Pull bsp-removal into release branch
[linux-2.6] / 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
31 #include <asm/atomic.h>
32
33 #ifdef CONFIG_BRIDGE_NETFILTER
34 #include "../bridge/br_private.h"
35 #endif
36
37 #define NFQNL_QMAX_DEFAULT 1024
38
39 #if 0
40 #define QDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
41                                         __FILE__, __LINE__, __FUNCTION__,  \
42                                         ## args)
43 #else
44 #define QDEBUG(x, ...)
45 #endif
46
47 struct nfqnl_queue_entry {
48         struct list_head list;
49         struct nf_info *info;
50         struct sk_buff *skb;
51         unsigned int id;
52 };
53
54 struct nfqnl_instance {
55         struct hlist_node hlist;                /* global list of queues */
56         atomic_t use;
57
58         int peer_pid;
59         unsigned int queue_maxlen;
60         unsigned int copy_range;
61         unsigned int queue_total;
62         unsigned int queue_dropped;
63         unsigned int queue_user_dropped;
64
65         atomic_t id_sequence;                   /* 'sequence' of pkt ids */
66
67         u_int16_t queue_num;                    /* number of this queue */
68         u_int8_t copy_mode;
69
70         spinlock_t lock;
71
72         struct list_head queue_list;            /* packets in queue */
73 };
74
75 typedef int (*nfqnl_cmpfn)(struct nfqnl_queue_entry *, unsigned long);
76
77 static DEFINE_RWLOCK(instances_lock);
78
79 #define INSTANCE_BUCKETS        16
80 static struct hlist_head instance_table[INSTANCE_BUCKETS];
81
82 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
83 {
84         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
85 }
86
87 static struct nfqnl_instance *
88 __instance_lookup(u_int16_t queue_num)
89 {
90         struct hlist_head *head;
91         struct hlist_node *pos;
92         struct nfqnl_instance *inst;
93
94         head = &instance_table[instance_hashfn(queue_num)];
95         hlist_for_each_entry(inst, pos, head, hlist) {
96                 if (inst->queue_num == queue_num)
97                         return inst;
98         }
99         return NULL;
100 }
101
102 static struct nfqnl_instance *
103 instance_lookup_get(u_int16_t queue_num)
104 {
105         struct nfqnl_instance *inst;
106
107         read_lock_bh(&instances_lock);
108         inst = __instance_lookup(queue_num);
109         if (inst)
110                 atomic_inc(&inst->use);
111         read_unlock_bh(&instances_lock);
112
113         return inst;
114 }
115
116 static void
117 instance_put(struct nfqnl_instance *inst)
118 {
119         if (inst && atomic_dec_and_test(&inst->use)) {
120                 QDEBUG("kfree(inst=%p)\n", inst);
121                 kfree(inst);
122         }
123 }
124
125 static struct nfqnl_instance *
126 instance_create(u_int16_t queue_num, int pid)
127 {
128         struct nfqnl_instance *inst;
129
130         QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
131
132         write_lock_bh(&instances_lock); 
133         if (__instance_lookup(queue_num)) {
134                 inst = NULL;
135                 QDEBUG("aborting, instance already exists\n");
136                 goto out_unlock;
137         }
138
139         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
140         if (!inst)
141                 goto out_unlock;
142
143         inst->queue_num = queue_num;
144         inst->peer_pid = pid;
145         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
146         inst->copy_range = 0xfffff;
147         inst->copy_mode = NFQNL_COPY_NONE;
148         atomic_set(&inst->id_sequence, 0);
149         /* needs to be two, since we _put() after creation */
150         atomic_set(&inst->use, 2);
151         spin_lock_init(&inst->lock);
152         INIT_LIST_HEAD(&inst->queue_list);
153
154         if (!try_module_get(THIS_MODULE))
155                 goto out_free;
156
157         hlist_add_head(&inst->hlist, 
158                        &instance_table[instance_hashfn(queue_num)]);
159
160         write_unlock_bh(&instances_lock);
161
162         QDEBUG("successfully created new instance\n");
163
164         return inst;
165
166 out_free:
167         kfree(inst);
168 out_unlock:
169         write_unlock_bh(&instances_lock);
170         return NULL;
171 }
172
173 static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
174
175 static void
176 _instance_destroy2(struct nfqnl_instance *inst, int lock)
177 {
178         /* first pull it out of the global list */
179         if (lock)
180                 write_lock_bh(&instances_lock);
181
182         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
183                 inst, inst->queue_num);
184         hlist_del(&inst->hlist);
185
186         if (lock)
187                 write_unlock_bh(&instances_lock);
