?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
netfilter: nfnetlink_queue: zero copy support
[linux-2.6.git] / net / netfilter / nfnetlink_queue_core.c
blobda91b8676ddb856c9b458e39fd8214f5590ed60e
1 /*
2 * This is a module which is used for queueing packets and communicating with
3 * userspace via nfnetlink.
4 *
5 * (C) 2005 by Harald Welte <laforge@netfilter.org>
6 * (C) 2007 by Patrick McHardy <kaber@trash.net>
7 *
8 * Based on the old ipv4-only ip_queue.c:
9 * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
10 * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 *
16 */
17 #include <linux/module.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/spinlock.h>
21 #include <linux/slab.h>
22 #include <linux/notifier.h>
23 #include <linux/netdevice.h>
24 #include <linux/netfilter.h>
25 #include <linux/proc_fs.h>
26 #include <linux/netfilter_ipv4.h>
27 #include <linux/netfilter_ipv6.h>
28 #include <linux/netfilter/nfnetlink.h>
29 #include <linux/netfilter/nfnetlink_queue.h>
30 #include <linux/list.h>
31 #include <net/sock.h>
32 #include <net/netfilter/nf_queue.h>
33 #include <net/netfilter/nfnetlink_queue.h>
34
35 #include <linux/atomic.h>
36
37 #ifdef CONFIG_BRIDGE_NETFILTER
38 #include "../bridge/br_private.h"
39 #endif
40
41 #define NFQNL_QMAX_DEFAULT 1024
42
43 struct nfqnl_instance {
44 struct hlist_node hlist; /* global list of queues */
45 struct rcu_head rcu;
46
47 int peer_portid;
48 unsigned int queue_maxlen;
49 unsigned int copy_range;
50 unsigned int queue_dropped;
51 unsigned int queue_user_dropped;
52
53
54 u_int16_t queue_num; /* number of this queue */
55 u_int8_t copy_mode;
56 u_int32_t flags; /* Set using NFQA_CFG_FLAGS */
57 /*
58 * Following fields are dirtied for each queued packet,
59 * keep them in same cache line if possible.
60 */
61 spinlock_t lock;
62 unsigned int queue_total;
63 unsigned int id_sequence; /* 'sequence' of pkt ids */
64 struct list_head queue_list; /* packets in queue */
65 };
66
67 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
68
69 static DEFINE_SPINLOCK(instances_lock);
70
71 #define INSTANCE_BUCKETS 16
72 static struct hlist_head instance_table[INSTANCE_BUCKETS] __read_mostly;
73
74 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
75 {
76 return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
77 }
78
79 static struct nfqnl_instance *
80 instance_lookup(u_int16_t queue_num)
81 {
82 struct hlist_head *head;
83 struct nfqnl_instance *inst;
84
85 head = &instance_table[instance_hashfn(queue_num)];
86 hlist_for_each_entry_rcu(inst, head, hlist) {
87 if (inst->queue_num == queue_num)
88 return inst;
89 }
90 return NULL;
91 }
92
93 static struct nfqnl_instance *
94 instance_create(u_int16_t queue_num, int portid)
95 {
96 struct nfqnl_instance *inst;
97 unsigned int h;
98 int err;
99
100 spin_lock(&instances_lock);
101 if (instance_lookup(queue_num)) {
102 err = -EEXIST;
103 goto out_unlock;
104 }
105
106 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
107 if (!inst) {
108 err = -ENOMEM;
109 goto out_unlock;
110 }
111
112 inst->queue_num = queue_num;
113 inst->peer_portid = portid;
114 inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
115 inst->copy_range = 0xfffff;
116 inst->copy_mode = NFQNL_COPY_NONE;
117 spin_lock_init(&inst->lock);
118 INIT_LIST_HEAD(&inst->queue_list);
119
120 if (!try_module_get(THIS_MODULE)) {
121 err = -EAGAIN;
122 goto out_free;
123 }
124
125 h = instance_hashfn(queue_num);
126 hlist_add_head_rcu(&inst->hlist, &instance_table[h]);
127
128 spin_unlock(&instances_lock);
129
130 return inst;
131
132 out_free:
133 kfree(inst);
134 out_unlock:
135 spin_unlock(&instances_lock);
136 return ERR_PTR(err);
137 }
138
139 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
140 unsigned long data);
141
142 static void
143 instance_destroy_rcu(struct rcu_head *head)
144 {
145 struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
146 rcu);
147
148 nfqnl_flush(inst, NULL, 0);
149 kfree(inst);
150 module_put(THIS_MODULE);
151 }
152
