Merge branch 'master'
[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         inst->lock = SPIN_LOCK_UNLOCKED;
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         unsigned int tmp_uint;
349
350         QDEBUG("entered\n");
351
352         /* all macros expand to constant values at compile time */
353         size =    NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
354                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
355                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
356 #ifdef CONFIG_BRIDGE_NETFILTER
357                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
358                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
359 #endif
360                 + NLMSG_SPACE(sizeof(u_int32_t))        /* mark */
361                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hw))
362                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_timestamp));
363
364         spin_lock_bh(&queue->lock);
365         
366         switch (queue->copy_mode) {
367         case NFQNL_COPY_META:
368         case NFQNL_COPY_NONE:
369                 data_len = 0;
370                 break;
371         
372         case NFQNL_COPY_PACKET:
373                 if (entry->skb->ip_summed == CHECKSUM_HW &&
374                     (*errp = skb_checksum_help(entry->skb,
375                                                entry->info->outdev == NULL))) {
376                         spin_unlock_bh(&queue->lock);
377                         return NULL;
378                 }
379                 if (queue->copy_range == 0 
380                     || queue->copy_range > entry->skb->len)
381                         data_len = entry->skb->len;
382                 else
383                         data_len = queue->copy_range;
384                 
385                 size += NLMSG_SPACE(data_len);
386                 break;
387         
388         default:
389                 *errp = -EINVAL;
390                 spin_unlock_bh(&queue->lock);
391                 return NULL;
392         }
393
394         spin_unlock_bh(&queue->lock);
395
396         skb = alloc_skb(size, GFP_ATOMIC);
397         if (!skb)
398                 goto nlmsg_failure;
399                 
400         old_tail= skb->tail;
401         nlh = NLMSG_PUT(skb, 0, 0, 
402                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
403                         sizeof(struct nfgenmsg));
404         nfmsg = NLMSG_DATA(nlh);
405         nfmsg->nfgen_family = entry->info->pf;
406         nfmsg->version = NFNETLINK_V0;
407         nfmsg->res_id = htons(queue->queue_num);
408
409         pmsg.packet_id          = htonl(entry->id);
410         pmsg.hw_protocol        = htons(entry->skb->protocol);
411         pmsg.hook               = entry->info->hook;
412
413         NFA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
414
415         if (entry->info->indev) {
416                 tmp_uint = htonl(entry->info->indev->ifindex);
417 #ifndef CONFIG_BRIDGE_NETFILTER
418                 NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
419 #else
420                 if (entry->info->pf == PF_BRIDGE) {
421                         /* Case 1: indev is physical input device, we need to
422                          * look for bridge group (when called from 
423                          * netfilter_bridge) */
424                         NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint), 
425                                 &tmp_uint);
426                         /* this is the bridge group "brX" */
427                         tmp_uint = htonl(entry->info->indev->br_port->br->dev->ifindex);
428                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
429                                 &tmp_uint);
430                 } else {
431                         /* Case 2: indev is bridge group, we need to look for
432                          * physical device (when called from ipv4) */
433                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
434                                 &tmp_uint);
435                         if (entry->skb->nf_bridge
436                             && entry->skb->nf_bridge->physindev) {
437                                 tmp_uint = htonl(entry->skb->nf_bridge->physindev->ifindex);
438                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
439                                         sizeof(tmp_uint), &tmp_uint);
440                         }
441                 }
442 #endif
443         }
444
445         if (entry->info->outdev) {
446                 tmp_uint = htonl(entry->info->outdev->ifindex);
447 #ifndef CONFIG_BRIDGE_NETFILTER
448                 NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
449 #else
