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 #include <net/ip.h>
28 #include <net/ipv6.h>
29
30 /* Use skb->cb to track consecutive/adjacent fragments coming at
31 * the end of the queue. Nodes in the rb-tree queue will
32 * contain "runs" of one or more adjacent fragments.
33 *
34 * Invariants:
35 * - next_frag is NULL at the tail of a "run";
36 * - the head of a "run" has the sum of all fragment lengths in frag_run_len.
37 */
38 struct ipfrag_skb_cb {
39 union {
40 struct inet_skb_parm h4;
41 struct inet6_skb_parm h6;
42 };
43 struct sk_buff *next_frag;
44 int frag_run_len;
45 };
46
47 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
48
fragcb_clear(struct sk_buff * skb)49 static void fragcb_clear(struct sk_buff *skb)
50 {
51 RB_CLEAR_NODE(&skb->rbnode);
52 FRAG_CB(skb)->next_frag = NULL;
53 FRAG_CB(skb)->frag_run_len = skb->len;
54 }
55
56 /* Append skb to the last "run". */
fragrun_append_to_last(struct inet_frag_queue * q,struct sk_buff * skb)57 static void fragrun_append_to_last(struct inet_frag_queue *q,
58 struct sk_buff *skb)
59 {
60 fragcb_clear(skb);
61
62 FRAG_CB(q->last_run_head)->frag_run_len += skb->len;
63 FRAG_CB(q->fragments_tail)->next_frag = skb;
64 q->fragments_tail = skb;
65 }
66
67 /* Create a new "run" with the skb. */
fragrun_create(struct inet_frag_queue * q,struct sk_buff * skb)68 static void fragrun_create(struct inet_frag_queue *q, struct sk_buff *skb)
69 {
70 BUILD_BUG_ON(sizeof(struct ipfrag_skb_cb) > sizeof(skb->cb));
71 fragcb_clear(skb);
72
73 if (q->last_run_head)
74 rb_link_node(&skb->rbnode, &q->last_run_head->rbnode,
75 &q->last_run_head->rbnode.rb_right);
76 else
77 rb_link_node(&skb->rbnode, NULL, &q->rb_fragments.rb_node);
78 rb_insert_color(&skb->rbnode, &q->rb_fragments);
79
80 q->fragments_tail = skb;
81 q->last_run_head = skb;
82 }
83
84 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
85 * Value : 0xff if frame should be dropped.
86 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
87 */
88 const u8 ip_frag_ecn_table[16] = {
89 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
90 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
91 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
92 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
93
94 /* invalid combinations : drop frame */
95 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
96 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
97 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
98 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
99 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
100 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
101 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
102 };
103 EXPORT_SYMBOL(ip_frag_ecn_table);
104
inet_frags_init(struct inet_frags * f)105 int inet_frags_init(struct inet_frags *f)
106 {
107 f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0,
108 NULL);
109 if (!f->frags_cachep)
110 return -ENOMEM;
111
112 return 0;
113 }
114 EXPORT_SYMBOL(inet_frags_init);
115
inet_frags_fini(struct inet_frags * f)116 void inet_frags_fini(struct inet_frags *f)
117 {
118 /* We must wait that all inet_frag_destroy_rcu() have completed. */
119 rcu_barrier();
120
121 kmem_cache_destroy(f->frags_cachep);
122 f->frags_cachep = NULL;
123 }
124 EXPORT_SYMBOL(inet_frags_fini);
125
inet_frags_free_cb(void * ptr,void * arg)126 static void inet_frags_free_cb(void *ptr, void *arg)
127 {
128 struct inet_frag_queue *fq = ptr;
129
130 /* If we can not cancel the timer, it means this frag_queue
131 * is already disappearing, we have nothing to do.
132 * Otherwise, we own a refcount until the end of this function.