153 static void
154 __instance_destroy(struct nfqnl_instance *inst)
155 {
156 hlist_del_rcu(&inst->hlist);
157 call_rcu(&inst->rcu, instance_destroy_rcu);
158 }
159
160 static void
161 instance_destroy(struct nfqnl_instance *inst)
162 {
163 spin_lock(&instances_lock);
164 __instance_destroy(inst);
165 spin_unlock(&instances_lock);
166 }
167
168 static inline void
169 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
170 {
171 list_add_tail(&entry->list, &queue->queue_list);
172 queue->queue_total++;
173 }
174
175 static void
176 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
177 {
178 list_del(&entry->list);
179 queue->queue_total--;
180 }
181
182 static struct nf_queue_entry *
183 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
184 {
185 struct nf_queue_entry *entry = NULL, *i;
186
187 spin_lock_bh(&queue->lock);
188
189 list_for_each_entry(i, &queue->queue_list, list) {
190 if (i->id == id) {
191 entry = i;
192 break;
193 }
194 }
195
196 if (entry)
197 __dequeue_entry(queue, entry);
198
199 spin_unlock_bh(&queue->lock);
200
201 return entry;
202 }
203
204 static void
205 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
206 {
207 struct nf_queue_entry *entry, *next;
208
209 spin_lock_bh(&queue->lock);
210 list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
211 if (!cmpfn || cmpfn(entry, data)) {
212 list_del(&entry->list);
213 queue->queue_total--;
214 nf_reinject(entry, NF_DROP);
215 }
216 }
217 spin_unlock_bh(&queue->lock);
218 }
219
220 static void
221 nfqnl_zcopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
222 {
223 int i, j = 0;
224 int plen = 0; /* length of skb->head fragment */
225 struct page *page;
226 unsigned int offset;
227
228 /* dont bother with small payloads */
229 if (len <= skb_tailroom(to)) {
230 skb_copy_bits(from, 0, skb_put(to, len), len);
231 return;
232 }
233
234 if (hlen) {
235 skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
236 len -= hlen;
237 } else {
238 plen = min_t(int, skb_headlen(from), len);
239 if (plen) {
240 page = virt_to_head_page(from->head);
241 offset = from->data - (unsigned char *)page_address(page);
242 __skb_fill_page_desc(to, 0, page, offset, plen);
243 get_page(page);
244 j = 1;
245 len -= plen;
246 }
247 }
248
249 to->truesize += len + plen;
250 to->len += len + plen;
251 to->data_len += len + plen;
252
253 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
254 if (!len)
255 break;
256 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
257 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
258 len -= skb_shinfo(to)->frags[j].size;
259 skb_frag_ref(to, j);
260 j++;
261 }
262 skb_shinfo(to)->nr_frags = j;
263 }
264
265 static struct sk_buff *
266 nfqnl_build_packet_message(struct nfqnl_instance *queue,
267 struct nf_queue_entry *entry,
268 __be32 **packet_id_ptr)
269 {
270 size_t size;
271 size_t data_len = 0, cap_len = 0;
272 int hlen = 0;
273 struct sk_buff *skb;
274 struct nlattr *nla;
275 struct nfqnl_msg_packet_hdr *pmsg;
276 struct nlmsghdr *nlh;
277 struct nfgenmsg *nfmsg;
278 struct sk_buff *entskb = entry->skb;
279 struct net_device *indev;
280 struct net_device *outdev;
281 struct nf_conn *ct = NULL;
282 enum ip_conntrack_info uninitialized_var(ctinfo);
283
284 size = NLMSG_SPACE(sizeof(struct nfgenmsg))
285 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
286 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
287 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
288 #ifdef CONFIG_BRIDGE_NETFILTER
289 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
290 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
291 #endif
292 + nla_total_size(sizeof(u_int32_t)) /* mark */
293 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
294 + nla_total_size(sizeof(u_int32_t)); /* cap_len */
295
296 if (entskb->tstamp.tv64)
297 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
298
299 outdev = entry->outdev;
300
301 switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
302 case NFQNL_COPY_META:
303 case NFQNL_COPY_NONE:
