[NETFILTER]: nf_queue: move queueing related functions/struct to seperate header
[linux-3.10.git] / net / netfilter / nfnetlink_queue.c
1 /*
2  * This is a module which is used for queueing packets and communicating with
3  * userspace via nfetlink.
4  *
5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
6  *
7  * Based on the old ipv4-only ip_queue.c:
8  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
9  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of the GNU General Public License version 2 as
13  * published by the Free Software Foundation.
14  *
15  */
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/init.h>
19 #include <linux/spinlock.h>
20 #include <linux/notifier.h>
21 #include <linux/netdevice.h>
22 #include <linux/netfilter.h>
23 #include <linux/proc_fs.h>
24 #include <linux/netfilter_ipv4.h>
25 #include <linux/netfilter_ipv6.h>
26 #include <linux/netfilter/nfnetlink.h>
27 #include <linux/netfilter/nfnetlink_queue.h>
28 #include <linux/list.h>
29 #include <net/sock.h>
30 #include <net/netfilter/nf_queue.h>
31
32 #include <asm/atomic.h>
33
34 #ifdef CONFIG_BRIDGE_NETFILTER
35 #include "../bridge/br_private.h"
36 #endif
37
38 #define NFQNL_QMAX_DEFAULT 1024
39
40 #if 0
41 #define QDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
42                                         __FILE__, __LINE__, __FUNCTION__,  \
43                                         ## args)
44 #else
45 #define QDEBUG(x, ...)
46 #endif
47
48 struct nfqnl_queue_entry {
49         struct list_head list;
50         struct nf_info *info;
51         struct sk_buff *skb;
52         unsigned int id;
53 };
54
55 struct nfqnl_instance {
56         struct hlist_node hlist;                /* global list of queues */
57         atomic_t use;
58
59         int peer_pid;
60         unsigned int queue_maxlen;
61         unsigned int copy_range;
62         unsigned int queue_total;
63         unsigned int queue_dropped;
64         unsigned int queue_user_dropped;
65
66         atomic_t id_sequence;                   /* 'sequence' of pkt ids */
67
68         u_int16_t queue_num;                    /* number of this queue */
69         u_int8_t copy_mode;
70
71         spinlock_t lock;
72
73         struct list_head queue_list;            /* packets in queue */
74 };
75
76 typedef int (*nfqnl_cmpfn)(struct nfqnl_queue_entry *, unsigned long);
77
78 static DEFINE_RWLOCK(instances_lock);
79
80 #define INSTANCE_BUCKETS        16
81 static struct hlist_head instance_table[INSTANCE_BUCKETS];
82
83 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
84 {
85         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
86 }
87
88 static struct nfqnl_instance *
89 __instance_lookup(u_int16_t queue_num)
90 {
91         struct hlist_head *head;
92         struct hlist_node *pos;
93         struct nfqnl_instance *inst;
94
95         head = &instance_table[instance_hashfn(queue_num)];
96         hlist_for_each_entry(inst, pos, head, hlist) {
97                 if (inst->queue_num == queue_num)
98                         return inst;
99         }
100         return NULL;
101 }
102
103 static struct nfqnl_instance *
104 instance_lookup_get(u_int16_t queue_num)
105 {
106         struct nfqnl_instance *inst;
107
108         read_lock_bh(&instances_lock);
109         inst = __instance_lookup(queue_num);
110         if (inst)
111                 atomic_inc(&inst->use);
112         read_unlock_bh(&instances_lock);
113
114         return inst;
115 }
116
117 static void
118 instance_put(struct nfqnl_instance *inst)
119 {
120         if (inst && atomic_dec_and_test(&inst->use)) {
121                 QDEBUG("kfree(inst=%p)\n", inst);
122                 kfree(inst);
123         }
124 }
125
126 static struct nfqnl_instance *
127 instance_create(u_int16_t queue_num, int pid)
128 {
129         struct nfqnl_instance *inst;
130
131         QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
132
133         write_lock_bh(&instances_lock);
134         if (__instance_lookup(queue_num)) {
135                 inst = NULL;
136                 QDEBUG("aborting, instance already exists\n");
137                 goto out_unlock;
138         }
139
140         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
141         if (!inst)
142                 goto out_unlock;
143
144         inst->queue_num = queue_num;
145         inst->peer_pid = pid;
146         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
147         inst->copy_range = 0xfffff;
148         inst->copy_mode = NFQNL_COPY_NONE;
149         atomic_set(&inst->id_sequence, 0);
150         /* needs to be two, since we _put() after creation */
151         atomic_set(&inst->use, 2);
152         spin_lock_init(&inst->lock);
153         INIT_LIST_HEAD(&inst->queue_list);
154
155         if (!try_module_get(THIS_MODULE))
156                 goto out_free;
157