133 */
134 if (!del_timer(&fq->timer))
135 return;
136
137 spin_lock_bh(&fq->lock);
138 if (!(fq->flags & INET_FRAG_COMPLETE)) {
139 fq->flags |= INET_FRAG_COMPLETE;
140 refcount_dec(&fq->refcnt);
141 }
142 spin_unlock_bh(&fq->lock);
143
144 inet_frag_put(fq);
145 }
146
inet_frags_exit_net(struct netns_frags * nf)147 void inet_frags_exit_net(struct netns_frags *nf)
148 {
149 nf->high_thresh = 0; /* prevent creation of new frags */
150
151 rhashtable_free_and_destroy(&nf->rhashtable, inet_frags_free_cb, NULL);
152 }
153 EXPORT_SYMBOL(inet_frags_exit_net);
154
inet_frag_kill(struct inet_frag_queue * fq)155 void inet_frag_kill(struct inet_frag_queue *fq)
156 {
157 if (del_timer(&fq->timer))
158 refcount_dec(&fq->refcnt);
159
160 if (!(fq->flags & INET_FRAG_COMPLETE)) {
161 struct netns_frags *nf = fq->net;
162
163 fq->flags |= INET_FRAG_COMPLETE;
164 rhashtable_remove_fast(&nf->rhashtable, &fq->node, nf->f->rhash_params);
165 refcount_dec(&fq->refcnt);
166 }
167 }
168 EXPORT_SYMBOL(inet_frag_kill);
169
inet_frag_destroy_rcu(struct rcu_head * head)170 static void inet_frag_destroy_rcu(struct rcu_head *head)
171 {
172 struct inet_frag_queue *q = container_of(head, struct inet_frag_queue,
173 rcu);
174 struct inet_frags *f = q->net->f;
175
176 if (f->destructor)
177 f->destructor(q);
178 kmem_cache_free(f->frags_cachep, q);
179 }
180
inet_frag_rbtree_purge(struct rb_root * root)181 unsigned int inet_frag_rbtree_purge(struct rb_root *root)
182 {
183 struct rb_node *p = rb_first(root);
184 unsigned int sum = 0;
185
186 while (p) {
187 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
188
189 p = rb_next(p);
190 rb_erase(&skb->rbnode, root);
191 while (skb) {
192 struct sk_buff *next = FRAG_CB(skb)->next_frag;
193
194 sum += skb->truesize;
195 kfree_skb(skb);
196 skb = next;
197 }
198 }
199 return sum;
200 }
201 EXPORT_SYMBOL(inet_frag_rbtree_purge);
202
inet_frag_destroy(struct inet_frag_queue * q)203 void inet_frag_destroy(struct inet_frag_queue *q)
204 {
205 struct sk_buff *fp;
206 struct netns_frags *nf;
207 unsigned int sum, sum_truesize = 0;
208 struct inet_frags *f;
209
210 WARN_ON(!(q->flags & INET_FRAG_COMPLETE));
211 WARN_ON(del_timer(&q->timer) != 0);
212
213 /* Release all fragment data. */
214 fp = q->fragments;
215 nf = q->net;
216 f = nf->f;
217 if (fp) {
218 do {
219 struct sk_buff *xp = fp->next;
220
221 sum_truesize += fp->truesize;
222 kfree_skb(fp);
223 fp = xp;
224 } while (fp);
225 } else {
226 sum_truesize = inet_frag_rbtree_purge(&q->rb_fragments);
227 }
228 sum = sum_truesize + f->qsize;
229
230 call_rcu(&q->rcu, inet_frag_destroy_rcu);
231
232 sub_frag_mem_limit(nf, sum);
233 }
234 EXPORT_SYMBOL(inet_frag_destroy);
235
inet_frag_alloc(struct netns_frags * nf,struct inet_frags * f,void * arg)236 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
237 struct inet_frags *f,
238 void *arg)
239 {
240 struct inet_frag_queue *q;
241
242 if (!nf->high_thresh || frag_mem_limit(nf) > nf->high_thresh)
243 return NULL;
244
245 q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
246 if (!q)
247 return NULL;
248
249 q->net = nf;
250 f->constructor(q, arg);
251 add_frag_mem_limit(nf, f->qsize);
252
253 timer_setup(&q->timer, f->frag_expire, 0);
254 spin_lock_init(&q->lock);
255 refcount_set(&q->refcnt, 3);
256
257 return q;
258 }
259
inet_frag_create(struct netns_frags * nf,void * arg,struct inet_frag_queue ** prev)260 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
261 void *arg,
262 struct inet_frag_queue **prev)
263 {
264 struct inet_frags *f = nf->f;
265 struct inet_frag_queue *q;
266
267 q = inet_frag_alloc(nf, f, arg);
268 if (!q) {
269 *prev = ERR_PTR(-ENOMEM);
270 return NULL;
271 }
272 mod_timer(&q->timer, jiffies + nf->timeout);
273
274 *prev = rhashtable_lookup_get_insert_key(&nf->rhashtable, &q->key,
275 &q->node, f->rhash_params);
276 if (*prev) {
277 q->flags |= INET_FRAG_COMPLETE;
278 inet_frag_kill(q);
279 inet_frag_destroy(q);
280 return NULL;
281 }
282 return q;
283 }
284
