tcp_bbr: reset full pipe detection on loss recovery undo
[linux-2.6-block.git] / net / ipv4 / inet_fragment.c
CommitLineData
7eb95156
PE
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>
321a3a99 19#include <linux/random.h>
1e4b8287
PE
20#include <linux/skbuff.h>
21#include <linux/rtnetlink.h>
5a0e3ad6 22#include <linux/slab.h>
7eb95156 23
5a3da1fe 24#include <net/sock.h>
7eb95156 25#include <net/inet_frag.h>
be991971
HFS
26#include <net/inet_ecn.h>
27
b13d3cbf
FW
28#define INETFRAGS_EVICT_BUCKETS 128
29#define INETFRAGS_EVICT_MAX 512
30
e3a57d18
FW
31/* don't rebuild inetfrag table with new secret more often than this */
32#define INETFRAGS_MIN_REBUILD_INTERVAL (5 * HZ)
33
be991971
HFS
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 */
38const 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};
53EXPORT_SYMBOL(ip_frag_ecn_table);
7eb95156 54
fb3cfe6e
FW
55static unsigned int
56inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q)
57{
58 return f->hashfn(q) & (INETFRAGS_HASHSZ - 1);
59}
60
e3a57d18
FW
61static 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
67static void inet_frag_secret_rebuild(struct inet_frags *f)
321a3a99 68{
321a3a99
PE
69 int i;
70
ab1c724f 71 write_seqlock_bh(&f->rnd_seqlock);
e3a57d18
FW
72
73 if (!inet_frag_may_rebuild(f))
74 goto out;
19952cc4 75
321a3a99 76 get_random_bytes(&f->rnd, sizeof(u32));
e3a57d18 77
321a3a99 78 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
19952cc4 79 struct inet_frag_bucket *hb;
321a3a99 80 struct inet_frag_queue *q;
b67bfe0d 81 struct hlist_node *n;
321a3a99 82
19952cc4 83 hb = &f->hash[i];
ab1c724f
FW
84 spin_lock(&hb->chain_lock);
85
19952cc4 86 hlist_for_each_entry_safe(q, n, &hb->chain, list) {
fb3cfe6e 87 unsigned int hval = inet_frag_hashfn(f, q);
321a3a99
PE
88
89 if (hval != i) {
19952cc4
JDB
90 struct inet_frag_bucket *hb_dest;
91
321a3a99
PE
92 hlist_del(&q->list);
93
94 /* Relink to new hash chain. */
19952cc4 95 hb_dest = &f->hash[hval];
ab1c724f
FW
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);
19952cc4 107 hlist_add_head(&q->list, &hb_dest->chain);
ab1c724f 108 spin_unlock(&hb_dest->chain_lock);
321a3a99
PE
109 }
110 }
ab1c724f 111 spin_unlock(&hb->chain_lock);
321a3a99 112 }
321a3a99 113
e3a57d18
FW
114 f->rebuild = false;
115 f->last_rebuild_jiffies = jiffies;
116out:
ab1c724f 117 write_sequnlock_bh(&f->rnd_seqlock);
321a3a99
PE
118}
119
b13d3cbf
FW
120static 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
126static unsigned int
127inet_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
b13d3cbf
FW
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
5719b296
FW
140 if (!del_timer(&fq->timer))
141 continue;
b13d3cbf 142
d1fe1944 143 hlist_add_head(&fq->list_evictor, &expired);
b13d3cbf
FW
144 ++evicted;
145 }
146
147 spin_unlock(&hb->chain_lock);
148
d1fe1944 149 hlist_for_each_entry_safe(fq, n, &expired, list_evictor)
78802011 150 f->frag_expire(&fq->timer);
b13d3cbf
FW
151
152 return evicted;
153}
154
155static 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
ab1c724f 165 local_bh_disable();
b13d3cbf 166
6aa7de05 167 for (i = READ_ONCE(f->next_bucket); budget; --budget) {
b13d3cbf
FW
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
ab1c724f
FW
176 local_bh_enable();
177
e3a57d18
FW
178 if (f->rebuild && inet_frag_may_rebuild(f))
179 inet_frag_secret_rebuild(f);
b13d3cbf
FW
180}
181
182static 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
d4ad4d22 188int inet_frags_init(struct inet_frags *f)
7eb95156
PE
189{
190 int i;
191
b13d3cbf
FW
192 INIT_WORK(&f->frags_work, inet_frag_worker);
193
19952cc4
JDB
194 for (i = 0; i < INETFRAGS_HASHSZ; i++) {
195 struct inet_frag_bucket *hb = &f->hash[i];
7eb95156 196
19952cc4
JDB