450                 if (entry->info->pf == PF_BRIDGE) {
451                         /* Case 1: outdev is physical output device, we need to
452                          * look for bridge group (when called from 
453                          * netfilter_bridge) */
454                         NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
455                                 &tmp_uint);
456                         /* this is the bridge group "brX" */
457                         tmp_uint = htonl(entry->info->outdev->br_port->br->dev->ifindex);
458                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
459                                 &tmp_uint);
460                 } else {
461                         /* Case 2: outdev is bridge group, we need to look for
462                          * physical output device (when called from ipv4) */
463                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
464                                 &tmp_uint);
465                         if (entry->skb->nf_bridge
466                             && entry->skb->nf_bridge->physoutdev) {
467                                 tmp_uint = htonl(entry->skb->nf_bridge->physoutdev->ifindex);
468                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
469                                         sizeof(tmp_uint), &tmp_uint);
470                         }
471                 }
472 #endif
473         }
474
475         if (entry->skb->nfmark) {
476                 tmp_uint = htonl(entry->skb->nfmark);
477                 NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
478         }
479
480         if (entry->info->indev && entry->skb->dev
481             && entry->skb->dev->hard_header_parse) {
482                 struct nfqnl_msg_packet_hw phw;
483
484                 phw.hw_addrlen =
485                         entry->skb->dev->hard_header_parse(entry->skb,
486                                                            phw.hw_addr);
487                 phw.hw_addrlen = htons(phw.hw_addrlen);
488                 NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
489         }
490
491         if (entry->skb->tstamp.off_sec) {
492                 struct nfqnl_msg_packet_timestamp ts;
493
494                 ts.sec = cpu_to_be64(entry->skb->tstamp.off_sec);
495                 ts.usec = cpu_to_be64(entry->skb->tstamp.off_usec);
496
497                 NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
498         }
499
500         if (data_len) {
501                 struct nfattr *nfa;
502                 int size = NFA_LENGTH(data_len);
503
504                 if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
505                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
506                         goto nlmsg_failure;
507                 }
508
509                 nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
510                 nfa->nfa_type = NFQA_PAYLOAD;
511                 nfa->nfa_len = size;
512
513                 if (skb_copy_bits(entry->skb, 0, NFA_DATA(nfa), data_len))
514                         BUG();
515         }
516                 
517         nlh->nlmsg_len = skb->tail - old_tail;
518         return skb;
519
520 nlmsg_failure:
521 nfattr_failure:
522         if (skb)
523                 kfree_skb(skb);
524         *errp = -EINVAL;
525         if (net_ratelimit())
526                 printk(KERN_ERR "nf_queue: error creating packet message\n");
527         return NULL;
528 }
529
530 static int
531 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info, 
532                      unsigned int queuenum, void *data)
533 {
534         int status = -EINVAL;
535         struct sk_buff *nskb;
536         struct nfqnl_instance *queue;
537         struct nfqnl_queue_entry *entry;
538
539         QDEBUG("entered\n");
540
541         queue = instance_lookup_get(queuenum);
542         if (!queue) {
543                 QDEBUG("no queue instance matching\n");
544                 return -EINVAL;
545         }
546
547         if (queue->copy_mode == NFQNL_COPY_NONE) {
548                 QDEBUG("mode COPY_NONE, aborting\n");
549                 status = -EAGAIN;
550                 goto err_out_put;
551         }
552
553         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
554         if (entry == NULL) {
555                 if (net_ratelimit())
556                         printk(KERN_ERR 
557                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
558                 status = -ENOMEM;
559                 goto err_out_put;
560         }
561
562         entry->info = info;
563         entry->skb = skb;
564         entry->id = atomic_inc_return(&queue->id_sequence);
565
566         nskb = nfqnl_build_packet_message(queue, entry, &status);
567         if (nskb == NULL)