188
189         /* then flush all pending skbs from the queue */
190         nfqnl_flush(inst, NF_DROP);
191
192         /* and finally put the refcount */
193         instance_put(inst);
194
195         module_put(THIS_MODULE);
196 }
197
198 static inline void
199 __instance_destroy(struct nfqnl_instance *inst)
200 {
201         _instance_destroy2(inst, 0);
202 }
203
204 static inline void
205 instance_destroy(struct nfqnl_instance *inst)
206 {
207         _instance_destroy2(inst, 1);
208 }
209
210
211
212 static void
213 issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
214 {
215         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
216
217         /* TCP input path (and probably other bits) assume to be called
218          * from softirq context, not from syscall, like issue_verdict is
219          * called.  TCP input path deadlocks with locks taken from timer
220          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
221
222         local_bh_disable();
223         nf_reinject(entry->skb, entry->info, verdict);
224         local_bh_enable();
225
226         kfree(entry);
227 }
228
229 static inline void
230 __enqueue_entry(struct nfqnl_instance *queue,
231                       struct nfqnl_queue_entry *entry)
232 {
233        list_add(&entry->list, &queue->queue_list);
234        queue->queue_total++;
235 }
236
237 /*
238  * Find and return a queued entry matched by cmpfn, or return the last
239  * entry if cmpfn is NULL.
240  */
241 static inline struct nfqnl_queue_entry *
242 __find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, 
243                    unsigned long data)
244 {
245         struct list_head *p;
246
247         list_for_each_prev(p, &queue->queue_list) {
248                 struct nfqnl_queue_entry *entry = (struct nfqnl_queue_entry *)p;
249                 
250                 if (!cmpfn || cmpfn(entry, data))
251                         return entry;
252         }
253         return NULL;
254 }
255
256 static inline void
257 __dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
258 {
259         list_del(&entry->list);
260         q->queue_total--;
261 }
262
263 static inline struct nfqnl_queue_entry *
264 __find_dequeue_entry(struct nfqnl_instance *queue,
265                      nfqnl_cmpfn cmpfn, unsigned long data)
266 {
267         struct nfqnl_queue_entry *entry;
268
269         entry = __find_entry(queue, cmpfn, data);
270         if (entry == NULL)
271                 return NULL;
272
273         __dequeue_entry(queue, entry);
274         return entry;
275 }
276
277
278 static inline void
279 __nfqnl_flush(struct nfqnl_instance *queue, int verdict)
280 {
281         struct nfqnl_queue_entry *entry;
282         
283         while ((entry = __find_dequeue_entry(queue, NULL, 0)))
284                 issue_verdict(entry, verdict);
285 }
286
287 static inline int
288 __nfqnl_set_mode(struct nfqnl_instance *queue,
289                  unsigned char mode, unsigned int range)
290 {
291         int status = 0;
292         
293         switch (mode) {
294         case NFQNL_COPY_NONE:
295         case NFQNL_COPY_META:
296                 queue->copy_mode = mode;
297                 queue->copy_range = 0;
298                 break;
299                 
300         case NFQNL_COPY_PACKET:
301                 queue->copy_mode = mode;
302                 /* we're using struct nfattr which has 16bit nfa_len */
303                 if (range > 0xffff)
304                         queue->copy_range = 0xffff;
305                 else
306                         queue->copy_range = range;
307                 break;
308                 
309         default:
310                 status = -EINVAL;
311
312         }
313         return status;
314 }
315
316 static struct nfqnl_queue_entry *
317 find_dequeue_entry(struct nfqnl_instance *queue,
318                          nfqnl_cmpfn cmpfn, unsigned long data)
319 {
320         struct nfqnl_queue_entry *entry;
321         
322         spin_lock_bh(&queue->lock);
323         entry = __find_dequeue_entry(queue, cmpfn, data);
324         spin_unlock_bh(&queue->lock);
325
326         return entry;
327 }
328
329 static void
330 nfqnl_flush(struct nfqnl_instance *queue, int verdict)
331 {
332         spin_lock_bh(&queue->lock);
333         __nfqnl_flush(queue, verdict);
334         spin_unlock_bh(&queue->lock);
335 }
336
337 static struct sk_buff *
338 nfqnl_build_packet_message(struct nfqnl_instance *queue,
339                            struct nfqnl_queue_entry *entry, int *errp)
340 {
341         unsigned char *old_tail;
342         size_t size;
343         size_t data_len = 0;
344         struct sk_buff *skb;
345         struct nfqnl_msg_packet_hdr pmsg;
346         struct nlmsghdr *nlh;
347         struct nfgenmsg *nfmsg;