304 break;
305
306 case NFQNL_COPY_PACKET:
307 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
308 skb_checksum_help(entskb))
309 return NULL;
310
311 data_len = ACCESS_ONCE(queue->copy_range);
312 if (data_len == 0 || data_len > entskb->len)
313 data_len = entskb->len;
314
315
316 if (!entskb->head_frag ||
317 skb_headlen(entskb) < L1_CACHE_BYTES ||
318 skb_shinfo(entskb)->nr_frags >= MAX_SKB_FRAGS)
319 hlen = skb_headlen(entskb);
320
321 if (skb_has_frag_list(entskb))
322 hlen = entskb->len;
323 hlen = min_t(int, data_len, hlen);
324 size += sizeof(struct nlattr) + hlen;
325 cap_len = entskb->len;
326 break;
327 }
328
329 if (queue->flags & NFQA_CFG_F_CONNTRACK)
330 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
331
332 skb = alloc_skb(size, GFP_ATOMIC);
333 if (!skb)
334 return NULL;
335
336 nlh = nlmsg_put(skb, 0, 0,
337 NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
338 sizeof(struct nfgenmsg), 0);
339 if (!nlh) {
340 kfree_skb(skb);
341 return NULL;
342 }
343 nfmsg = nlmsg_data(nlh);
344 nfmsg->nfgen_family = entry->pf;
345 nfmsg->version = NFNETLINK_V0;
346 nfmsg->res_id = htons(queue->queue_num);
347
348 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
349 pmsg = nla_data(nla);
350 pmsg->hw_protocol = entskb->protocol;
351 pmsg->hook = entry->hook;
352 *packet_id_ptr = &pmsg->packet_id;
353
354 indev = entry->indev;
355 if (indev) {
356 #ifndef CONFIG_BRIDGE_NETFILTER
357 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
358 goto nla_put_failure;
359 #else
360 if (entry->pf == PF_BRIDGE) {
361 /* Case 1: indev is physical input device, we need to
362 * look for bridge group (when called from
363 * netfilter_bridge) */
364 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
365 htonl(indev->ifindex)) ||
366 /* this is the bridge group "brX" */
367 /* rcu_read_lock()ed by __nf_queue */
368 nla_put_be32(skb, NFQA_IFINDEX_INDEV,
369 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
370 goto nla_put_failure;
371 } else {
372 /* Case 2: indev is bridge group, we need to look for
373 * physical device (when called from ipv4) */
374 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
375 htonl(indev->ifindex)))
376 goto nla_put_failure;
377 if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
378 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
379 htonl(entskb->nf_bridge->physindev->ifindex)))
380 goto nla_put_failure;
381 }
382 #endif
383 }
384
385 if (outdev) {
386 #ifndef CONFIG_BRIDGE_NETFILTER
387 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
388 goto nla_put_failure;
389 #else
390 if (entry->pf == PF_BRIDGE) {
391 /* Case 1: outdev is physical output device, we need to
392 * look for bridge group (when called from
393 * netfilter_bridge) */
394 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
395 htonl(outdev->ifindex)) ||
396 /* this is the bridge group "brX" */
397 /* rcu_read_lock()ed by __nf_queue */
398 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
399 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
400 goto nla_put_failure;
401 } else {
402 /* Case 2: outdev is bridge group, we need to look for
403 * physical output device (when called from ipv4) */
404 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
405 htonl(outdev->ifindex)))
406 goto nla_put_failure;
407 if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
408 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
409 htonl(entskb->nf_bridge->physoutdev->ifindex)))
410 goto nla_put_failure;
411 }
412 #endif
413 }
414
415 if (entskb->mark &&
416 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
417 goto nla_put_failure;
418
419 if (indev && entskb->dev &&
420 entskb->mac_header != entskb->network_header) {
421 struct nfqnl_msg_packet_hw phw;
422 int len = dev_parse_header(entskb, phw.hw_addr);
423 if (len) {
424 phw.hw_addrlen = htons(len);
425 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
426 goto nla_put_failure;
427 }
428 }
429
430 if (entskb->tstamp.tv64) {
