2 * inet fragments management
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.
9 * Authors: Pavel Emelyanov <xemul@openvz.org>
10 * Started as consolidation of ipv4/ip_fragment.c,
11 * ipv6/reassembly. and ipv6 nf conntrack reassembly
14 #include <linux/list.h>
15 #include <linux/spinlock.h>
16 #include <linux/module.h>
17 #include <linux/timer.h>
19 #include <linux/random.h>
20 #include <linux/skbuff.h>
21 #include <linux/rtnetlink.h>
22 #include <linux/slab.h>
25 #include <net/inet_frag.h>
26 #include <net/inet_ecn.h>
28 #define INETFRAGS_EVICT_BUCKETS 128
29 #define INETFRAGS_EVICT_MAX 512
31 /* don't rebuild inetfrag table with new secret more often than this */
32 #define INETFRAGS_MIN_REBUILD_INTERVAL (5 * HZ)
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
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,
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,
53 EXPORT_SYMBOL(ip_frag_ecn_table);
56 inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q)
58 return f->hashfn(q) & (INETFRAGS_HASHSZ - 1);
61 static bool inet_frag_may_rebuild(struct inet_frags *f)
63 return time_after(jiffies,
64 f->last_rebuild_jiffies + INETFRAGS_MIN_REBUILD_INTERVAL);
67 static void inet_frag_secret_rebuild(struct inet_frags *f)
71 /* Per bucket lock NOT needed here, due to write lock protection */
72 write_lock_bh(&f->lock);
74 if (!inet_frag_may_rebuild(f))
77 get_random_bytes(&f->rnd, sizeof(u32));
79 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
80 struct inet_frag_bucket *hb;
81 struct inet_frag_queue *q;
85 hlist_for_each_entry_safe(q, n, &hb->chain, list) {
86 unsigned int hval = inet_frag_hashfn(f, q);
89 struct inet_frag_bucket *hb_dest;
93 /* Relink to new hash chain. */
94 hb_dest = &f->hash[hval];
95 hlist_add_head(&q->list, &hb_dest->chain);
101 f->last_rebuild_jiffies = jiffies;
103 write_unlock_bh(&f->lock);
106 static bool inet_fragq_should_evict(const struct inet_frag_queue *q)
108 return q->net->low_thresh == 0 ||
109 frag_mem_limit(q->net) >= q->net->low_thresh;
113 inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb)
115 struct inet_frag_queue *fq;
116 struct hlist_node *n;
117 unsigned int evicted = 0;
121 spin_lock(&hb->chain_lock);
123 hlist_for_each_entry_safe(fq, n, &hb->chain, list) {
124 if (!inet_fragq_should_evict(fq))
127 if (!del_timer(&fq->timer)) {
128 /* q expiring right now thus increment its refcount so
129 * it won't be freed under us and wait until the timer
130 * has finished executing then destroy it
132 atomic_inc(&fq->refcnt);
133 spin_unlock(&hb->chain_lock);
134 del_timer_sync(&fq->timer);
135 WARN_ON(atomic_read(&fq->refcnt) != 1);
136 inet_frag_put(fq, f);
140 /* suppress xmit of (icmp) error packet */
141 fq->last_in &= ~INET_FRAG_FIRST_IN;
142 fq->last_in |= INET_FRAG_EVICTED;
143 hlist_del(&fq->list);
144 hlist_add_head(&fq->list, &expired);
148 spin_unlock(&hb->chain_lock);
150 hlist_for_each_entry_safe(fq, n, &expired, list)
151 f->frag_expire((unsigned long) fq);
156 static void inet_frag_worker(struct work_struct *work)
158 unsigned int budget = INETFRAGS_EVICT_BUCKETS;
159 unsigned int i, evicted = 0;
160 struct inet_frags *f;
162 f = container_of(work, struct inet_frags, frags_work);
164 BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ);
166 read_lock_bh(&f->lock);
168 for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) {
169 evicted += inet_evict_bucket(f, &f->hash[i]);
170 i = (i + 1) & (INETFRAGS_HASHSZ - 1);
171 if (evicted > INETFRAGS_EVICT_MAX)
177 read_unlock_bh(&f->lock);
178 if (f->rebuild && inet_frag_may_rebuild(f))
179 inet_frag_secret_rebuild(f);
182 static void inet_frag_schedule_worker(struct inet_frags *f)
184 if (unlikely(!work_pending(&f->frags_work)))
185 schedule_work(&f->frags_work);
188 void inet_frags_init(struct inet_frags *f)
192 INIT_WORK(&f->frags_work, inet_frag_worker);
194 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
195 struct inet_frag_bucket *hb = &f->hash[i];
197 spin_lock_init(&hb->chain_lock);
198 INIT_HLIST_HEAD(&hb->chain);
200 rwlock_init(&f->lock);
201 f->last_rebuild_jiffies = 0;
203 EXPORT_SYMBOL(inet_frags_init);
205 void inet_frags_init_net(struct netns_frags *nf)
207 init_frag_mem_limit(nf);
209 EXPORT_SYMBOL(inet_frags_init_net);
211 void inet_frags_fini(struct inet_frags *f)
