63 				   (jiffies ^ (jiffies >> 6)));
64 
65 	setup_timer(&f->secret_timer, inet_frag_secret_rebuild,
66 			(unsigned long)f);
67 	f->secret_timer.expires = jiffies + f->secret_interval;
68 	add_timer(&f->secret_timer);
69 }
70 EXPORT_SYMBOL(inet_frags_init);
71 
72 void inet_frags_init_net(struct netns_frags *nf)
73 {
74 	nf->nqueues = 0;
75 	atomic_set(&nf->mem, 0);
76 	INIT_LIST_HEAD(&nf->lru_list);
77 }
78 EXPORT_SYMBOL(inet_frags_init_net);
79 
80 void inet_frags_fini(struct inet_frags *f)
81 {
82 	del_timer(&f->secret_timer);
83 }
84 EXPORT_SYMBOL(inet_frags_fini);
85 
86 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
87 {
88 	nf->low_thresh = 0;
89 
90 	local_bh_disable();
91 	inet_frag_evictor(nf, f);
92 	local_bh_enable();
93 }
94 EXPORT_SYMBOL(inet_frags_exit_net);
95 
96 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
97 {
98 	write_lock(&f->lock);
99 	hlist_del(&fq->list);
100 	list_del(&fq->lru_list);
101 	fq->net->nqueues--;
102 	write_unlock(&f->lock);
103 }
104 
105 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
106 {
107 	if (del_timer(&fq->timer))
108 		atomic_dec(&fq->refcnt);
109 
110 	if (!(fq->last_in & INET_FRAG_COMPLETE)) {
111 		fq_unlink(fq, f);
112 		atomic_dec(&fq->refcnt);
113 		fq->last_in |= INET_FRAG_COMPLETE;
114 	}
115 }
116 
117 EXPORT_SYMBOL(inet_frag_kill);
118 
119 static 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 
131 void 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 }
159 EXPORT_SYMBOL(inet_frag_destroy);
160 
161 int 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 }
191 EXPORT_SYMBOL(inet_frag_evictor);
192 
193 static 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 
237 static 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 
256 static 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 
268 struct 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 }
286 EXPORT_SYMBOL(inet_frag_find);
287