158         hlist_add_head(&inst->hlist,
159                        &instance_table[instance_hashfn(queue_num)]);
160
161         write_unlock_bh(&instances_lock);
162
163         QDEBUG("successfully created new instance\n");
164
165         return inst;
166
167 out_free:
168         kfree(inst);
169 out_unlock:
170         write_unlock_bh(&instances_lock);
171         return NULL;
172 }
173
174 static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
175
176 static void
177 _instance_destroy2(struct nfqnl_instance *inst, int lock)
178 {
179         /* first pull it out of the global list */
180         if (lock)
181                 write_lock_bh(&instances_lock);
182
183         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
184                 inst, inst->queue_num);
185         hlist_del(&inst->hlist);
186
187         if (lock)
188                 write_unlock_bh(&instances_lock);
189
190         /* then flush all pending skbs from the queue */
191         nfqnl_flush(inst, NF_DROP);
192
193         /* and finally put the refcount */
194         instance_put(inst);
195
196         module_put(THIS_MODULE);
197 }
198
199 static inline void
200 __instance_destroy(struct nfqnl_instance *inst)
201 {
202         _instance_destroy2(inst, 0);
203 }
204
205 static inline void
206 instance_destroy(struct nfqnl_instance *inst)
207 {
208         _instance_destroy2(inst, 1);
209 }
210
211
212
213 static void
214 issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
215 {
216         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
217
218         /* TCP input path (and probably other bits) assume to be called
219          * from softirq context, not from syscall, like issue_verdict is
220          * called.  TCP input path deadlocks with locks taken from timer
221          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
222
223         local_bh_disable();
224         nf_reinject(entry->skb, entry->info, verdict);
225         local_bh_enable();
226
227         kfree(entry);
228 }
229
230 static inline void
231 __enqueue_entry(struct nfqnl_instance *queue,
232                       struct nfqnl_queue_entry *entry)
233 {
234        list_add(&entry->list, &queue->queue_list);
235        queue->queue_total++;
236 }
237
238 /*
239  * Find and return a queued entry matched by cmpfn, or return the last
240  * entry if cmpfn is NULL.
241  */
242 static inline struct nfqnl_queue_entry *
243 __find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
244                    unsigned long data)
245 {
246         struct list_head *p;
247
248         list_for_each_prev(p, &queue->queue_list) {
249                 struct nfqnl_queue_entry *entry = (struct nfqnl_queue_entry *)p;
250
251                 if (!cmpfn || cmpfn(entry, data))
252                         return entry;
253         }
254         return NULL;
255 }
256
257 static inline void
258 __dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
259 {
260         list_del(&entry->list);
261         q->queue_total--;
262 }
263
264 static inline struct nfqnl_queue_entry *
265 __find_dequeue_entry(struct nfqnl_instance *queue,
266                      nfqnl_cmpfn cmpfn, unsigned long data)
267 {
268         struct nfqnl_queue_entry *entry;
269
270         entry = __find_entry(queue, cmpfn, data);
271         if (entry == NULL)
272                 return NULL;
273
274         __dequeue_entry(queue, entry);
275         return entry;
276 }
277
278
279 static inline void
280 __nfqnl_flush(struct nfqnl_instance *queue, int verdict)
281 {
282         struct nfqnl_queue_entry *entry;
283
284         while ((entry = __find_dequeue_entry(queue, NULL, 0)))
285                 issue_verdict(entry, verdict);
286 }
287
288 static inline int
289 __nfqnl_set_mode(struct nfqnl_instance *queue,
290                  unsigned char mode, unsigned int range)
291 {
292         int status = 0;
293
294         switch (mode) {
295         case NFQNL_COPY_NONE:
296         case NFQNL_COPY_META:
297                 queue->copy_mode = mode;
298                 queue->copy_range = 0;
299                 break;
300
301         case NFQNL_COPY_PACKET:
302                 queue->copy_mode = mode;
303                 /* we're using struct nlattr which has 16bit nla_len */
304                 if (range > 0xffff)
305                         queue->copy_range = 0xffff;
306                 else
307                         queue->copy_range = range;
308                 break;
309
310         default:
311                 status = -EINVAL;
312
313         }
314         return status;
315 }
316
317 static struct nfqnl_queue_entry *
318 find_dequeue_entry(struct nfqnl_instance *queue,
319                          nfqnl_cmpfn cmpfn, unsigned long data)
320 {
321         struct nfqnl_queue_entry *entry;
322
323         spin_lock_bh(&queue->lock);