348         struct nf_info *entinf = entry->info;
349         struct sk_buff *entskb = entry->skb;
350         struct net_device *indev;
351         struct net_device *outdev;
352         unsigned int tmp_uint;
353
354         QDEBUG("entered\n");
355
356         /* all macros expand to constant values at compile time */
357         size =    NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
358                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
359                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
360 #ifdef CONFIG_BRIDGE_NETFILTER
361                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
362                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
363 #endif
364                 + NLMSG_SPACE(sizeof(u_int32_t))        /* mark */
365                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hw))
366                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_timestamp));
367
368         outdev = entinf->outdev;
369
370         spin_lock_bh(&queue->lock);
371         
372         switch (queue->copy_mode) {
373         case NFQNL_COPY_META:
374         case NFQNL_COPY_NONE:
375                 data_len = 0;
376                 break;
377         
378         case NFQNL_COPY_PACKET:
379                 if (entskb->ip_summed == CHECKSUM_HW &&
380                     (*errp = skb_checksum_help(entskb,
381                                                outdev == NULL))) {
382                         spin_unlock_bh(&queue->lock);
383                         return NULL;
384                 }
385                 if (queue->copy_range == 0 
386                     || queue->copy_range > entskb->len)
387                         data_len = entskb->len;
388                 else
389                         data_len = queue->copy_range;
390                 
391                 size += NLMSG_SPACE(data_len);
392                 break;
393         
394         default:
395                 *errp = -EINVAL;
396                 spin_unlock_bh(&queue->lock);
397                 return NULL;
398         }
399
400         spin_unlock_bh(&queue->lock);
401
402         skb = alloc_skb(size, GFP_ATOMIC);
403         if (!skb)
404                 goto nlmsg_failure;
405                 
406         old_tail= skb->tail;
407         nlh = NLMSG_PUT(skb, 0, 0, 
408                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
409                         sizeof(struct nfgenmsg));
410         nfmsg = NLMSG_DATA(nlh);
411         nfmsg->nfgen_family = entinf->pf;
412         nfmsg->version = NFNETLINK_V0;
413         nfmsg->res_id = htons(queue->queue_num);
414
415         pmsg.packet_id          = htonl(entry->id);
416         pmsg.hw_protocol        = htons(entskb->protocol);
417         pmsg.hook               = entinf->hook;
418
419         NFA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
420
421         indev = entinf->indev;
422         if (indev) {
423                 tmp_uint = htonl(indev->ifindex);
424 #ifndef CONFIG_BRIDGE_NETFILTER
425                 NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
426 #else
427                 if (entinf->pf == PF_BRIDGE) {
428                         /* Case 1: indev is physical input device, we need to
429                          * look for bridge group (when called from 
430                          * netfilter_bridge) */
431                         NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint), 
432                                 &tmp_uint);
433                         /* this is the bridge group "brX" */
434                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
435                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
436                                 &tmp_uint);
437                 } else {
438                         /* Case 2: indev is bridge group, we need to look for
439                          * physical device (when called from ipv4) */
440                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
441                                 &tmp_uint);
442                         if (entskb->nf_bridge
443                             && entskb->nf_bridge->physindev) {
444                                 tmp_uint = htonl(entskb->nf_bridge->physindev->ifindex);
445                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
446                                         sizeof(tmp_uint), &tmp_uint);
447                         }
448                 }
449 #endif
450         }
451
452         if (outdev) {
453                 tmp_uint = htonl(outdev->ifindex);
454 #ifndef CONFIG_BRIDGE_NETFILTER
455                 NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
456 #else
457                 if (entinf->pf == PF_BRIDGE) {
