Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
[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 nfgenmsg)) +
358                 + NFA_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
359                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
360                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
361 #ifdef CONFIG_BRIDGE_NETFILTER
362                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
363                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
364 #endif
365                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
366                 + NFA_SPACE(sizeof(struct nfqnl_msg_packet_hw))
367                 + NFA_SPACE(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 += NFA_SPACE(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        = htons(entskb->protocol);
418         pmsg.hook               = entinf->hook;
419
420         NFA_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                 NFA_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                         NFA_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                         NFA_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                         NFA_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                                 NFA_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                 NFA_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                         NFA_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                         NFA_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                         NFA_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                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
477                                         sizeof(tmp_uint), &tmp_uint);
478                         }
479                 }
480 #endif
481         }
482
483         if (entskb->nfmark) {
484                 tmp_uint = htonl(entskb->nfmark);
485                 NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
486         }
487
488         if (indev && entskb->dev
489             && entskb->dev->hard_header_parse) {
490                 struct nfqnl_msg_packet_hw phw;
491
492                 phw.hw_addrlen =
493                         entskb->dev->hard_header_parse(entskb,
494                                                            phw.hw_addr);
495                 phw.hw_addrlen = htons(phw.hw_addrlen);
496                 NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
497         }
498
499         if (entskb->tstamp.off_sec) {
500                 struct nfqnl_msg_packet_timestamp ts;
501
502                 ts.sec = cpu_to_be64(entskb->tstamp.off_sec);
503                 ts.usec = cpu_to_be64(entskb->tstamp.off_usec);
504
505                 NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
506         }
507
508         if (data_len) {
509                 struct nfattr *nfa;
510                 int size = NFA_LENGTH(data_len);
511
512                 if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
513                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
514                         goto nlmsg_failure;
515                 }
516
517                 nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
518                 nfa->nfa_type = NFQA_PAYLOAD;
519                 nfa->nfa_len = size;
520
521                 if (skb_copy_bits(entskb, 0, NFA_DATA(nfa), data_len))
522                         BUG();
523         }
524                 
525         nlh->nlmsg_len = skb->tail - old_tail;
526         return skb;
527
528 nlmsg_failure:
529 nfattr_failure:
530         if (skb)
531                 kfree_skb(skb);
532         *errp = -EINVAL;
533         if (net_ratelimit())
534                 printk(KERN_ERR "nf_queue: error creating packet message\n");
535         return NULL;
536 }
537
538 static int
539 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info, 
540                      unsigned int queuenum, void *data)
541 {
542         int status = -EINVAL;
543         struct sk_buff *nskb;
544         struct nfqnl_instance *queue;
545         struct nfqnl_queue_entry *entry;
546
547         QDEBUG("entered\n");
548
549         queue = instance_lookup_get(queuenum);
550         if (!queue) {
551                 QDEBUG("no queue instance matching\n");
552                 return -EINVAL;
553         }
554
555         if (queue->copy_mode == NFQNL_COPY_NONE) {
556                 QDEBUG("mode COPY_NONE, aborting\n");
557                 status = -EAGAIN;
558                 goto err_out_put;
559         }
560
561         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
562         if (entry == NULL) {
563                 if (net_ratelimit())
564                         printk(KERN_ERR 
565                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
566                 status = -ENOMEM;
567                 goto err_out_put;
568         }
569
570         entry->info = info;
571         entry->skb = skb;
572         entry->id = atomic_inc_return(&queue->id_sequence);
573
574         nskb = nfqnl_build_packet_message(queue, entry, &status);
575         if (nskb == NULL)
576                 goto err_out_free;
577                 
578         spin_lock_bh(&queue->lock);
579         
580         if (!queue->peer_pid)
581                 goto err_out_free_nskb; 
582
583         if (queue->queue_total >= queue->queue_maxlen) {
584                 queue->queue_dropped++;
585                 status = -ENOSPC;
586                 if (net_ratelimit())
587                           printk(KERN_WARNING "nf_queue: full at %d entries, "
588                                  "dropping packets(s). Dropped: %d\n", 
589                                  queue->queue_total, queue->queue_dropped);
590                 goto err_out_free_nskb;
591         }
592
593         /* nfnetlink_unicast will either free the nskb or add it to a socket */
594         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
595         if (status < 0) {
596                 queue->queue_user_dropped++;
597                 goto err_out_unlock;
598         }
599
600         __enqueue_entry(queue, entry);
601
602         spin_unlock_bh(&queue->lock);
603         instance_put(queue);
604         return status;
605
606 err_out_free_nskb:
607         kfree_skb(nskb); 
608         
609 err_out_unlock:
610         spin_unlock_bh(&queue->lock);
611
612 err_out_free:
613         kfree(entry);
614 err_out_put:
615         instance_put(queue);
616         return status;
617 }
618
619 static int
620 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