568                 goto err_out_free;
569                 
570         spin_lock_bh(&queue->lock);
571         
572         if (!queue->peer_pid)
573                 goto err_out_free_nskb; 
574
575         if (queue->queue_total >= queue->queue_maxlen) {
576                 queue->queue_dropped++;
577                 status = -ENOSPC;
578                 if (net_ratelimit())
579                           printk(KERN_WARNING "ip_queue: full at %d entries, "
580                                  "dropping packets(s). Dropped: %d\n", 
581                                  queue->queue_total, queue->queue_dropped);
582                 goto err_out_free_nskb;
583         }
584
585         /* nfnetlink_unicast will either free the nskb or add it to a socket */
586         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
587         if (status < 0) {
588                 queue->queue_user_dropped++;
589                 goto err_out_unlock;
590         }
591
592         __enqueue_entry(queue, entry);
593
594         spin_unlock_bh(&queue->lock);
595         instance_put(queue);
596         return status;
597
598 err_out_free_nskb:
599         kfree_skb(nskb); 
600         
601 err_out_unlock:
602         spin_unlock_bh(&queue->lock);
603
604 err_out_free:
605         kfree(entry);
606 err_out_put:
607         instance_put(queue);
608         return status;
609 }
610
611 static int
612 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
613 {
614         int diff;
615
616         diff = data_len - e->skb->len;
617         if (diff < 0)
618                 skb_trim(e->skb, data_len);
619         else if (diff > 0) {
620                 if (data_len > 0xFFFF)
621                         return -EINVAL;
622                 if (diff > skb_tailroom(e->skb)) {
623                         struct sk_buff *newskb;
624                         
625                         newskb = skb_copy_expand(e->skb,
626                                                  skb_headroom(e->skb),
627                                                  diff,
628                                                  GFP_ATOMIC);
629                         if (newskb == NULL) {
630                                 printk(KERN_WARNING "ip_queue: OOM "
631                                       "in mangle, dropping packet\n");
632                                 return -ENOMEM;
633                         }
634                         if (e->skb->sk)
635                                 skb_set_owner_w(newskb, e->skb->sk);
636                         kfree_skb(e->skb);
637                         e->skb = newskb;
638                 }
639                 skb_put(e->skb, diff);
640         }
641         if (!skb_make_writable(&e->skb, data_len))
642                 return -ENOMEM;
643         memcpy(e->skb->data, data, data_len);
644         e->skb->ip_summed = CHECKSUM_NONE;
645         return 0;
646 }
647
648 static inline int
649 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
650 {
651         return (id == e->id);
652 }
653
654 static int
655 nfqnl_set_mode(struct nfqnl_instance *queue,
656                unsigned char mode, unsigned int range)
657 {
658         int status;
659
660         spin_lock_bh(&queue->lock);
661         status = __nfqnl_set_mode(queue, mode, range);
662         spin_unlock_bh(&queue->lock);
663
664         return status;
665 }
666
667 static int
668 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
669 {
670         if (entry->info->indev)
671                 if (entry->info->indev->ifindex == ifindex)
672                         return 1;
673                         
674         if (entry->info->outdev)
675                 if (entry->info->outdev->ifindex == ifindex)
676                         return 1;
677
678         return 0;
679 }
680
681 /* drop all packets with either indev or outdev == ifindex from all queue
682  * instances */
683 static void
684 nfqnl_dev_drop(int ifindex)
685 {
686         int i;
687         
688         QDEBUG("entering for ifindex %u\n", ifindex);
689
690         /* this only looks like we have to hold the readlock for a way too long
691          * time, issue_verdict(),  nf_reinject(), ... - but we always only
692          * issue NF_DROP, which is processed directly in nf_reinject() */
693         read_lock_bh(&instances_lock);
694
695         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
696                 struct hlist_node *tmp;
697                 struct nfqnl_instance *inst;
698                 struct hlist_head *head = &instance_table[i];
699
700                 hlist_for_each_entry(inst, tmp, head, hlist) {
701                         struct nfqnl_queue_entry *entry;
702                         while ((entry = find_dequeue_entry(inst, dev_cmp, 
703                                                            ifindex)) != NULL)