285 /* TODO : call from rcu_read_lock() and no longer use refcount_inc_not_zero() */
inet_frag_find(struct netns_frags * nf,void * key)286 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, void *key)
287 {
288 struct inet_frag_queue *fq = NULL, *prev;
289
290 rcu_read_lock();
291
292 prev = rhashtable_lookup(&nf->rhashtable, key, nf->f->rhash_params);
293 if (!prev)
294 fq = inet_frag_create(nf, key, &prev);
295 if (prev && !IS_ERR(prev)) {
296 fq = prev;
297 if (!refcount_inc_not_zero(&fq->refcnt))
298 fq = NULL;
299 }
300 rcu_read_unlock();
301
302 return fq;
303 }
304 EXPORT_SYMBOL(inet_frag_find);
305
inet_frag_queue_insert(struct inet_frag_queue * q,struct sk_buff * skb,int offset,int end)306 int inet_frag_queue_insert(struct inet_frag_queue *q, struct sk_buff *skb,
307 int offset, int end)
308 {
309 struct sk_buff *last = q->fragments_tail;
310
311 /* RFC5722, Section 4, amended by Errata ID : 3089
312 * When reassembling an IPv6 datagram, if
313 * one or more its constituent fragments is determined to be an
314 * overlapping fragment, the entire datagram (and any constituent
315 * fragments) MUST be silently discarded.
316 *
317 * Duplicates, however, should be ignored (i.e. skb dropped, but the
318 * queue/fragments kept for later reassembly).
319 */
320 if (!last)
321 fragrun_create(q, skb); /* First fragment. */
322 else if (last->ip_defrag_offset + last->len < end) {
323 /* This is the common case: skb goes to the end. */
324 /* Detect and discard overlaps. */
325 if (offset < last->ip_defrag_offset + last->len)
326 return IPFRAG_OVERLAP;
327 if (offset == last->ip_defrag_offset + last->len)
328 fragrun_append_to_last(q, skb);
329 else
330 fragrun_create(q, skb);
331 } else {
332 /* Binary search. Note that skb can become the first fragment,
333 * but not the last (covered above).
334 */
335 struct rb_node **rbn, *parent;
336
337 rbn = &q->rb_fragments.rb_node;
338 do {
339 struct sk_buff *curr;
340 int curr_run_end;
341
342 parent = *rbn;
343 curr = rb_to_skb(parent);
344 curr_run_end = curr->ip_defrag_offset +
345 FRAG_CB(curr)->frag_run_len;
346 if (end <= curr->ip_defrag_offset)
347 rbn = &parent->rb_left;
348 else if (offset >= curr_run_end)
349 rbn = &parent->rb_right;
350 else if (offset >= curr->ip_defrag_offset &&
351 end <= curr_run_end)
352 return IPFRAG_DUP;
353 else
354 return IPFRAG_OVERLAP;
355 } while (*rbn);
356 /* Here we have parent properly set, and rbn pointing to
357 * one of its NULL left/right children. Insert skb.
358 */
359 fragcb_clear(skb);
360 rb_link_node(&skb->rbnode, parent, rbn);
361 rb_insert_color(&skb->rbnode, &q->rb_fragments);
362 }
363
364 skb->ip_defrag_offset = offset;
365
366 return IPFRAG_OK;
367 }
368 EXPORT_SYMBOL(inet_frag_queue_insert);
369
inet_frag_reasm_prepare(struct inet_frag_queue * q,struct sk_buff * skb,struct sk_buff * parent)370 void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb,
371 struct sk_buff *parent)
372 {
373 struct sk_buff *fp, *head = skb_rb_first(&q->rb_fragments);
374 struct sk_buff **nextp;
375 int delta;
376
377 if (head != skb) {
378 fp = skb_clone(skb, GFP_ATOMIC);
379 if (!fp)
380 return NULL;
381 FRAG_CB(fp)->next_frag = FRAG_CB(skb)->next_frag;
382 if (RB_EMPTY_NODE(&skb->rbnode))
383 FRAG_CB(parent)->next_frag = fp;
384 else
385 rb_replace_node(&skb->rbnode, &fp->rbnode,
386 &q->rb_fragments);
387 if (q->fragments_tail == skb)
388 q->fragments_tail = fp;
389 skb_morph(skb, head);
390 FRAG_CB(skb)->next_frag = FRAG_CB(head)->next_frag;
391 rb_replace_node(&head->rbnode, &skb->rbnode,
392 &q->rb_fragments);
393 consume_skb(head);
394 head = skb;
395 }
396 WARN_ON(head->ip_defrag_offset != 0);
397
398 delta = -head->truesize;
399
400 /* Head of list must not be cloned. */
401 if (skb_unclone(head, GFP_ATOMIC))
402 return NULL;
403
404 delta += head->truesize;
405 if (delta)
406 add_frag_mem_limit(q->net, delta);
407
408 /* If the first fragment is fragmented itself, we split
409 * it to two chunks: the first with data and paged part
410 * and the second, holding only fragments.