458                         /* Case 1: outdev is physical output device, we need to
459                          * look for bridge group (when called from 
460                          * netfilter_bridge) */
461                         NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
462                                 &tmp_uint);
463                         /* this is the bridge group "brX" */
464                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
465                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
466                                 &tmp_uint);
467                 } else {
468                         /* Case 2: outdev is bridge group, we need to look for
469                          * physical output device (when called from ipv4) */
470                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
471                                 &tmp_uint);
472                         if (entskb->nf_bridge
473                             && entskb->nf_bridge->physoutdev) {
474                                 tmp_uint = htonl(entskb->nf_bridge->physoutdev->ifindex);
475                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
476                                         sizeof(tmp_uint), &tmp_uint);
477                         }
478                 }
479 #endif
480         }
481
482         if (entskb->nfmark) {
483                 tmp_uint = htonl(entskb->nfmark);
484                 NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
485         }
486
487         if (indev && entskb->dev
488             && entskb->dev->hard_header_parse) {
489                 struct nfqnl_msg_packet_hw phw;
490
491                 phw.hw_addrlen =
492                         entskb->dev->hard_header_parse(entskb,
493                                                            phw.hw_addr);
494                 phw.hw_addrlen = htons(phw.hw_addrlen);
495                 NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
496         }
497
498         if (entskb->tstamp.off_sec) {
499                 struct nfqnl_msg_packet_timestamp ts;
500
501                 ts.sec = cpu_to_be64(entskb->tstamp.off_sec);
502                 ts.usec = cpu_to_be64(entskb->tstamp.off_usec);
503
504                 NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
505         }
506
507         if (data_len) {
508                 struct nfattr *nfa;
509                 int size = NFA_LENGTH(data_len);
510
511                 if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
512                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
513                         goto nlmsg_failure;
514                 }
515
516                 nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
517                 nfa->nfa_type = NFQA_PAYLOAD;
518                 nfa->nfa_len = size;
519
520                 if (skb_copy_bits(entskb, 0, NFA_DATA(nfa), data_len))
521                         BUG();
522         }
523                 
524         nlh->nlmsg_len = skb->tail - old_tail;
525         return skb;
526
527 nlmsg_failure:
528 nfattr_failure:
529         if (skb)
530                 kfree_skb(skb);
531         *errp = -EINVAL;
532         if (net_ratelimit())
533                 printk(KERN_ERR "nf_queue: error creating packet message\n");
534         return NULL;
535 }
536
537 static int
538 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info, 
539                      unsigned int queuenum, void *data)
540 {
541         int status = -EINVAL;
542         struct sk_buff *nskb;
543         struct nfqnl_instance *queue;
544         struct nfqnl_queue_entry *entry;
545
546         QDEBUG("entered\n");
547
548         queue = instance_lookup_get(queuenum);
549         if (!queue) {
550                 QDEBUG("no queue instance matching\n");
551                 return -EINVAL;
552         }
553
554         if (queue->copy_mode == NFQNL_COPY_NONE) {
555                 QDEBUG("mode COPY_NONE, aborting\n");
556                 status = -EAGAIN;
557                 goto err_out_put;
558         }
559
560         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
561         if (entry == NULL) {
562                 if (net_ratelimit())
563                         printk(KERN_ERR 
564                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
565                 status = -ENOMEM;
566                 goto err_out_put;
567         }
568
569         entry->info = info;
570         entry->skb = skb;
571         entry->id = atomic_inc_return(&queue->id_sequence);
572
573         nskb = nfqnl_build_packet_message(queue, entry, &status);
574         if (nskb == NULL)
575                 goto err_out_free;
576                 
577         spin_lock_bh(&queue->lock);
