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 write_seqlock_bh(&f->rnd_seqlock);
73 if (!inet_frag_may_rebuild(f))
76 get_random_bytes(&f->rnd, sizeof(u32));
78 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
79 struct inet_frag_bucket *hb;
80 struct inet_frag_queue *q;
84 spin_lock(&hb->chain_lock);
86 hlist_for_each_entry_safe(q, n, &hb->chain, list) {
87 unsigned int hval = inet_frag_hashfn(f, q);
90 struct inet_frag_bucket *hb_dest;
94 /* Relink to new hash chain. */
95 hb_dest = &f->hash[hval];
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.
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);
111 spin_unlock(&hb->chain_lock);
115 f->last_rebuild_jiffies = jiffies;
117 write_sequnlock_bh(&f->rnd_seqlock);
120 static bool inet_fragq_should_evict(const struct inet_frag_queue *q)
122 return q->net->low_thresh == 0 ||
123 frag_mem_limit(q->net) >= q->net->low_thresh;
127 inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb)
129 struct inet_frag_queue *fq;
130 struct hlist_node *n;
131 unsigned int evicted = 0;
135 spin_lock(&hb->chain_lock);
137 hlist_for_each_entry_safe(fq, n, &hb->chain, list) {
138 if (!inet_fragq_should_evict(fq))
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
146 atomic_inc(&fq->refcnt);
147 spin_unlock(&hb->chain_lock);
148 del_timer_sync(&fq->timer);
149 WARN_ON(atomic_read(&fq->refcnt) != 1);
150 inet_frag_put(fq, f);
154 fq->flags |= INET_FRAG_EVICTED;
155 hlist_del(&fq->list);
156 hlist_add_head(&fq->list, &expired);
160 spin_unlock(&hb->chain_lock);
162 hlist_for_each_entry_safe(fq, n, &expired, list)
163 f->frag_expire((unsigned long) fq);
168 static void inet_frag_worker(struct work_struct *work)
170 unsigned int budget = INETFRAGS_EVICT_BUCKETS;
171 unsigned int i, evicted = 0;
172 struct inet_frags *f;
174 f = container_of(work, struct inet_frags, frags_work);
176 BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ);
180 for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) {
181 evicted += inet_evict_bucket(f, &f->hash[i]);
182 i = (i + 1) & (INETFRAGS_HASHSZ - 1);
183 if (evicted > INETFRAGS_EVICT_MAX)
191 if (f->rebuild && inet_frag_may_rebuild(f))
192 inet_frag_secret_rebuild(f);
195 static void inet_frag_schedule_worker(struct inet_frags *f)
197 if (unlikely(!work_pending(&f->frags_work)))
198 schedule_work(&f->frags_work);
201 int inet_frags_init(struct inet_frags *f)
205 INIT_WORK(&f->frags_work, inet_frag_worker);
207 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
208 struct inet_frag_bucket *hb = &f->hash[i];
210 spin_lock_init(&hb->chain_lock);
211 INIT_HLIST_HEAD(&hb->chain);
214 seqlock_init(&f->rnd_seqlock);
215 f->last_rebuild_jiffies = 0;
216 f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0,
218 if (!f->frags_cachep)
223 EXPORT_SYMBOL(inet_frags_init);
225 void inet_frags_init_net(struct netns_frags *nf)
227 init_frag_mem_limit(nf);
229 EXPORT_SYMBOL(inet_frags_init_net);
231 void inet_frags_fini(struct inet_frags *f)
233 cancel_work_sync(&f->frags_work);
234 kmem_cache_destroy(f->frags_cachep);
236 EXPORT_SYMBOL(inet_frags_fini);
238 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
247 seq = read_seqbegin(&f->rnd_seqlock);
249 for (i = 0; i < INETFRAGS_HASHSZ ; i++)
250 inet_evict_bucket(f, &f->hash[i]);
252 if (read_seqretry(&f->rnd_seqlock, seq))
257 percpu_counter_destroy(&nf->mem);
259 EXPORT_SYMBOL(inet_frags_exit_net);
261 static struct inet_frag_bucket *
262 get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f)
263 __acquires(hb->chain_lock)
265 struct inet_frag_bucket *hb;
266 unsigned int seq, hash;
269 seq = read_seqbegin(&f->rnd_seqlock);
271 hash = inet_frag_hashfn(f, fq);
274 spin_lock(&hb->chain_lock);
275 if (read_seqretry(&f->rnd_seqlock, seq)) {
276 spin_unlock(&hb->chain_lock);
283 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
285 struct inet_frag_bucket *hb;
287 hb = get_frag_bucket_locked(fq, f);
288 hlist_del(&fq->list);
289 spin_unlock(&hb->chain_lock);
292 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
294 if (del_timer(&fq->timer))
295 atomic_dec(&fq->refcnt);
297 if (!(fq->flags & INET_FRAG_COMPLETE)) {
299 atomic_dec(&fq->refcnt);
300 fq->flags |= INET_FRAG_COMPLETE;
303 EXPORT_SYMBOL(inet_frag_kill);
305 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
313 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f)
316 struct netns_frags *nf;
317 unsigned int sum, sum_truesize = 0;
319 WARN_ON(!(q->flags & INET_FRAG_COMPLETE));
320 WARN_ON(del_timer(&q->timer) != 0);
322 /* Release all fragment data. */
326 struct sk_buff *xp = fp->next;
328 sum_truesize += fp->truesize;
329 frag_kfree_skb(nf, f, fp);
332 sum = sum_truesize + f->qsize;
333 sub_frag_mem_limit(q, sum);
337 kmem_cache_free(f->frags_cachep, q);
339 EXPORT_SYMBOL(inet_frag_destroy);
341 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
342 struct inet_frag_queue *qp_in,
343 struct inet_frags *f,
346 struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f);
347 struct inet_frag_queue *qp;
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.
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);
365 if (!mod_timer(&qp->timer, jiffies + nf->timeout))
366 atomic_inc(&qp->refcnt);
368 atomic_inc(&qp->refcnt);
369 hlist_add_head(&qp->list, &hb->chain);
371 spin_unlock(&hb->chain_lock);
376 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
377 struct inet_frags *f,
380 struct inet_frag_queue *q;
382 if (frag_mem_limit(nf) > nf->high_thresh) {
383 inet_frag_schedule_worker(f);
387 q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
392 f->constructor(q, arg);
393 add_frag_mem_limit(q, f->qsize);
395 setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
396 spin_lock_init(&q->lock);
397 atomic_set(&q->refcnt, 1);
402 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
403 struct inet_frags *f,
406 struct inet_frag_queue *q;
408 q = inet_frag_alloc(nf, f, arg);
412 return inet_frag_intern(nf, q, f, arg);
415 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
416 struct inet_frags *f, void *key,
419 struct inet_frag_bucket *hb;
420 struct inet_frag_queue *q;
423 if (frag_mem_limit(nf) > nf->low_thresh)
424 inet_frag_schedule_worker(f);
426 hash &= (INETFRAGS_HASHSZ - 1);
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);
438 spin_unlock(&hb->chain_lock);
440 if (depth <= INETFRAGS_MAXDEPTH)
441 return inet_frag_create(nf, f, key);
443 if (inet_frag_may_rebuild(f)) {
446 inet_frag_schedule_worker(f);
449 return ERR_PTR(-ENOBUFS);
451 EXPORT_SYMBOL(inet_frag_find);
453 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
456 static const char msg[] = "inet_frag_find: Fragment hash bucket"
457 " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
458 ". Dropping fragment.\n";
460 if (PTR_ERR(q) == -ENOBUFS)
461 LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg);
463 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);