431 struct nfqnl_msg_packet_timestamp ts;
432 struct timeval tv = ktime_to_timeval(entskb->tstamp);
433 ts.sec = cpu_to_be64(tv.tv_sec);
434 ts.usec = cpu_to_be64(tv.tv_usec);
435
436 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
437 goto nla_put_failure;
438 }
439
440 if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
441 goto nla_put_failure;
442
443 if (cap_len > 0 && nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
444 goto nla_put_failure;
445
446 if (data_len) {
447 struct nlattr *nla;
448
449 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
450 goto nla_put_failure;
451
452 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
453 nla->nla_type = NFQA_PAYLOAD;
454 nla->nla_len = nla_attr_size(data_len);
455
456 nfqnl_zcopy(skb, entskb, data_len, hlen);
457 }
458
459 nlh->nlmsg_len = skb->len;
460 return skb;
461
462 nla_put_failure:
463 kfree_skb(skb);
464 net_err_ratelimited("nf_queue: error creating packet message\n");
465 return NULL;
466 }
467
468 static int
469 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
470 {
471 struct sk_buff *nskb;
472 struct nfqnl_instance *queue;
473 int err = -ENOBUFS;
474 __be32 *packet_id_ptr;
475 int failopen = 0;
476
477 /* rcu_read_lock()ed by nf_hook_slow() */
478 queue = instance_lookup(queuenum);
479 if (!queue) {
480 err = -ESRCH;
481 goto err_out;
482 }
483
484 if (queue->copy_mode == NFQNL_COPY_NONE) {
485 err = -EINVAL;
486 goto err_out;
487 }
488
489 nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
490 if (nskb == NULL) {
491 err = -ENOMEM;
492 goto err_out;
493 }
494 spin_lock_bh(&queue->lock);
495
496 if (!queue->peer_portid) {
497 err = -EINVAL;
498 goto err_out_free_nskb;
499 }
500 if (queue->queue_total >= queue->queue_maxlen) {
501 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
502 failopen = 1;
503 err = 0;
504 } else {
505 queue->queue_dropped++;
506 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
507 queue->queue_total);
508 }
509 goto err_out_free_nskb;
510 }
511 entry->id = ++queue->id_sequence;
512 *packet_id_ptr = htonl(entry->id);
513
514 /* nfnetlink_unicast will either free the nskb or add it to a socket */
515 err = nfnetlink_unicast(nskb, &init_net, queue->peer_portid, MSG_DONTWAIT);
516 if (err < 0) {
517 queue->queue_user_dropped++;
518 goto err_out_unlock;
519 }
520
521 __enqueue_entry(queue, entry);
522
523 spin_unlock_bh(&queue->lock);
524 return 0;
525
526 err_out_free_nskb:
527 kfree_skb(nskb);
528 err_out_unlock:
529 spin_unlock_bh(&queue->lock);
530 if (failopen)
531 nf_reinject(entry, NF_ACCEPT);
532 err_out:
533 return err;
534 }
535
536 static int
537 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
538 {
539 struct sk_buff *nskb;
540
541 if (diff < 0) {
542 if (pskb_trim(e->skb, data_len))
543 return -ENOMEM;
544 } else if (diff > 0) {
545 if (data_len > 0xFFFF)
546 return -EINVAL;
547 if (diff > skb_tailroom(e->skb)) {
548 nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
549 diff, GFP_ATOMIC);
550 if (!nskb) {
551 printk(KERN_WARNING "nf_queue: OOM "
552 "in mangle, dropping packet\n");
553 return -ENOMEM;
554 }
555 kfree_skb(e->skb);
556 e->skb = nskb;
557 }
558 skb_put(e->skb, diff);
559 }
560 if (!skb_make_writable(e->skb, data_len))
561 return -ENOMEM;
562 skb_copy_to_linear_data(e->skb, data, data_len);
563 e->skb->ip_summed = CHECKSUM_NONE;
564 return 0;
565 }
566
567 static int
568 nfqnl_set_mode(struct nfqnl_instance *queue,
569 unsigned char mode, unsigned int range)
570 {
571 int status = 0;
572
573 spin_lock_bh(&queue->lock);
574 switch (mode) {
575 case NFQNL_COPY_NONE:
576 case NFQNL_COPY_META:
577 queue->copy_mode = mode;
578 queue->copy_range = 0;
579 break;
580
581 case NFQNL_COPY_PACKET:
582 queue->copy_mode = mode;
583 /* We're using struct nlattr which has 16bit nla_len. Note that
584 * nla_len includes the header length. Thus, the maximum packet
585 * length that we support is 65531 bytes. We send truncated
586 * packets if the specified length is larger than that.