578         
579         if (!queue->peer_pid)
580                 goto err_out_free_nskb; 
581
582         if (queue->queue_total >= queue->queue_maxlen) {
583                 queue->queue_dropped++;
584                 status = -ENOSPC;
585                 if (net_ratelimit())
586                           printk(KERN_WARNING "ip_queue: full at %d entries, "
587                                  "dropping packets(s). Dropped: %d\n", 
588                                  queue->queue_total, queue->queue_dropped);
589                 goto err_out_free_nskb;
590         }
591
592         /* nfnetlink_unicast will either free the nskb or add it to a socket */
593         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
594         if (status < 0) {
595                 queue->queue_user_dropped++;
596                 goto err_out_unlock;
597         }
598
599         __enqueue_entry(queue, entry);
600
601         spin_unlock_bh(&queue->lock);
602         instance_put(queue);
603         return status;
604
605 err_out_free_nskb:
606         kfree_skb(nskb); 
607         
608 err_out_unlock:
609         spin_unlock_bh(&queue->lock);
610
611 err_out_free:
612         kfree(entry);
613 err_out_put:
614         instance_put(queue);
615         return status;
616 }
617
618 static int
619 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
620 {
621         int diff;
622
623         diff = data_len - e->skb->len;
624         if (diff < 0)
625                 skb_trim(e->skb, data_len);
626         else if (diff > 0) {
627                 if (data_len > 0xFFFF)
628                         return -EINVAL;
629                 if (diff > skb_tailroom(e->skb)) {
630                         struct sk_buff *newskb;
631                         
632                         newskb = skb_copy_expand(e->skb,
633                                                  skb_headroom(e->skb),
634                                                  diff,
635                                                  GFP_ATOMIC);
636                         if (newskb == NULL) {
637                                 printk(KERN_WARNING "ip_queue: OOM "
638                                       "in mangle, dropping packet\n");
639                                 return -ENOMEM;
640                         }
641                         if (e->skb->sk)
642                                 skb_set_owner_w(newskb, e->skb->sk);
643                         kfree_skb(e->skb);
644                         e->skb = newskb;
645                 }
646                 skb_put(e->skb, diff);
647         }
648         if (!skb_make_writable(&e->skb, data_len))
649                 return -ENOMEM;
650         memcpy(e->skb->data, data, data_len);
651         e->skb->ip_summed = CHECKSUM_NONE;
652         return 0;
653 }
654
655 static inline int
656 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
657 {
658         return (id == e->id);
659 }
660
661 static int
662 nfqnl_set_mode(struct nfqnl_instance *queue,
663                unsigned char mode, unsigned int range)
664 {
665         int status;
666
667         spin_lock_bh(&queue->lock);
668         status = __nfqnl_set_mode(queue, mode, range);
669         spin_unlock_bh(&queue->lock);
670
671         return status;
672 }
673
674 static int
675 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
676 {
677         struct nf_info *entinf = entry->info;
678         
679         if (entinf->indev)
680                 if (entinf->indev->ifindex == ifindex)
681                         return 1;
682                         
683         if (entinf->outdev)
684                 if (entinf->outdev->ifindex == ifindex)
685                         return 1;
686
687         return 0;
688 }
689
690 /* drop all packets with either indev or outdev == ifindex from all queue
691  * instances */
692 static void
693 nfqnl_dev_drop(int ifindex)
694 {
695         int i;
696         
697         QDEBUG("entering for ifindex %u\n", ifindex);
698
699         /* this only looks like we have to hold the readlock for a way too long
700          * time, issue_verdict(),  nf_reinject(), ... - but we always only
701          * issue NF_DROP, which is processed directly in nf_reinject() */
702         read_lock_bh(&instances_lock);
703
704         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
705                 struct hlist_node *tmp;
706                 struct nfqnl_instance *inst;
707                 struct hlist_head *head = &instance_table[i];
708
709                 hlist_for_each_entry(inst, tmp, head, hlist) {
710                         struct nfqnl_queue_entry *entry;
711                         while ((entry = find_dequeue_entry(inst, dev_cmp, 
712                                                            ifindex)) != NULL)