324         entry = __find_dequeue_entry(queue, cmpfn, data);
325         spin_unlock_bh(&queue->lock);
326
327         return entry;
328 }
329
330 static void
331 nfqnl_flush(struct nfqnl_instance *queue, int verdict)
332 {
333         spin_lock_bh(&queue->lock);
334         __nfqnl_flush(queue, verdict);
335         spin_unlock_bh(&queue->lock);
336 }
337
338 static struct sk_buff *
339 nfqnl_build_packet_message(struct nfqnl_instance *queue,
340                            struct nfqnl_queue_entry *entry, int *errp)
341 {
342         sk_buff_data_t old_tail;
343         size_t size;
344         size_t data_len = 0;
345         struct sk_buff *skb;
346         struct nfqnl_msg_packet_hdr pmsg;
347         struct nlmsghdr *nlh;
348         struct nfgenmsg *nfmsg;
349         struct nf_info *entinf = entry->info;
350         struct sk_buff *entskb = entry->skb;
351         struct net_device *indev;
352         struct net_device *outdev;
353         __be32 tmp_uint;
354
355         QDEBUG("entered\n");
356
357         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
358                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
359                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
360                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
361 #ifdef CONFIG_BRIDGE_NETFILTER
362                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
363                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
364 #endif
365                 + nla_total_size(sizeof(u_int32_t))     /* mark */
366                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
367                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
368
369         outdev = entinf->outdev;
370
371         spin_lock_bh(&queue->lock);
372
373         switch (queue->copy_mode) {
374         case NFQNL_COPY_META:
375         case NFQNL_COPY_NONE:
376                 data_len = 0;
377                 break;
378
379         case NFQNL_COPY_PACKET:
380                 if ((entskb->ip_summed == CHECKSUM_PARTIAL ||
381                      entskb->ip_summed == CHECKSUM_COMPLETE) &&
382                     (*errp = skb_checksum_help(entskb))) {
383                         spin_unlock_bh(&queue->lock);
384                         return NULL;
385                 }
386                 if (queue->copy_range == 0
387                     || queue->copy_range > entskb->len)
388                         data_len = entskb->len;
389                 else
390                         data_len = queue->copy_range;
391
392                 size += nla_total_size(data_len);
393                 break;
394
395         default:
396                 *errp = -EINVAL;
397                 spin_unlock_bh(&queue->lock);
398                 return NULL;
399         }
400
401         spin_unlock_bh(&queue->lock);
402
403         skb = alloc_skb(size, GFP_ATOMIC);
404         if (!skb)
405                 goto nlmsg_failure;
406
407         old_tail = skb->tail;
408         nlh = NLMSG_PUT(skb, 0, 0,
409                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
410                         sizeof(struct nfgenmsg));
411         nfmsg = NLMSG_DATA(nlh);
412         nfmsg->nfgen_family = entinf->pf;
413         nfmsg->version = NFNETLINK_V0;
414         nfmsg->res_id = htons(queue->queue_num);
415
416         pmsg.packet_id          = htonl(entry->id);
417         pmsg.hw_protocol        = entskb->protocol;
418         pmsg.hook               = entinf->hook;
419
420         NLA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
421
422         indev = entinf->indev;
423         if (indev) {
424                 tmp_uint = htonl(indev->ifindex);
425 #ifndef CONFIG_BRIDGE_NETFILTER
426                 NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
427 #else
428                 if (entinf->pf == PF_BRIDGE) {
429                         /* Case 1: indev is physical input device, we need to
430                          * look for bridge group (when called from
431                          * netfilter_bridge) */
432                         NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint),
433                                 &tmp_uint);
434                         /* this is the bridge group "brX" */
435                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
436                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
437                                 &tmp_uint);
438                 } else {
439                         /* Case 2: indev is bridge group, we need to look for
440                          * physical device (when called from ipv4) */
441                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
442                                 &tmp_uint);
443                         if (entskb->nf_bridge
444                             && entskb->nf_bridge->physindev) {