704                                 issue_verdict(entry, NF_DROP);
705                 }
706         }
707
708         read_unlock_bh(&instances_lock);
709 }
710
711 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
712
713 static int
714 nfqnl_rcv_dev_event(struct notifier_block *this,
715                     unsigned long event, void *ptr)
716 {
717         struct net_device *dev = ptr;
718
719         /* Drop any packets associated with the downed device */
720         if (event == NETDEV_DOWN)
721                 nfqnl_dev_drop(dev->ifindex);
722         return NOTIFY_DONE;
723 }
724
725 static struct notifier_block nfqnl_dev_notifier = {
726         .notifier_call  = nfqnl_rcv_dev_event,
727 };
728
729 static int
730 nfqnl_rcv_nl_event(struct notifier_block *this,
731                    unsigned long event, void *ptr)
732 {
733         struct netlink_notify *n = ptr;
734
735         if (event == NETLINK_URELEASE &&
736             n->protocol == NETLINK_NETFILTER && n->pid) {
737                 int i;
738
739                 /* destroy all instances for this pid */
740                 write_lock_bh(&instances_lock);
741                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
742                         struct hlist_node *tmp, *t2;
743                         struct nfqnl_instance *inst;
744                         struct hlist_head *head = &instance_table[i];
745
746                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
747                                 if (n->pid == inst->peer_pid)
748                                         __instance_destroy(inst);
749                         }
750                 }
751                 write_unlock_bh(&instances_lock);
752         }
753         return NOTIFY_DONE;
754 }
755
756 static struct notifier_block nfqnl_rtnl_notifier = {
757         .notifier_call  = nfqnl_rcv_nl_event,
758 };
759
760 static const int nfqa_verdict_min[NFQA_MAX] = {
761         [NFQA_VERDICT_HDR-1]    = sizeof(struct nfqnl_msg_verdict_hdr),
762         [NFQA_MARK-1]           = sizeof(u_int32_t),
763         [NFQA_PAYLOAD-1]        = 0,
764 };
765
766 static int
767 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
768                    struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
769 {
770         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
771         u_int16_t queue_num = ntohs(nfmsg->res_id);
772
773         struct nfqnl_msg_verdict_hdr *vhdr;
774         struct nfqnl_instance *queue;
775         unsigned int verdict;
776         struct nfqnl_queue_entry *entry;
777         int err;
778
779         if (nfattr_bad_size(nfqa, NFQA_MAX, nfqa_verdict_min)) {
780                 QDEBUG("bad attribute size\n");
781                 return -EINVAL;
782         }
783
784         queue = instance_lookup_get(queue_num);
785         if (!queue)
786                 return -ENODEV;
787
788         if (queue->peer_pid != NETLINK_CB(skb).pid) {
789                 err = -EPERM;
790                 goto err_out_put;
791         }
792
793         if (!nfqa[NFQA_VERDICT_HDR-1]) {
794                 err = -EINVAL;
795                 goto err_out_put;
796         }
797
798         vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
799         verdict = ntohl(vhdr->verdict);
800
801         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
802                 err = -EINVAL;
803                 goto err_out_put;
804         }
805
806         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
807         if (entry == NULL) {
808                 err = -ENOENT;
809                 goto err_out_put;
810         }
811
812         if (nfqa[NFQA_PAYLOAD-1]) {
813                 if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
814                                  NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
815                         verdict = NF_DROP;
816         }
817
818         if (nfqa[NFQA_MARK-1])
819                 skb->nfmark = ntohl(*(u_int32_t *)NFA_DATA(nfqa[NFQA_MARK-1]));
820                 
821         issue_verdict(entry, verdict);
822         instance_put(queue);
823         return 0;
824
825 err_out_put:
826         instance_put(queue);
827         return err;
828 }
829
830 static int
831 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
832                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
833 {
834         return -ENOTSUPP;
835 }
836
837 static const int nfqa_cfg_min[NFQA_CFG_MAX] = {
838         [NFQA_CFG_CMD-1]        = sizeof(struct nfqnl_msg_config_cmd),
839         [NFQA_CFG_PARAMS-1]     = sizeof(struct nfqnl_msg_config_params),
840 };
841
842 static struct nf_queue_handler nfqh = {
843         .name   = "nf_queue",
844         .outfn  = &nfqnl_enqueue_packet,
845 };