411 */
412 if (skb_has_frag_list(head)) {
413 struct sk_buff *clone;
414 int i, plen = 0;
415
416 clone = alloc_skb(0, GFP_ATOMIC);
417 if (!clone)
418 return NULL;
419 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
420 skb_frag_list_init(head);
421 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
422 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
423 clone->data_len = head->data_len - plen;
424 clone->len = clone->data_len;
425 head->truesize += clone->truesize;
426 clone->csum = 0;
427 clone->ip_summed = head->ip_summed;
428 add_frag_mem_limit(q->net, clone->truesize);
429 skb_shinfo(head)->frag_list = clone;
430 nextp = &clone->next;
431 } else {
432 nextp = &skb_shinfo(head)->frag_list;
433 }
434
435 return nextp;
436 }
437 EXPORT_SYMBOL(inet_frag_reasm_prepare);
438
inet_frag_reasm_finish(struct inet_frag_queue * q,struct sk_buff * head,void * reasm_data)439 void inet_frag_reasm_finish(struct inet_frag_queue *q, struct sk_buff *head,
440 void *reasm_data)
441 {
442 struct sk_buff **nextp = (struct sk_buff **)reasm_data;
443 struct rb_node *rbn;
444 struct sk_buff *fp;
445
446 skb_push(head, head->data - skb_network_header(head));
447
448 /* Traverse the tree in order, to build frag_list. */
449 fp = FRAG_CB(head)->next_frag;
450 rbn = rb_next(&head->rbnode);
451 rb_erase(&head->rbnode, &q->rb_fragments);
452 while (rbn || fp) {
453 /* fp points to the next sk_buff in the current run;
454 * rbn points to the next run.
455 */
456 /* Go through the current run. */
457 while (fp) {
458 *nextp = fp;
459 nextp = &fp->next;
460 fp->prev = NULL;
461 memset(&fp->rbnode, 0, sizeof(fp->rbnode));
462 fp->sk = NULL;
463 head->data_len += fp->len;
464 head->len += fp->len;
465 if (head->ip_summed != fp->ip_summed)
466 head->ip_summed = CHECKSUM_NONE;
467 else if (head->ip_summed == CHECKSUM_COMPLETE)
468 head->csum = csum_add(head->csum, fp->csum);
469 head->truesize += fp->truesize;
470 fp = FRAG_CB(fp)->next_frag;
471 }
472 /* Move to the next run. */
473 if (rbn) {
474 struct rb_node *rbnext = rb_next(rbn);
475
476 fp = rb_to_skb(rbn);
477 rb_erase(rbn, &q->rb_fragments);
478 rbn = rbnext;
479 }
480 }
481 sub_frag_mem_limit(q->net, head->truesize);
482
483 *nextp = NULL;
484 head->next = NULL;
485 head->prev = NULL;
486 head->tstamp = q->stamp;
487 }
488 EXPORT_SYMBOL(inet_frag_reasm_finish);
489
inet_frag_pull_head(struct inet_frag_queue * q)490 struct sk_buff *inet_frag_pull_head(struct inet_frag_queue *q)
491 {
492 struct sk_buff *head;
493
494 if (q->fragments) {
495 head = q->fragments;
496 q->fragments = head->next;
497 } else {
498 struct sk_buff *skb;
499
500 head = skb_rb_first(&q->rb_fragments);
501 if (!head)
502 return NULL;
503 skb = FRAG_CB(head)->next_frag;
504 if (skb)
505 rb_replace_node(&head->rbnode, &skb->rbnode,
506 &q->rb_fragments);
507 else
508 rb_erase(&head->rbnode, &q->rb_fragments);
509 memset(&head->rbnode, 0, sizeof(head->rbnode));
510 barrier();
511 }
512 if (head == q->fragments_tail)
513 q->fragments_tail = NULL;
514
515 sub_frag_mem_limit(q->net, head->truesize);
516
517 return head;
518 }
519 EXPORT_SYMBOL(inet_frag_pull_head);
520