445                                 tmp_uint = htonl(entskb->nf_bridge->physindev->ifindex);
446                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
447                                         sizeof(tmp_uint), &tmp_uint);
448                         }
449                 }
450 #endif
451         }
452
453         if (outdev) {
454                 tmp_uint = htonl(outdev->ifindex);
455 #ifndef CONFIG_BRIDGE_NETFILTER
456                 NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
457 #else
458                 if (entinf->pf == PF_BRIDGE) {
459                         /* Case 1: outdev is physical output device, we need to
460                          * look for bridge group (when called from
461                          * netfilter_bridge) */
462                         NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
463                                 &tmp_uint);
464                         /* this is the bridge group "brX" */
465                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
466                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
467                                 &tmp_uint);
468                 } else {
469                         /* Case 2: outdev is bridge group, we need to look for
470                          * physical output device (when called from ipv4) */
471                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
472                                 &tmp_uint);
473                         if (entskb->nf_bridge
474                             && entskb->nf_bridge->physoutdev) {
475                                 tmp_uint = htonl(entskb->nf_bridge->physoutdev->ifindex);
476                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
477                                         sizeof(tmp_uint), &tmp_uint);
478                         }
479                 }
480 #endif
481         }
482
483         if (entskb->mark) {
484                 tmp_uint = htonl(entskb->mark);
485                 NLA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
486         }
487
488         if (indev && entskb->dev) {
489                 struct nfqnl_msg_packet_hw phw;
490                 int len = dev_parse_header(entskb, phw.hw_addr);
491                 if (len) {
492                         phw.hw_addrlen = htons(len);
493                         NLA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
494                 }
495         }
496
497         if (entskb->tstamp.tv64) {
498                 struct nfqnl_msg_packet_timestamp ts;
499                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
500                 ts.sec = cpu_to_be64(tv.tv_sec);
501                 ts.usec = cpu_to_be64(tv.tv_usec);
502
503                 NLA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
504         }
505
506         if (data_len) {
507                 struct nlattr *nla;
508                 int size = nla_attr_size(data_len);
509
510                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
511                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
512                         goto nlmsg_failure;
513                 }
514
515                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
516                 nla->nla_type = NFQA_PAYLOAD;
517                 nla->nla_len = size;
518
519                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
520                         BUG();
521         }
522
523         nlh->nlmsg_len = skb->tail - old_tail;
524         return skb;
525
526 nlmsg_failure:
527 nla_put_failure:
528         if (skb)
529                 kfree_skb(skb);
530         *errp = -EINVAL;
531         if (net_ratelimit())
532                 printk(KERN_ERR "nf_queue: error creating packet message\n");
533         return NULL;
534 }
535
536 static int
537 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info,
538                      unsigned int queuenum)
539 {
540         int status = -EINVAL;
541         struct sk_buff *nskb;
542         struct nfqnl_instance *queue;
543         struct nfqnl_queue_entry *entry;
544
545         QDEBUG("entered\n");
546
547         queue = instance_lookup_get(queuenum);
548         if (!queue) {
549                 QDEBUG("no queue instance matching\n");
550                 return -EINVAL;
551         }
552
553         if (queue->copy_mode == NFQNL_COPY_NONE) {
554                 QDEBUG("mode COPY_NONE, aborting\n");
555                 status = -EAGAIN;
556                 goto err_out_put;
557         }
558
559         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
560         if (entry == NULL) {
561                 if (net_ratelimit())
562                         printk(KERN_ERR
563                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