846
847 static int
848 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
849                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
850 {
851         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
852         u_int16_t queue_num = ntohs(nfmsg->res_id);
853         struct nfqnl_instance *queue;
854         int ret = 0;
855
856         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
857
858         if (nfattr_bad_size(nfqa, NFQA_CFG_MAX, nfqa_cfg_min)) {
859                 QDEBUG("bad attribute size\n");
860                 return -EINVAL;
861         }
862
863         queue = instance_lookup_get(queue_num);
864         if (nfqa[NFQA_CFG_CMD-1]) {
865                 struct nfqnl_msg_config_cmd *cmd;
866                 cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
867                 QDEBUG("found CFG_CMD\n");
868
869                 switch (cmd->command) {
870                 case NFQNL_CFG_CMD_BIND:
871                         if (queue)
872                                 return -EBUSY;
873
874                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
875                         if (!queue)
876                                 return -EINVAL;
877                         break;
878                 case NFQNL_CFG_CMD_UNBIND:
879                         if (!queue)
880                                 return -ENODEV;
881
882                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
883                                 ret = -EPERM;
884                                 goto out_put;
885                         }
886
887                         instance_destroy(queue);
888                         break;
889                 case NFQNL_CFG_CMD_PF_BIND:
890                         QDEBUG("registering queue handler for pf=%u\n",
891                                 ntohs(cmd->pf));
892                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
893                         break;
894                 case NFQNL_CFG_CMD_PF_UNBIND:
895                         QDEBUG("unregistering queue handler for pf=%u\n",
896                                 ntohs(cmd->pf));
897                         /* This is a bug and a feature.  We can unregister
898                          * other handlers(!) */
899                         ret = nf_unregister_queue_handler(ntohs(cmd->pf));
900                         break;
901                 default:
902                         ret = -EINVAL;
903                         break;
904                 }
905         } else {
906                 if (!queue) {
907                         QDEBUG("no config command, and no instance ENOENT\n");
908                         ret = -ENOENT;
909                         goto out_put;
910                 }
911
912                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
913                         QDEBUG("no config command, and wrong pid\n");
914                         ret = -EPERM;
915                         goto out_put;
916                 }
917         }
918
919         if (nfqa[NFQA_CFG_PARAMS-1]) {
920                 struct nfqnl_msg_config_params *params;
921                 params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
922
923                 nfqnl_set_mode(queue, params->copy_mode,
924                                 ntohl(params->copy_range));
925         }
926
927 out_put:
928         instance_put(queue);
929         return ret;
930 }
931
932 static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
933         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
934                                     .attr_count = NFQA_MAX,
935                                     .cap_required = CAP_NET_ADMIN },
936         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
937                                     .attr_count = NFQA_MAX,
938                                     .cap_required = CAP_NET_ADMIN },
939         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
940                                     .attr_count = NFQA_CFG_MAX,
941                                     .cap_required = CAP_NET_ADMIN },
942 };
943
944 static struct nfnetlink_subsystem nfqnl_subsys = {
945         .name           = "nf_queue",
946         .subsys_id      = NFNL_SUBSYS_QUEUE,
947         .cb_count       = NFQNL_MSG_MAX,
948         .cb             = nfqnl_cb,
949 };
950
951 #ifdef CONFIG_PROC_FS
952 struct iter_state {
953         unsigned int bucket;
954 };
955
956 static struct hlist_node *get_first(struct seq_file *seq)
957 {
958         struct iter_state *st = seq->private;
959
960         if (!st)
961                 return NULL;
962
963         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
964                 if (!hlist_empty(&instance_table[st->bucket]))