587 */
588 if (range > 0xffff - NLA_HDRLEN)
589 queue->copy_range = 0xffff - NLA_HDRLEN;
590 else
591 queue->copy_range = range;
592 break;
593
594 default:
595 status = -EINVAL;
596
597 }
598 spin_unlock_bh(&queue->lock);
599
600 return status;
601 }
602
603 static int
604 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
605 {
606 if (entry->indev)
607 if (entry->indev->ifindex == ifindex)
608 return 1;
609 if (entry->outdev)
610 if (entry->outdev->ifindex == ifindex)
611 return 1;
612 #ifdef CONFIG_BRIDGE_NETFILTER
613 if (entry->skb->nf_bridge) {
614 if (entry->skb->nf_bridge->physindev &&
615 entry->skb->nf_bridge->physindev->ifindex == ifindex)
616 return 1;
617 if (entry->skb->nf_bridge->physoutdev &&
618 entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
619 return 1;
620 }
621 #endif
622 return 0;
623 }
624
625 /* drop all packets with either indev or outdev == ifindex from all queue
626 * instances */
627 static void
628 nfqnl_dev_drop(int ifindex)
629 {
630 int i;
631
632 rcu_read_lock();
633
634 for (i = 0; i < INSTANCE_BUCKETS; i++) {
635 struct nfqnl_instance *inst;
636 struct hlist_head *head = &instance_table[i];
637
638 hlist_for_each_entry_rcu(inst, head, hlist)
639 nfqnl_flush(inst, dev_cmp, ifindex);
640 }
641
642 rcu_read_unlock();
643 }
644
645 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
646
647 static int
648 nfqnl_rcv_dev_event(struct notifier_block *this,
649 unsigned long event, void *ptr)
650 {
651 struct net_device *dev = ptr;
652
653 if (!net_eq(dev_net(dev), &init_net))
654 return NOTIFY_DONE;
655
656 /* Drop any packets associated with the downed device */
657 if (event == NETDEV_DOWN)
658 nfqnl_dev_drop(dev->ifindex);
659 return NOTIFY_DONE;
660 }
661
662 static struct notifier_block nfqnl_dev_notifier = {
663 .notifier_call = nfqnl_rcv_dev_event,
664 };
665
666 static int
667 nfqnl_rcv_nl_event(struct notifier_block *this,
668 unsigned long event, void *ptr)
669 {
670 struct netlink_notify *n = ptr;
671
672 if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
673 int i;
674
675 /* destroy all instances for this portid */
676 spin_lock(&instances_lock);
677 for (i = 0; i < INSTANCE_BUCKETS; i++) {
678 struct hlist_node *t2;
679 struct nfqnl_instance *inst;
680 struct hlist_head *head = &instance_table[i];
681
682 hlist_for_each_entry_safe(inst, t2, head, hlist) {
683 if ((n->net == &init_net) &&
684 (n->portid == inst->peer_portid))
685 __instance_destroy(inst);
686 }
687 }
688 spin_unlock(&instances_lock);
689 }
690 return NOTIFY_DONE;
691 }
692
693 static struct notifier_block nfqnl_rtnl_notifier = {
694 .notifier_call = nfqnl_rcv_nl_event,
695 };
696
697 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
698 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
699 [NFQA_MARK] = { .type = NLA_U32 },
700 [NFQA_PAYLOAD] = { .type = NLA_UNSPEC },
701 [NFQA_CT] = { .type = NLA_UNSPEC },
702 };
703
704 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
705 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
706 [NFQA_MARK] = { .type = NLA_U32 },
707 };
708
709 static struct nfqnl_instance *verdict_instance_lookup(u16 queue_num, int nlportid)
710 {
711 struct nfqnl_instance *queue;
712
713 queue = instance_lookup(queue_num);
714 if (!queue)
715 return ERR_PTR(-ENODEV);
716
717 if (queue->peer_portid != nlportid)
718 return ERR_PTR(-EPERM);
719
720 return queue;
721 }
722
723 static struct nfqnl_msg_verdict_hdr*
724 verdicthdr_get(const struct nlattr * const nfqa[])
725 {
726 struct nfqnl_msg_verdict_hdr *vhdr;
727 unsigned int verdict;
728
729 if (!nfqa[NFQA_VERDICT_HDR])
730 return NULL;
731
732 vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