564                 status = -ENOMEM;
565                 goto err_out_put;
566         }
567
568         entry->info = info;
569         entry->skb = skb;
570         entry->id = atomic_inc_return(&queue->id_sequence);
571
572         nskb = nfqnl_build_packet_message(queue, entry, &status);
573         if (nskb == NULL)
574                 goto err_out_free;
575
576         spin_lock_bh(&queue->lock);
577
578         if (!queue->peer_pid)
579                 goto err_out_free_nskb;
580
581         if (queue->queue_total >= queue->queue_maxlen) {
582                 queue->queue_dropped++;
583                 status = -ENOSPC;
584                 if (net_ratelimit())
585                           printk(KERN_WARNING "nf_queue: full at %d entries, "
586                                  "dropping packets(s). Dropped: %d\n",
587                                  queue->queue_total, queue->queue_dropped);
588                 goto err_out_free_nskb;
589         }
590
591         /* nfnetlink_unicast will either free the nskb or add it to a socket */
592         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
593         if (status < 0) {
594                 queue->queue_user_dropped++;
595                 goto err_out_unlock;
596         }
597
598         __enqueue_entry(queue, entry);
599
600         spin_unlock_bh(&queue->lock);
601         instance_put(queue);
602         return status;
603
604 err_out_free_nskb:
605         kfree_skb(nskb);
606
607 err_out_unlock:
608         spin_unlock_bh(&queue->lock);
609
610 err_out_free:
611         kfree(entry);
612 err_out_put:
613         instance_put(queue);
614         return status;
615 }
616
617 static int
618 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
619 {
620         int diff;
621         int err;
622
623         diff = data_len - e->skb->len;
624         if (diff < 0) {
625                 if (pskb_trim(e->skb, data_len))
626                         return -ENOMEM;
627         } else if (diff > 0) {
628                 if (data_len > 0xFFFF)
629                         return -EINVAL;
630                 if (diff > skb_tailroom(e->skb)) {
631                         err = pskb_expand_head(e->skb, 0,
632                                                diff - skb_tailroom(e->skb),
633                                                GFP_ATOMIC);
634                         if (err) {
635                                 printk(KERN_WARNING "nf_queue: OOM "
636                                       "in mangle, dropping packet\n");
637                                 return err;
638                         }
639                 }
640                 skb_put(e->skb, diff);
641         }
642         if (!skb_make_writable(e->skb, data_len))
643                 return -ENOMEM;
644         skb_copy_to_linear_data(e->skb, data, data_len);
645         e->skb->ip_summed = CHECKSUM_NONE;
646         return 0;
647 }
648
649 static inline int
650 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
651 {
652         return (id == e->id);
653 }
654
655 static int
656 nfqnl_set_mode(struct nfqnl_instance *queue,
657                unsigned char mode, unsigned int range)
658 {
659         int status;
660
661         spin_lock_bh(&queue->lock);
662         status = __nfqnl_set_mode(queue, mode, range);
663         spin_unlock_bh(&queue->lock);
664
665         return status;
666 }
667
668 static int
669 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
670 {
671         struct nf_info *entinf = entry->info;
672
673         if (entinf->indev)
674                 if (entinf->indev->ifindex == ifindex)
675                         return 1;
676         if (entinf->outdev)
677                 if (entinf->outdev->ifindex == ifindex)
678                         return 1;
679 #ifdef CONFIG_BRIDGE_NETFILTER
680         if (entry->skb->nf_bridge) {
681                 if (entry->skb->nf_bridge->physindev &&
682                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
683                         return 1;
684                 if (entry->skb->nf_bridge->physoutdev &&
685                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
686                         return 1;
687         }
688 #endif
689         return 0;
690 }
691
692 /* drop all packets with either indev or outdev == ifindex from all queue
693  * instances */
694 static void
695 nfqnl_dev_drop(int ifindex)
696 {
697         int i;
698
699         QDEBUG("entering for ifindex %u\n", ifindex);
700
701         /* this only looks like we have to hold the readlock for a way too long
702          * time, issue_verdict(),  nf_reinject(), ... - but we always only
703          * issue NF_DROP, which is processed directly in nf_reinject() */
704         read_lock_bh(&instances_lock);
705
706         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