965                         return instance_table[st->bucket].first;
966         }
967         return NULL;
968 }
969
970 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
971 {
972         struct iter_state *st = seq->private;
973
974         h = h->next;
975         while (!h) {
976                 if (++st->bucket >= INSTANCE_BUCKETS)
977                         return NULL;
978
979                 h = instance_table[st->bucket].first;
980         }
981         return h;
982 }
983
984 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
985 {
986         struct hlist_node *head;
987         head = get_first(seq);
988
989         if (head)
990                 while (pos && (head = get_next(seq, head)))
991                         pos--;
992         return pos ? NULL : head;
993 }
994
995 static void *seq_start(struct seq_file *seq, loff_t *pos)
996 {
997         read_lock_bh(&instances_lock);
998         return get_idx(seq, *pos);
999 }
1000
1001 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1002 {
1003         (*pos)++;
1004         return get_next(s, v);
1005 }
1006
1007 static void seq_stop(struct seq_file *s, void *v)
1008 {
1009         read_unlock_bh(&instances_lock);
1010 }
1011
1012 static int seq_show(struct seq_file *s, void *v)
1013 {
1014         const struct nfqnl_instance *inst = v;
1015
1016         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1017                           inst->queue_num,
1018                           inst->peer_pid, inst->queue_total,
1019                           inst->copy_mode, inst->copy_range,
1020                           inst->queue_dropped, inst->queue_user_dropped,
1021                           atomic_read(&inst->id_sequence),
1022                           atomic_read(&inst->use));
1023 }
1024
1025 static struct seq_operations nfqnl_seq_ops = {
1026         .start  = seq_start,
1027         .next   = seq_next,
1028         .stop   = seq_stop,
1029         .show   = seq_show,
1030 };
1031
1032 static int nfqnl_open(struct inode *inode, struct file *file)
1033 {
1034         struct seq_file *seq;
1035         struct iter_state *is;
1036         int ret;
1037
1038         is = kzalloc(sizeof(*is), GFP_KERNEL);
1039         if (!is)
1040                 return -ENOMEM;
1041         ret = seq_open(file, &nfqnl_seq_ops);
1042         if (ret < 0)
1043                 goto out_free;
1044         seq = file->private_data;
1045         seq->private = is;
1046         return ret;
1047 out_free:
1048         kfree(is);
1049         return ret;
1050 }
1051
1052 static struct file_operations nfqnl_file_ops = {
1053         .owner   = THIS_MODULE,
1054         .open    = nfqnl_open,
1055         .read    = seq_read,
1056         .llseek  = seq_lseek,
1057         .release = seq_release_private,
1058 };
1059
1060 #endif /* PROC_FS */
1061
1062 static int
1063 init_or_cleanup(int init)
1064 {
1065         int i, status = -ENOMEM;
1066 #ifdef CONFIG_PROC_FS
1067         struct proc_dir_entry *proc_nfqueue;
1068 #endif
1069         
1070         if (!init)
1071                 goto cleanup;
1072
1073         for (i = 0; i < INSTANCE_BUCKETS; i++)
1074                 INIT_HLIST_HEAD(&instance_table[i]);
1075
1076         netlink_register_notifier(&nfqnl_rtnl_notifier);
1077         status = nfnetlink_subsys_register(&nfqnl_subsys);
1078         if (status < 0) {
1079                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1080                 goto cleanup_netlink_notifier;
1081         }
1082
1083 #ifdef CONFIG_PROC_FS
1084         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1085                                          proc_net_netfilter);
1086         if (!proc_nfqueue)
1087                 goto cleanup_subsys;
1088         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1089 #endif
1090
1091         register_netdevice_notifier(&nfqnl_dev_notifier);
1092
1093         return status;
1094
1095 cleanup:
1096         nf_unregister_queue_handlers(&nfqh);
1097         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1098 #ifdef CONFIG_PROC_FS
1099         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1100 cleanup_subsys:
1101 #endif  
1102         nfnetlink_subsys_unregister(&nfqnl_subsys);
1103 cleanup_netlink_notifier:
1104         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1105         return status;
1106 }
1107
1108 static int __init init(void)
1109 {
1110         
1111         return init_or_cleanup(1);
1112 }
1113
1114 static void __exit fini(void)
1115 {
1116         init_or_cleanup(0);
1117 }
1118
1119 MODULE_DESCRIPTION("netfilter packet queue handler");
1120 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1121 MODULE_LICENSE("GPL");
1122 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1123
1124 module_init(init);
1125 module_exit(fini);