733 verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
734 if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
735 return NULL;
736 return vhdr;
737 }
738
739 static int nfq_id_after(unsigned int id, unsigned int max)
740 {
741 return (int)(id - max) > 0;
742 }
743
744 static int
745 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
746 const struct nlmsghdr *nlh,
747 const struct nlattr * const nfqa[])
748 {
749 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
750 struct nf_queue_entry *entry, *tmp;
751 unsigned int verdict, maxid;
752 struct nfqnl_msg_verdict_hdr *vhdr;
753 struct nfqnl_instance *queue;
754 LIST_HEAD(batch_list);
755 u16 queue_num = ntohs(nfmsg->res_id);
756
757 queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).portid);
758 if (IS_ERR(queue))
759 return PTR_ERR(queue);
760
761 vhdr = verdicthdr_get(nfqa);
762 if (!vhdr)
763 return -EINVAL;
764
765 verdict = ntohl(vhdr->verdict);
766 maxid = ntohl(vhdr->id);
767
768 spin_lock_bh(&queue->lock);
769
770 list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
771 if (nfq_id_after(entry->id, maxid))
772 break;
773 __dequeue_entry(queue, entry);
774 list_add_tail(&entry->list, &batch_list);
775 }
776
777 spin_unlock_bh(&queue->lock);
778
779 if (list_empty(&batch_list))
780 return -ENOENT;
781
782 list_for_each_entry_safe(entry, tmp, &batch_list, list) {
783 if (nfqa[NFQA_MARK])
784 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
785 nf_reinject(entry, verdict);
786 }
787 return 0;
788 }
789
790 static int
791 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
792 const struct nlmsghdr *nlh,
793 const struct nlattr * const nfqa[])
794 {
795 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
796 u_int16_t queue_num = ntohs(nfmsg->res_id);
797
798 struct nfqnl_msg_verdict_hdr *vhdr;
799 struct nfqnl_instance *queue;
800 unsigned int verdict;
801 struct nf_queue_entry *entry;
802 enum ip_conntrack_info uninitialized_var(ctinfo);
803 struct nf_conn *ct = NULL;
804
805 queue = instance_lookup(queue_num);
806 if (!queue)
807
808 queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).portid);
809 if (IS_ERR(queue))
810 return PTR_ERR(queue);
811
812 vhdr = verdicthdr_get(nfqa);
813 if (!vhdr)
814 return -EINVAL;
815
816 verdict = ntohl(vhdr->verdict);
817
818 entry = find_dequeue_entry(queue, ntohl(vhdr->id));
819 if (entry == NULL)
820 return -ENOENT;
821
822 rcu_read_lock();
823 if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
824 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
825
826 if (nfqa[NFQA_PAYLOAD]) {
827 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
828 int diff = payload_len - entry->skb->len;
829
830 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
831 payload_len, entry, diff) < 0)
832 verdict = NF_DROP;
833
834 if (ct)
835 nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
836 }
837 rcu_read_unlock();
838
839 if (nfqa[NFQA_MARK])
840 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
841
842 nf_reinject(entry, verdict);
843 return 0;
844 }
845
846 static int
847 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
848 const struct nlmsghdr *nlh,
849 const struct nlattr * const nfqa[])
850 {
851 return -ENOTSUPP;
852 }
853
854 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
855 [NFQA_CFG_CMD] = { .len = sizeof(struct nfqnl_msg_config_cmd) },
856 [NFQA_CFG_PARAMS] = { .len = sizeof(struct nfqnl_msg_config_params) },
857 };
858
859 static const struct nf_queue_handler nfqh = {
860 .outfn = &nfqnl_enqueue_packet,
861 };
862
863 static int
864 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
865 const struct nlmsghdr *nlh,
866 const struct nlattr * const nfqa[])