707                 struct hlist_node *tmp;
708                 struct nfqnl_instance *inst;
709                 struct hlist_head *head = &instance_table[i];
710
711                 hlist_for_each_entry(inst, tmp, head, hlist) {
712                         struct nfqnl_queue_entry *entry;
713                         while ((entry = find_dequeue_entry(inst, dev_cmp,
714                                                            ifindex)) != NULL)
715                                 issue_verdict(entry, NF_DROP);
716                 }
717         }
718
719         read_unlock_bh(&instances_lock);
720 }
721
722 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
723
724 static int
725 nfqnl_rcv_dev_event(struct notifier_block *this,
726                     unsigned long event, void *ptr)
727 {
728         struct net_device *dev = ptr;
729
730         if (dev->nd_net != &init_net)
731                 return NOTIFY_DONE;
732
733         /* Drop any packets associated with the downed device */
734         if (event == NETDEV_DOWN)
735                 nfqnl_dev_drop(dev->ifindex);
736         return NOTIFY_DONE;
737 }
738
739 static struct notifier_block nfqnl_dev_notifier = {
740         .notifier_call  = nfqnl_rcv_dev_event,
741 };
742
743 static int
744 nfqnl_rcv_nl_event(struct notifier_block *this,
745                    unsigned long event, void *ptr)
746 {
747         struct netlink_notify *n = ptr;
748
749         if (event == NETLINK_URELEASE &&
750             n->protocol == NETLINK_NETFILTER && n->pid) {
751                 int i;
752
753                 /* destroy all instances for this pid */
754                 write_lock_bh(&instances_lock);
755                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
756                         struct hlist_node *tmp, *t2;
757                         struct nfqnl_instance *inst;
758                         struct hlist_head *head = &instance_table[i];
759
760                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
761                                 if ((n->net == &init_net) &&
762                                     (n->pid == inst->peer_pid))
763                                         __instance_destroy(inst);
764                         }
765                 }
766                 write_unlock_bh(&instances_lock);
767         }
768         return NOTIFY_DONE;
769 }
770
771 static struct notifier_block nfqnl_rtnl_notifier = {
772         .notifier_call  = nfqnl_rcv_nl_event,
773 };
774
775 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
776         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
777         [NFQA_MARK]             = { .type = NLA_U32 },
778         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
779 };
780
781 static int
782 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
783                    struct nlmsghdr *nlh, struct nlattr *nfqa[])
784 {
785         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
786         u_int16_t queue_num = ntohs(nfmsg->res_id);
787
788         struct nfqnl_msg_verdict_hdr *vhdr;
789         struct nfqnl_instance *queue;
790         unsigned int verdict;
791         struct nfqnl_queue_entry *entry;
792         int err;
793
794         queue = instance_lookup_get(queue_num);
795         if (!queue)
796                 return -ENODEV;
797
798         if (queue->peer_pid != NETLINK_CB(skb).pid) {
799                 err = -EPERM;
800                 goto err_out_put;
801         }
802
803         if (!nfqa[NFQA_VERDICT_HDR]) {
804                 err = -EINVAL;
805                 goto err_out_put;
806         }
807
808         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
809         verdict = ntohl(vhdr->verdict);
810
811         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
812                 err = -EINVAL;
813                 goto err_out_put;
814         }
815
816         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
817         if (entry == NULL) {
818                 err = -ENOENT;
819                 goto err_out_put;
820         }
821
822         if (nfqa[NFQA_PAYLOAD]) {
823                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
824                                  nla_len(nfqa[NFQA_PAYLOAD]), entry) < 0)
825                         verdict = NF_DROP;
826         }
827
828         if (nfqa[NFQA_MARK])
829                 entry->skb->mark = ntohl(*(__be32 *)
830                                          nla_data(nfqa[NFQA_MARK]));
831
832         issue_verdict(entry, verdict);
833         instance_put(queue);
834         return 0;
835
836 err_out_put:
837         instance_put(queue);
838         return err;
839 }
840
841 static int
842 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
843                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
844 {
845         return -ENOTSUPP;
846 }