713                                 issue_verdict(entry, NF_DROP);
714                 }
715         }
716
717         read_unlock_bh(&instances_lock);
718 }
719
720 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
721
722 static int
723 nfqnl_rcv_dev_event(struct notifier_block *this,
724                     unsigned long event, void *ptr)
725 {
726         struct net_device *dev = ptr;
727
728         /* Drop any packets associated with the downed device */
729         if (event == NETDEV_DOWN)
730                 nfqnl_dev_drop(dev->ifindex);
731         return NOTIFY_DONE;
732 }
733
734 static struct notifier_block nfqnl_dev_notifier = {
735         .notifier_call  = nfqnl_rcv_dev_event,
736 };
737
738 static int
739 nfqnl_rcv_nl_event(struct notifier_block *this,
740                    unsigned long event, void *ptr)
741 {
742         struct netlink_notify *n = ptr;
743
744         if (event == NETLINK_URELEASE &&
745             n->protocol == NETLINK_NETFILTER && n->pid) {
746                 int i;
747
748                 /* destroy all instances for this pid */
749                 write_lock_bh(&instances_lock);
750                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
751                         struct hlist_node *tmp, *t2;
752                         struct nfqnl_instance *inst;
753                         struct hlist_head *head = &instance_table[i];
754
755                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
756                                 if (n->pid == inst->peer_pid)
757                                         __instance_destroy(inst);
758                         }
759                 }
760                 write_unlock_bh(&instances_lock);
761         }
762         return NOTIFY_DONE;
763 }
764
765 static struct notifier_block nfqnl_rtnl_notifier = {
766         .notifier_call  = nfqnl_rcv_nl_event,
767 };
768
769 static const int nfqa_verdict_min[NFQA_MAX] = {
770         [NFQA_VERDICT_HDR-1]    = sizeof(struct nfqnl_msg_verdict_hdr),
771         [NFQA_MARK-1]           = sizeof(u_int32_t),
772         [NFQA_PAYLOAD-1]        = 0,
773 };
774
775 static int
776 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
777                    struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
778 {
779         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
780         u_int16_t queue_num = ntohs(nfmsg->res_id);
781
782         struct nfqnl_msg_verdict_hdr *vhdr;
783         struct nfqnl_instance *queue;
784         unsigned int verdict;
785         struct nfqnl_queue_entry *entry;
786         int err;
787
788         if (nfattr_bad_size(nfqa, NFQA_MAX, nfqa_verdict_min)) {
789                 QDEBUG("bad attribute size\n");
790                 return -EINVAL;
791         }
792
793         queue = instance_lookup_get(queue_num);
794         if (!queue)
795                 return -ENODEV;
796
797         if (queue->peer_pid != NETLINK_CB(skb).pid) {
798                 err = -EPERM;
799                 goto err_out_put;
800         }
801
802         if (!nfqa[NFQA_VERDICT_HDR-1]) {
803                 err = -EINVAL;
804                 goto err_out_put;
805         }
806
807         vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
808         verdict = ntohl(vhdr->verdict);
809
810         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
811                 err = -EINVAL;
812                 goto err_out_put;
813         }
814
815         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
816         if (entry == NULL) {
817                 err = -ENOENT;
818                 goto err_out_put;
819         }
820
821         if (nfqa[NFQA_PAYLOAD-1]) {
822                 if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
823                                  NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
824                         verdict = NF_DROP;
825         }
826
827         if (nfqa[NFQA_MARK-1])
828                 entry->skb->nfmark = ntohl(*(u_int32_t *)
829                                            NFA_DATA(nfqa[NFQA_MARK-1]));
830                 
831         issue_verdict(entry, verdict);
832         instance_put(queue);
833         return 0;
834
835 err_out_put:
836         instance_put(queue);
837         return err;
838 }
839
840 static int
841 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
842                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
843 {
844         return -ENOTSUPP;
845 }
846
847 static const int nfqa_cfg_min[NFQA_CFG_MAX] = {
848         [NFQA_CFG_CMD-1]        = sizeof(struct nfqnl_msg_config_cmd),
849         [NFQA_CFG_PARAMS-1]     = sizeof(struct nfqnl_msg_config_params),
850 };
851
852 static struct nf_queue_handler nfqh = {
853         .name   = "nf_queue",