213 cancel_work_sync(&f->frags_work);
215 EXPORT_SYMBOL(inet_frags_fini);
217 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
223 read_lock_bh(&f->lock);
225 for (i = 0; i < INETFRAGS_HASHSZ ; i++)
226 inet_evict_bucket(f, &f->hash[i]);
228 read_unlock_bh(&f->lock);
230 percpu_counter_destroy(&nf->mem);
232 EXPORT_SYMBOL(inet_frags_exit_net);
234 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
236 struct inet_frag_bucket *hb;
240 hash = inet_frag_hashfn(f, fq);
243 spin_lock(&hb->chain_lock);
244 hlist_del(&fq->list);
245 spin_unlock(&hb->chain_lock);
247 read_unlock(&f->lock);
250 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
252 if (del_timer(&fq->timer))
253 atomic_dec(&fq->refcnt);
255 if (!(fq->last_in & INET_FRAG_COMPLETE)) {
257 atomic_dec(&fq->refcnt);
258 fq->last_in |= INET_FRAG_COMPLETE;
261 EXPORT_SYMBOL(inet_frag_kill);
263 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
271 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f)
274 struct netns_frags *nf;
275 unsigned int sum, sum_truesize = 0;
277 WARN_ON(!(q->last_in & INET_FRAG_COMPLETE));
278 WARN_ON(del_timer(&q->timer) != 0);
280 /* Release all fragment data. */
284 struct sk_buff *xp = fp->next;
286 sum_truesize += fp->truesize;
287 frag_kfree_skb(nf, f, fp);
290 sum = sum_truesize + f->qsize;
291 sub_frag_mem_limit(q, sum);
297 EXPORT_SYMBOL(inet_frag_destroy);
299 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
300 struct inet_frag_queue *qp_in, struct inet_frags *f,
303 struct inet_frag_bucket *hb;
304 struct inet_frag_queue *qp;
307 read_lock(&f->lock); /* Protects against hash rebuild */
309 * While we stayed w/o the lock other CPU could update
310 * the rnd seed, so we need to re-calculate the hash
311 * chain. Fortunatelly the qp_in can be used to get one.
313 hash = inet_frag_hashfn(f, qp_in);
315 spin_lock(&hb->chain_lock);
318 /* With SMP race we have to recheck hash table, because
319 * such entry could be created on other cpu, while we
320 * released the hash bucket lock.
322 hlist_for_each_entry(qp, &hb->chain, list) {
323 if (qp->net == nf && f->match(qp, arg)) {
324 atomic_inc(&qp->refcnt);
325 spin_unlock(&hb->chain_lock);
326 read_unlock(&f->lock);
327 qp_in->last_in |= INET_FRAG_COMPLETE;
328 inet_frag_put(qp_in, f);
334 if (!mod_timer(&qp->timer, jiffies + nf->timeout))
335 atomic_inc(&qp->refcnt);
337 atomic_inc(&qp->refcnt);
338 hlist_add_head(&qp->list, &hb->chain);
340 spin_unlock(&hb->chain_lock);
341 read_unlock(&f->lock);
346 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
347 struct inet_frags *f, void *arg)
349 struct inet_frag_queue *q;
351 if (frag_mem_limit(nf) > nf->high_thresh) {
352 inet_frag_schedule_worker(f);
356 q = kzalloc(f->qsize, GFP_ATOMIC);
361 f->constructor(q, arg);
362 add_frag_mem_limit(q, f->qsize);
364 setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
365 spin_lock_init(&q->lock);
366 atomic_set(&q->refcnt, 1);
371 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
372 struct inet_frags *f, void *arg)
374 struct inet_frag_queue *q;
376 q = inet_frag_alloc(nf, f, arg);
380 return inet_frag_intern(nf, q, f, arg);
383 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
384 struct inet_frags *f, void *key, unsigned int hash)
387 struct inet_frag_bucket *hb;
388 struct inet_frag_queue *q;
391 if (frag_mem_limit(nf) > nf->low_thresh)
392 inet_frag_schedule_worker(f);
394 hash &= (INETFRAGS_HASHSZ - 1);
397 spin_lock(&hb->chain_lock);
398 hlist_for_each_entry(q, &hb->chain, list) {
399 if (q->net == nf && f->match(q, key)) {
400 atomic_inc(&q->refcnt);
401 spin_unlock(&hb->chain_lock);
402 read_unlock(&f->lock);
407 spin_unlock(&hb->chain_lock);
408 read_unlock(&f->lock);
410 if (depth <= INETFRAGS_MAXDEPTH)
411 return inet_frag_create(nf, f, key);
413 if (inet_frag_may_rebuild(f)) {
415 inet_frag_schedule_worker(f);
418 return ERR_PTR(-ENOBUFS);
420 EXPORT_SYMBOL(inet_frag_find);
422 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
425 static const char msg[] = "inet_frag_find: Fragment hash bucket"
426 " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
427 ". Dropping fragment.\n";
429 if (PTR_ERR(q) == -ENOBUFS)
430 LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg);
432 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);