847
848 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
849         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
850         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
851 };
852
853 static const struct nf_queue_handler nfqh = {
854         .name   = "nf_queue",
855         .outfn  = &nfqnl_enqueue_packet,
856 };
857
858 static int
859 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
860                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
861 {
862         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
863         u_int16_t queue_num = ntohs(nfmsg->res_id);
864         struct nfqnl_instance *queue;
865         int ret = 0;
866
867         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
868
869         queue = instance_lookup_get(queue_num);
870         if (nfqa[NFQA_CFG_CMD]) {
871                 struct nfqnl_msg_config_cmd *cmd;
872                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
873                 QDEBUG("found CFG_CMD\n");
874
875                 switch (cmd->command) {
876                 case NFQNL_CFG_CMD_BIND:
877                         if (queue)
878                                 return -EBUSY;
879
880                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
881                         if (!queue)
882                                 return -EINVAL;
883                         break;
884                 case NFQNL_CFG_CMD_UNBIND:
885                         if (!queue)
886                                 return -ENODEV;
887
888                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
889                                 ret = -EPERM;
890                                 goto out_put;
891                         }
892
893                         instance_destroy(queue);
894                         break;
895                 case NFQNL_CFG_CMD_PF_BIND:
896                         QDEBUG("registering queue handler for pf=%u\n",
897                                 ntohs(cmd->pf));
898                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
899                         break;
900                 case NFQNL_CFG_CMD_PF_UNBIND:
901                         QDEBUG("unregistering queue handler for pf=%u\n",
902                                 ntohs(cmd->pf));
903                         ret = nf_unregister_queue_handler(ntohs(cmd->pf), &nfqh);
904                         break;
905                 default:
906                         ret = -EINVAL;
907                         break;
908                 }
909         } else {
910                 if (!queue) {
911                         QDEBUG("no config command, and no instance ENOENT\n");
912                         ret = -ENOENT;
913                         goto out_put;
914                 }
915
916                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
917                         QDEBUG("no config command, and wrong pid\n");
918                         ret = -EPERM;
919                         goto out_put;
920                 }
921         }
922
923         if (nfqa[NFQA_CFG_PARAMS]) {
924                 struct nfqnl_msg_config_params *params;
925
926                 if (!queue) {
927                         ret = -ENOENT;
928                         goto out_put;
929                 }
930                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
931                 nfqnl_set_mode(queue, params->copy_mode,
932                                 ntohl(params->copy_range));
933         }
934
935         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
936                 __be32 *queue_maxlen;
937                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
938                 spin_lock_bh(&queue->lock);
939                 queue->queue_maxlen = ntohl(*queue_maxlen);
940                 spin_unlock_bh(&queue->lock);
941         }
942
943 out_put:
944         instance_put(queue);
945         return ret;
946 }
947
948 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
949         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
950                                     .attr_count = NFQA_MAX, },
951         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
952                                     .attr_count = NFQA_MAX,
953                                     .policy = nfqa_verdict_policy },
954         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
955                                     .attr_count = NFQA_CFG_MAX,
956                                     .policy = nfqa_cfg_policy },
957 };
958
959 static const struct nfnetlink_subsystem nfqnl_subsys = {
960         .name           = "nf_queue",
961         .subsys_id      = NFNL_SUBSYS_QUEUE,
962         .cb_count       = NFQNL_MSG_MAX,
963         .cb             = nfqnl_cb,
964 };
965
966 #ifdef CONFIG_PROC_FS