854         .outfn  = &nfqnl_enqueue_packet,
855 };
856
857 static int
858 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
859                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
860 {
861         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
862         u_int16_t queue_num = ntohs(nfmsg->res_id);
863         struct nfqnl_instance *queue;
864         int ret = 0;
865
866         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
867
868         if (nfattr_bad_size(nfqa, NFQA_CFG_MAX, nfqa_cfg_min)) {
869                 QDEBUG("bad attribute size\n");
870                 return -EINVAL;
871         }
872
873         queue = instance_lookup_get(queue_num);
874         if (nfqa[NFQA_CFG_CMD-1]) {
875                 struct nfqnl_msg_config_cmd *cmd;
876                 cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
877                 QDEBUG("found CFG_CMD\n");
878
879                 switch (cmd->command) {
880                 case NFQNL_CFG_CMD_BIND:
881                         if (queue)
882                                 return -EBUSY;
883
884                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
885                         if (!queue)
886                                 return -EINVAL;
887                         break;
888                 case NFQNL_CFG_CMD_UNBIND:
889                         if (!queue)
890                                 return -ENODEV;
891
892                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
893                                 ret = -EPERM;
894                                 goto out_put;
895                         }
896
897                         instance_destroy(queue);
898                         break;
899                 case NFQNL_CFG_CMD_PF_BIND:
900                         QDEBUG("registering queue handler for pf=%u\n",
901                                 ntohs(cmd->pf));
902                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
903                         break;
904                 case NFQNL_CFG_CMD_PF_UNBIND:
905                         QDEBUG("unregistering queue handler for pf=%u\n",
906                                 ntohs(cmd->pf));
907                         /* This is a bug and a feature.  We can unregister
908                          * other handlers(!) */
909                         ret = nf_unregister_queue_handler(ntohs(cmd->pf));
910                         break;
911                 default:
912                         ret = -EINVAL;
913                         break;
914                 }
915         } else {
916                 if (!queue) {
917                         QDEBUG("no config command, and no instance ENOENT\n");
918                         ret = -ENOENT;
919                         goto out_put;
920                 }
921
922                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
923                         QDEBUG("no config command, and wrong pid\n");
924                         ret = -EPERM;
925                         goto out_put;
926                 }
927         }
928
929         if (nfqa[NFQA_CFG_PARAMS-1]) {
930                 struct nfqnl_msg_config_params *params;
931
932                 if (!queue) {
933                         ret = -ENOENT;
934                         goto out_put;
935                 }
936                 params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
937                 nfqnl_set_mode(queue, params->copy_mode,
938                                 ntohl(params->copy_range));
939         }
940
941 out_put:
942         instance_put(queue);
943         return ret;
944 }
945
946 static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
947         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
948                                     .attr_count = NFQA_MAX, },
949         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
950                                     .attr_count = NFQA_MAX, },
951         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
952                                     .attr_count = NFQA_CFG_MAX, },
953 };
954
955 static struct nfnetlink_subsystem nfqnl_subsys = {
956         .name           = "nf_queue",
957         .subsys_id      = NFNL_SUBSYS_QUEUE,
958         .cb_count       = NFQNL_MSG_MAX,
959         .cb             = nfqnl_cb,
960 };
961
962 #ifdef CONFIG_PROC_FS
963 struct iter_state {
964         unsigned int bucket;
965 };
966
967 static struct hlist_node *get_first(struct seq_file *seq)
968 {
969         struct iter_state *st = seq->private;
970
971         if (!st)
972                 return NULL;
973
974         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
975                 if (!hlist_empty(&instance_table[st->bucket]))
976                         return instance_table[st->bucket].first;