197 spin_lock_init(&hb->chain_lock);
198 INIT_HLIST_HEAD(&hb->chain);
199 }
ab1c724f
FW
200
201 seqlock_init(&f->rnd_seqlock);
e3a57d18 202 f->last_rebuild_jiffies = 0;
d4ad4d22
NA
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;
7eb95156
PE
209}
210EXPORT_SYMBOL(inet_frags_init);
211
212void inet_frags_fini(struct inet_frags *f)
213{
b13d3cbf 214 cancel_work_sync(&f->frags_work);
d4ad4d22 215 kmem_cache_destroy(f->frags_cachep);
7eb95156
PE
216}
217EXPORT_SYMBOL(inet_frags_fini);
277e650d 218
81566e83
PE
219void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
220{
ab1c724f 221 unsigned int seq;
b13d3cbf
FW
222 int i;
223
81566e83 224 nf->low_thresh = 0;
e8e16b70 225
ab1c724f 226evict_again:
5719b296 227 local_bh_disable();
ab1c724f 228 seq = read_seqbegin(&f->rnd_seqlock);
b13d3cbf
FW
229
230 for (i = 0; i < INETFRAGS_HASHSZ ; i++)
231 inet_evict_bucket(f, &f->hash[i]);
232
ab1c724f 233 local_bh_enable();
5719b296
FW
234 cond_resched();
235
236 if (read_seqretry(&f->rnd_seqlock, seq) ||
fb452a1a 237 sum_frag_mem_limit(nf))
5719b296 238 goto evict_again;
81566e83
PE
239}
240EXPORT_SYMBOL(inet_frags_exit_net);
241
ab1c724f
FW
242static struct inet_frag_bucket *
243get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f)
244__acquires(hb->chain_lock)
277e650d 245{
19952cc4 246 struct inet_frag_bucket *hb;
ab1c724f
FW
247 unsigned int seq, hash;
248
249 restart:
250 seq = read_seqbegin(&f->rnd_seqlock);
19952cc4 251
fb3cfe6e 252 hash = inet_frag_hashfn(f, fq);
19952cc4
JDB
253 hb = &f->hash[hash];
254
255 spin_lock(&hb->chain_lock);
ab1c724f
FW
256 if (read_seqretry(&f->rnd_seqlock, seq)) {
257 spin_unlock(&hb->chain_lock);
258 goto restart;
259 }
260
261 return hb;
262}
263
264static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
265{
266 struct inet_frag_bucket *hb;
267
268 hb = get_frag_bucket_locked(fq, f);
d1fe1944 269 hlist_del(&fq->list);
5719b296 270 fq->flags |= INET_FRAG_COMPLETE;
19952cc4 271 spin_unlock(&hb->chain_lock);
277e650d
PE
272}
273
274void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
275{
276 if (del_timer(&fq->timer))
edcb6918 277 refcount_dec(&fq->refcnt);
277e650d 278
06aa8b8a 279 if (!(fq->flags & INET_FRAG_COMPLETE)) {
277e650d 280 fq_unlink(fq, f);
edcb6918 281 refcount_dec(&fq->refcnt);
277e650d
PE
282 }
283}
277e650d 284EXPORT_SYMBOL(inet_frag_kill);
1e4b8287 285
3fd588eb 286void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f)
1e4b8287
PE
287{
288 struct sk_buff *fp;
6ddc0822 289 struct netns_frags *nf;
d433673e 290 unsigned int sum, sum_truesize = 0;
1e4b8287 291
06aa8b8a 292 WARN_ON(!(q->flags & INET_FRAG_COMPLETE));
547b792c 293 WARN_ON(del_timer(&q->timer) != 0);
1e4b8287
PE
294
295 /* Release all fragment data. */
296 fp = q->fragments;
6ddc0822 297 nf = q->net;
1e4b8287
PE
298 while (fp) {
299 struct sk_buff *xp = fp->next;
300
d433673e 301 sum_truesize += fp->truesize;
a72a5e2d 302 kfree_skb(fp);
1e4b8287
PE
303 fp = xp;
304 }
d433673e 305 sum = sum_truesize + f->qsize;
1e4b8287 306
c9547709
PE
307 if (f->destructor)
308 f->destructor(q);
d4ad4d22 309 kmem_cache_free(f->frags_cachep, q);
5719b296
FW
310
311 sub_frag_mem_limit(nf, sum);
1e4b8287
PE
312}
313EXPORT_SYMBOL(inet_frag_destroy);
8e7999c4 314
ac18e750 315static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
f926e236
NA
316 struct inet_frag_queue *qp_in,
317 struct inet_frags *f,
318 void *arg)
2588fe1d 319{
ab1c724f 320 struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f);
2588fe1d 321 struct inet_frag_queue *qp;
19952cc4 322
2588fe1d
PE
323#ifdef CONFIG_SMP
324 /* With SMP race we have to recheck hash table, because
ab1c724f
FW
325 * such entry could have been created on other cpu before
326 * we acquired hash bucket lock.