967 struct iter_state {
968         unsigned int bucket;
969 };
970
971 static struct hlist_node *get_first(struct seq_file *seq)
972 {
973         struct iter_state *st = seq->private;
974
975         if (!st)
976                 return NULL;
977
978         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
979                 if (!hlist_empty(&instance_table[st->bucket]))
980                         return instance_table[st->bucket].first;
981         }
982         return NULL;
983 }
984
985 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
986 {
987         struct iter_state *st = seq->private;
988
989         h = h->next;
990         while (!h) {
991                 if (++st->bucket >= INSTANCE_BUCKETS)
992                         return NULL;
993
994                 h = instance_table[st->bucket].first;
995         }
996         return h;
997 }
998
999 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1000 {
1001         struct hlist_node *head;
1002         head = get_first(seq);
1003
1004         if (head)
1005                 while (pos && (head = get_next(seq, head)))
1006                         pos--;
1007         return pos ? NULL : head;
1008 }
1009
1010 static void *seq_start(struct seq_file *seq, loff_t *pos)
1011 {
1012         read_lock_bh(&instances_lock);
1013         return get_idx(seq, *pos);
1014 }
1015
1016 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1017 {
1018         (*pos)++;
1019         return get_next(s, v);
1020 }
1021
1022 static void seq_stop(struct seq_file *s, void *v)
1023 {
1024         read_unlock_bh(&instances_lock);
1025 }
1026
1027 static int seq_show(struct seq_file *s, void *v)
1028 {
1029         const struct nfqnl_instance *inst = v;
1030
1031         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1032                           inst->queue_num,
1033                           inst->peer_pid, inst->queue_total,
1034                           inst->copy_mode, inst->copy_range,
1035                           inst->queue_dropped, inst->queue_user_dropped,
1036                           atomic_read(&inst->id_sequence),
1037                           atomic_read(&inst->use));
1038 }
1039
1040 static const struct seq_operations nfqnl_seq_ops = {
1041         .start  = seq_start,
1042         .next   = seq_next,
1043         .stop   = seq_stop,
1044         .show   = seq_show,
1045 };
1046
1047 static int nfqnl_open(struct inode *inode, struct file *file)
1048 {
1049         return seq_open_private(file, &nfqnl_seq_ops,
1050                         sizeof(struct iter_state));
1051 }
1052
1053 static const struct file_operations nfqnl_file_ops = {
1054         .owner   = THIS_MODULE,
1055         .open    = nfqnl_open,
1056         .read    = seq_read,
1057         .llseek  = seq_lseek,
1058         .release = seq_release_private,
1059 };
1060
1061 #endif /* PROC_FS */
1062
1063 static int __init nfnetlink_queue_init(void)
1064 {
1065         int i, status = -ENOMEM;
1066 #ifdef CONFIG_PROC_FS
1067         struct proc_dir_entry *proc_nfqueue;
1068 #endif
1069
1070         for (i = 0; i < INSTANCE_BUCKETS; i++)
1071                 INIT_HLIST_HEAD(&instance_table[i]);
1072
1073         netlink_register_notifier(&nfqnl_rtnl_notifier);
1074         status = nfnetlink_subsys_register(&nfqnl_subsys);
1075         if (status < 0) {
1076                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1077                 goto cleanup_netlink_notifier;
1078         }
1079
1080 #ifdef CONFIG_PROC_FS
1081         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1082                                          proc_net_netfilter);
1083         if (!proc_nfqueue)
1084                 goto cleanup_subsys;
1085         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1086 #endif
1087
1088         register_netdevice_notifier(&nfqnl_dev_notifier);
1089         return status;
1090
1091 #ifdef CONFIG_PROC_FS
1092 cleanup_subsys:
1093         nfnetlink_subsys_unregister(&nfqnl_subsys);
1094 #endif
1095 cleanup_netlink_notifier:
1096         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1097         return status;
1098 }
1099
1100 static void __exit nfnetlink_queue_fini(void)
1101 {
1102         nf_unregister_queue_handlers(&nfqh);
1103         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1104 #ifdef CONFIG_PROC_FS
1105         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1106 #endif
1107         nfnetlink_subsys_unregister(&nfqnl_subsys);
1108         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1109 }
1110
1111 MODULE_DESCRIPTION("netfilter packet queue handler");
1112 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1113 MODULE_LICENSE("GPL");
1114 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1115
1116 module_init(nfnetlink_queue_init);
1117 module_exit(nfnetlink_queue_fini);