867 {
868 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
869 u_int16_t queue_num = ntohs(nfmsg->res_id);
870 struct nfqnl_instance *queue;
871 struct nfqnl_msg_config_cmd *cmd = NULL;
872 int ret = 0;
873
874 if (nfqa[NFQA_CFG_CMD]) {
875 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
876
877 /* Obsolete commands without queue context */
878 switch (cmd->command) {
879 case NFQNL_CFG_CMD_PF_BIND: return 0;
880 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
881 }
882 }
883
884 rcu_read_lock();
885 queue = instance_lookup(queue_num);
886 if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
887 ret = -EPERM;
888 goto err_out_unlock;
889 }
890
891 if (cmd != NULL) {
892 switch (cmd->command) {
893 case NFQNL_CFG_CMD_BIND:
894 if (queue) {
895 ret = -EBUSY;
896 goto err_out_unlock;
897 }
898 queue = instance_create(queue_num, NETLINK_CB(skb).portid);
899 if (IS_ERR(queue)) {
900 ret = PTR_ERR(queue);
901 goto err_out_unlock;
902 }
903 break;
904 case NFQNL_CFG_CMD_UNBIND:
905 if (!queue) {
906 ret = -ENODEV;
907 goto err_out_unlock;
908 }
909 instance_destroy(queue);
910 break;
911 case NFQNL_CFG_CMD_PF_BIND:
912 case NFQNL_CFG_CMD_PF_UNBIND:
913 break;
914 default:
915 ret = -ENOTSUPP;
916 break;
917 }
918 }
919
920 if (nfqa[NFQA_CFG_PARAMS]) {
921 struct nfqnl_msg_config_params *params;
922
923 if (!queue) {
924 ret = -ENODEV;
925 goto err_out_unlock;
926 }
927 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
928 nfqnl_set_mode(queue, params->copy_mode,
929 ntohl(params->copy_range));
930 }
931
932 if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
933 __be32 *queue_maxlen;
934
935 if (!queue) {
936 ret = -ENODEV;
937 goto err_out_unlock;
938 }
939 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
940 spin_lock_bh(&queue->lock);
941 queue->queue_maxlen = ntohl(*queue_maxlen);
942 spin_unlock_bh(&queue->lock);
943 }
944
945 if (nfqa[NFQA_CFG_FLAGS]) {
946 __u32 flags, mask;
947
948 if (!queue) {
949 ret = -ENODEV;
950 goto err_out_unlock;
951 }
952
953 if (!nfqa[NFQA_CFG_MASK]) {
954 /* A mask is needed to specify which flags are being
955 * changed.
956 */
957 ret = -EINVAL;
958 goto err_out_unlock;
959 }
960
961 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
962 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
963
964 if (flags >= NFQA_CFG_F_MAX) {
965 ret = -EOPNOTSUPP;
966 goto err_out_unlock;
967 }
968
969 spin_lock_bh(&queue->lock);
970 queue->flags &= ~mask;
971 queue->flags |= flags & mask;
972 spin_unlock_bh(&queue->lock);
973 }
974
975 err_out_unlock:
976 rcu_read_unlock();
977 return ret;
978 }
979
980 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
981 [NFQNL_MSG_PACKET] = { .call_rcu = nfqnl_recv_unsupp,
982 .attr_count = NFQA_MAX, },
983 [NFQNL_MSG_VERDICT] = { .call_rcu = nfqnl_recv_verdict,
984 .attr_count = NFQA_MAX,
985 .policy = nfqa_verdict_policy },
986 [NFQNL_MSG_CONFIG] = { .call = nfqnl_recv_config,
987 .attr_count = NFQA_CFG_MAX,
988 .policy = nfqa_cfg_policy },
989 [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
990 .attr_count = NFQA_MAX,
991 .policy = nfqa_verdict_batch_policy },
992 };
993
994 static const struct nfnetlink_subsystem nfqnl_subsys = {
995 .name = "nf_queue",
996 .subsys_id = NFNL_SUBSYS_QUEUE,
997 .cb_count = NFQNL_MSG_MAX,
998 .cb = nfqnl_cb,
999 };
1000
1001 #ifdef CONFIG_PROC_FS
1002 struct iter_state {
1003 unsigned int bucket;
1004 };
1005
1006 static struct hlist_node *get_first(struct seq_file *seq)
1007 {
1008 struct iter_state *st = seq->private;
1009
1010 if (!st)
1011 return NULL;
1012
1013 for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1014 if (!hlist_empty(&instance_table[st->bucket]))