977         }
978         return NULL;
979 }
980
981 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
982 {
983         struct iter_state *st = seq->private;
984
985         h = h->next;
986         while (!h) {
987                 if (++st->bucket >= INSTANCE_BUCKETS)
988                         return NULL;
989
990                 h = instance_table[st->bucket].first;
991         }
992         return h;
993 }
994
995 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
996 {
997         struct hlist_node *head;
998         head = get_first(seq);
999
1000         if (head)
1001                 while (pos && (head = get_next(seq, head)))
1002                         pos--;
1003         return pos ? NULL : head;
1004 }
1005
1006 static void *seq_start(struct seq_file *seq, loff_t *pos)
1007 {
1008         read_lock_bh(&instances_lock);
1009         return get_idx(seq, *pos);
1010 }
1011
1012 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1013 {
1014         (*pos)++;
1015         return get_next(s, v);
1016 }
1017
1018 static void seq_stop(struct seq_file *s, void *v)
1019 {
1020         read_unlock_bh(&instances_lock);
1021 }
1022
1023 static int seq_show(struct seq_file *s, void *v)
1024 {
1025         const struct nfqnl_instance *inst = v;
1026
1027         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1028                           inst->queue_num,
1029                           inst->peer_pid, inst->queue_total,
1030                           inst->copy_mode, inst->copy_range,
1031                           inst->queue_dropped, inst->queue_user_dropped,
1032                           atomic_read(&inst->id_sequence),
1033                           atomic_read(&inst->use));
1034 }
1035
1036 static struct seq_operations nfqnl_seq_ops = {
1037         .start  = seq_start,
1038         .next   = seq_next,
1039         .stop   = seq_stop,
1040         .show   = seq_show,
1041 };
1042
1043 static int nfqnl_open(struct inode *inode, struct file *file)
1044 {
1045         struct seq_file *seq;
1046         struct iter_state *is;
1047         int ret;
1048
1049         is = kzalloc(sizeof(*is), GFP_KERNEL);
1050         if (!is)
1051                 return -ENOMEM;
1052         ret = seq_open(file, &nfqnl_seq_ops);
1053         if (ret < 0)
1054                 goto out_free;
1055         seq = file->private_data;
1056         seq->private = is;
1057         return ret;
1058 out_free:
1059         kfree(is);
1060         return ret;
1061 }
1062
1063 static struct file_operations nfqnl_file_ops = {
1064         .owner   = THIS_MODULE,
1065         .open    = nfqnl_open,
1066         .read    = seq_read,
1067         .llseek  = seq_lseek,
1068         .release = seq_release_private,
1069 };
1070
1071 #endif /* PROC_FS */
1072
1073 static int
1074 init_or_cleanup(int init)
1075 {
1076         int i, status = -ENOMEM;
1077 #ifdef CONFIG_PROC_FS
1078         struct proc_dir_entry *proc_nfqueue;
1079 #endif
1080         
1081         if (!init)
1082                 goto cleanup;
1083
1084         for (i = 0; i < INSTANCE_BUCKETS; i++)
1085                 INIT_HLIST_HEAD(&instance_table[i]);
1086
1087         netlink_register_notifier(&nfqnl_rtnl_notifier);
1088         status = nfnetlink_subsys_register(&nfqnl_subsys);
1089         if (status < 0) {
1090                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1091                 goto cleanup_netlink_notifier;
1092         }
1093
1094 #ifdef CONFIG_PROC_FS
1095         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1096                                          proc_net_netfilter);
1097         if (!proc_nfqueue)
1098                 goto cleanup_subsys;
1099         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1100 #endif
1101
1102         register_netdevice_notifier(&nfqnl_dev_notifier);
1103
1104         return status;
1105
1106 cleanup:
1107         nf_unregister_queue_handlers(&nfqh);
1108         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1109 #ifdef CONFIG_PROC_FS
1110         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1111 cleanup_subsys:
1112 #endif  
1113         nfnetlink_subsys_unregister(&nfqnl_subsys);
1114 cleanup_netlink_notifier:
1115         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1116         return status;
1117 }
1118
1119 static int __init init(void)
1120 {
1121         
1122         return init_or_cleanup(1);
1123 }
1124
1125 static void __exit fini(void)
1126 {
1127         init_or_cleanup(0);
1128 }
1129
1130 MODULE_DESCRIPTION("netfilter packet queue handler");
1131 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1132 MODULE_LICENSE("GPL");
1133 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1134
1135 module_init(init);
1136 module_exit(fini);