2588fe1d 327 */
19952cc4 328 hlist_for_each_entry(qp, &hb->chain, list) {
ac18e750 329 if (qp->net == nf && f->match(qp, arg)) {
edcb6918 330 refcount_inc(&qp->refcnt);
19952cc4 331 spin_unlock(&hb->chain_lock);
06aa8b8a 332 qp_in->flags |= INET_FRAG_COMPLETE;
2588fe1d
PE
333 inet_frag_put(qp_in, f);
334 return qp;
335 }
336 }
337#endif
338 qp = qp_in;
b2fd5321 339 if (!mod_timer(&qp->timer, jiffies + nf->timeout))
edcb6918 340 refcount_inc(&qp->refcnt);
2588fe1d 341
edcb6918 342 refcount_inc(&qp->refcnt);
19952cc4 343 hlist_add_head(&qp->list, &hb->chain);
3fd588eb 344
19952cc4 345 spin_unlock(&hb->chain_lock);
24b9bf43 346
2588fe1d
PE
347 return qp;
348}
e521db9d 349
ac18e750 350static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
f926e236
NA
351 struct inet_frags *f,
352 void *arg)
e521db9d
PE
353{
354 struct inet_frag_queue *q;
355
30759219 356 if (!nf->high_thresh || frag_mem_limit(nf) > nf->high_thresh) {
b13d3cbf 357 inet_frag_schedule_worker(f);
86e93e47 358 return NULL;
b13d3cbf 359 }
86e93e47 360
d4ad4d22 361 q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
51456b29 362 if (!q)
e521db9d
PE
363 return NULL;
364
54db0cc2 365 q->net = nf;
c6fda282 366 f->constructor(q, arg);
0e60d245 367 add_frag_mem_limit(nf, f->qsize);
d433673e 368
78802011 369 timer_setup(&q->timer, f->frag_expire, 0);
e521db9d 370 spin_lock_init(&q->lock);
edcb6918 371 refcount_set(&q->refcnt, 1);
e521db9d
PE
372
373 return q;
374}
c6fda282 375
ac18e750 376static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
f926e236
NA
377 struct inet_frags *f,
378 void *arg)
c6fda282
PE
379{
380 struct inet_frag_queue *q;
381
ac18e750 382 q = inet_frag_alloc(nf, f, arg);
51456b29 383 if (!q)
c6fda282
PE
384 return NULL;
385
9a375803 386 return inet_frag_intern(nf, q, f, arg);
c6fda282 387}
abd6523d 388
ac18e750 389struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
f926e236
NA
390 struct inet_frags *f, void *key,
391 unsigned int hash)
abd6523d 392{
19952cc4 393 struct inet_frag_bucket *hb;
abd6523d 394 struct inet_frag_queue *q;
5a3da1fe 395 int depth = 0;
abd6523d 396
b13d3cbf
FW
397 if (frag_mem_limit(nf) > nf->low_thresh)
398 inet_frag_schedule_worker(f);
86e93e47 399
fb3cfe6e 400 hash &= (INETFRAGS_HASHSZ - 1);
19952cc4
JDB
401 hb = &f->hash[hash];
402
403 spin_lock(&hb->chain_lock);
404 hlist_for_each_entry(q, &hb->chain, list) {
ac18e750 405 if (q->net == nf && f->match(q, key)) {
edcb6918 406 refcount_inc(&q->refcnt);
19952cc4 407 spin_unlock(&hb->chain_lock);
abd6523d
PE
408 return q;
409 }
5a3da1fe 410 depth++;
abd6523d 411 }
19952cc4 412 spin_unlock(&hb->chain_lock);
abd6523d 413
5a3da1fe
HFS
414 if (depth <= INETFRAGS_MAXDEPTH)
415 return inet_frag_create(nf, f, key);
e3a57d18
FW
416
417 if (inet_frag_may_rebuild(f)) {
ab1c724f
FW
418 if (!f->rebuild)
419 f->rebuild = true;
e3a57d18
FW
420 inet_frag_schedule_worker(f);
421 }
422
423 return ERR_PTR(-ENOBUFS);
abd6523d
PE
424}
425EXPORT_SYMBOL(inet_frag_find);
5a3da1fe
HFS
426
427void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
428 const char *prefix)
429{
430 static const char msg[] = "inet_frag_find: Fragment hash bucket"
431 " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
432 ". Dropping fragment.\n";
433
434 if (PTR_ERR(q) == -ENOBUFS)
ba7a46f1 435 net_dbg_ratelimited("%s%s", prefix, msg);
5a3da1fe
HFS
436}
437EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);