1015 return instance_table[st->bucket].first;
1016 }
1017 return NULL;
1018 }
1019
1020 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1021 {
1022 struct iter_state *st = seq->private;
1023
1024 h = h->next;
1025 while (!h) {
1026 if (++st->bucket >= INSTANCE_BUCKETS)
1027 return NULL;
1028
1029 h = instance_table[st->bucket].first;
1030 }
1031 return h;
1032 }
1033
1034 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1035 {
1036 struct hlist_node *head;
1037 head = get_first(seq);
1038
1039 if (head)
1040 while (pos && (head = get_next(seq, head)))
1041 pos--;
1042 return pos ? NULL : head;
1043 }
1044
1045 static void *seq_start(struct seq_file *seq, loff_t *pos)
1046 __acquires(instances_lock)
1047 {
1048 spin_lock(&instances_lock);
1049 return get_idx(seq, *pos);
1050 }
1051
1052 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1053 {
1054 (*pos)++;
1055 return get_next(s, v);
1056 }
1057
1058 static void seq_stop(struct seq_file *s, void *v)
1059 __releases(instances_lock)
1060 {
1061 spin_unlock(&instances_lock);
1062 }
1063
1064 static int seq_show(struct seq_file *s, void *v)
1065 {
1066 const struct nfqnl_instance *inst = v;
1067
1068 return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1069 inst->queue_num,
1070 inst->peer_portid, inst->queue_total,
1071 inst->copy_mode, inst->copy_range,
1072 inst->queue_dropped, inst->queue_user_dropped,
1073 inst->id_sequence, 1);
1074 }
1075
1076 static const struct seq_operations nfqnl_seq_ops = {
1077 .start = seq_start,
1078 .next = seq_next,
1079 .stop = seq_stop,
1080 .show = seq_show,
1081 };
1082
1083 static int nfqnl_open(struct inode *inode, struct file *file)
1084 {
1085 return seq_open_private(file, &nfqnl_seq_ops,
1086 sizeof(struct iter_state));
1087 }
1088
1089 static const struct file_operations nfqnl_file_ops = {
1090 .owner = THIS_MODULE,
1091 .open = nfqnl_open,
1092 .read = seq_read,
1093 .llseek = seq_lseek,
1094 .release = seq_release_private,
1095 };
1096
1097 #endif /* PROC_FS */
1098
1099 static int __init nfnetlink_queue_init(void)
1100 {
1101 int i, status = -ENOMEM;
1102
1103 for (i = 0; i < INSTANCE_BUCKETS; i++)
1104 INIT_HLIST_HEAD(&instance_table[i]);
1105
1106 netlink_register_notifier(&nfqnl_rtnl_notifier);
1107 status = nfnetlink_subsys_register(&nfqnl_subsys);
1108 if (status < 0) {
1109 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1110 goto cleanup_netlink_notifier;
1111 }
1112
1113 #ifdef CONFIG_PROC_FS
1114 if (!proc_create("nfnetlink_queue", 0440,
1115 proc_net_netfilter, &nfqnl_file_ops))
1116 goto cleanup_subsys;
1117 #endif
1118
1119 register_netdevice_notifier(&nfqnl_dev_notifier);
1120 nf_register_queue_handler(&nfqh);
1121 return status;
1122
1123 #ifdef CONFIG_PROC_FS
1124 cleanup_subsys:
1125 nfnetlink_subsys_unregister(&nfqnl_subsys);
1126 #endif
1127 cleanup_netlink_notifier:
1128 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1129 return status;
1130 }
1131
1132 static void __exit nfnetlink_queue_fini(void)
1133 {
1134 nf_unregister_queue_handler();
1135 unregister_netdevice_notifier(&nfqnl_dev_notifier);
1136 #ifdef CONFIG_PROC_FS
1137 remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1138 #endif
1139 nfnetlink_subsys_unregister(&nfqnl_subsys);
1140 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1141
1142 rcu_barrier(); /* Wait for completion of call_rcu()'s */
1143 }
1144
1145 MODULE_DESCRIPTION("netfilter packet queue handler");
1146 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1147 MODULE_LICENSE("GPL");
1148 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1149
1150 module_init(nfnetlink_queue_init);
1151 module_exit(nfnetlink_queue_fini);
Linus' kernel tree