621 {
622         int diff;
623
624         diff = data_len - e->skb->len;
625         if (diff < 0)
626                 skb_trim(e->skb, data_len);
627         else if (diff > 0) {
628                 if (data_len > 0xFFFF)
629                         return -EINVAL;
630                 if (diff > skb_tailroom(e->skb)) {
631                         struct sk_buff *newskb;
632                         
633                         newskb = skb_copy_expand(e->skb,
634                                                  skb_headroom(e->skb),
635                                                  diff,
636                                                  GFP_ATOMIC);
637                         if (newskb == NULL) {
638                                 printk(KERN_WARNING "nf_queue: OOM "
639                                       "in mangle, dropping packet\n");
640                                 return -ENOMEM;
641                         }
642                         if (e->skb->sk)
643                                 skb_set_owner_w(newskb, e->skb->sk);
644                         kfree_skb(e->skb);
645                         e->skb = newskb;
646                 }
647                 skb_put(e->skb, diff);
648         }
649         if (!skb_make_writable(&e->skb, data_len))
650                 return -ENOMEM;
651         memcpy(e->skb->data, data, data_len);
652         e->skb->ip_summed = CHECKSUM_NONE;
653         return 0;
654 }
655
656 static inline int
657 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
658 {
659         return (id == e->id);
660 }
661
662 static int
663 nfqnl_set_mode(struct nfqnl_instance *queue,
664                unsigned char mode, unsigned int range)
665 {
666         int status;
667
668         spin_lock_bh(&queue->lock);
669         status = __nfqnl_set_mode(queue, mode, range);
670         spin_unlock_bh(&queue->lock);
671
672         return status;
673 }
674
675 static int
676 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
677 {
678         struct nf_info *entinf = entry->info;
679         
680         if (entinf->indev)
681                 if (entinf->indev->ifindex == ifindex)
682                         return 1;
683         if (entinf->outdev)
684                 if (entinf->outdev->ifindex == ifindex)
685                         return 1;
686 #ifdef CONFIG_BRIDGE_NETFILTER
687         if (entry->skb->nf_bridge) {
688                 if (entry->skb->nf_bridge->physindev &&
689                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
690                         return 1;
691                 if (entry->skb->nf_bridge->physoutdev &&
692                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
693                         return 1;
694         }
695 #endif
696         return 0;
697 }
698
699 /* drop all packets with either indev or outdev == ifindex from all queue
700  * instances */
701 static void
702 nfqnl_dev_drop(int ifindex)
703 {
704         int i;
705         
706         QDEBUG("entering for ifindex %u\n", ifindex);
707
708         /* this only looks like we have to hold the readlock for a way too long
709          * time, issue_verdict(),  nf_reinject(), ... - but we always only
710          * issue NF_DROP, which is processed directly in nf_reinject() */
711         read_lock_bh(&instances_lock);
712
713         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
714                 struct hlist_node *tmp;
715                 struct nfqnl_instance *inst;
716                 struct hlist_head *head = &instance_table[i];
717
718                 hlist_for_each_entry(inst, tmp, head, hlist) {
719                         struct nfqnl_queue_entry *entry;
720                         while ((entry = find_dequeue_entry(inst, dev_cmp, 
721                                                            ifindex)) != NULL)
722                                 issue_verdict(entry, NF_DROP);
723                 }
724         }
725
726         read_unlock_bh(&instances_lock);
727 }
728
729 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
730
731 static int
732 nfqnl_rcv_dev_event(struct notifier_block *this,
733                     unsigned long event, void *ptr)
734 {
735         struct net_device *dev = ptr;
736
737         /* Drop any packets associated with the downed device */
738         if (event == NETDEV_DOWN)
739                 nfqnl_dev_drop(dev->ifindex);
740         return NOTIFY_DONE;
741 }
742
743 static struct notifier_block nfqnl_dev_notifier = {
744         .notifier_call  = nfqnl_rcv_dev_event,
745 };
746
747 static int
748 nfqnl_rcv_nl_event(struct notifier_block *this,
749                    unsigned long event, void *ptr)
750 {
751         struct netlink_notify *n = ptr;
752
753         if (event == NETLINK_URELEASE &&
754             n->protocol == NETLINK_NETFILTER && n->pid) {
755                 int i;
756
757                 /* destroy all instances for this pid */
758                 write_lock_bh(&instances_lock);
759                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
760                         struct hlist_node *tmp, *t2;
761                         struct nfqnl_instance *inst;
762                         struct hlist_head *head = &instance_table[i];
763
764                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
765                                 if (n->pid == inst->peer_pid)
766                                         __instance_destroy(inst);
767                         }
768                 }
769                 write_unlock_bh(&instances_lock);
770         }
771         return NOTIFY_DONE;
772 }
773
774 static struct notifier_block nfqnl_rtnl_notifier = {
775         .notifier_call  = nfqnl_rcv_nl_event,
776 };
777
778 static const int nfqa_verdict_min[NFQA_MAX] = {
779         [NFQA_VERDICT_HDR-1]    = sizeof(struct nfqnl_msg_verdict_hdr),
780         [NFQA_MARK-1]           = sizeof(u_int32_t),
781         [NFQA_PAYLOAD-1]        = 0,
782 };
783
784 static int
785 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
786                    struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
787 {
788         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
789         u_int16_t queue_num = ntohs(nfmsg->res_id);
790
791         struct nfqnl_msg_verdict_hdr *vhdr;
792         struct nfqnl_instance *queue;
793         unsigned int verdict;
794         struct nfqnl_queue_entry *entry;
795         int err;
796
797         if (nfattr_bad_size(nfqa, NFQA_MAX, nfqa_verdict_min)) {
798                 QDEBUG("bad attribute size\n");
799                 return -EINVAL;
800         }
801
802         queue = instance_lookup_get(queue_num);
803         if (!queue)
804                 return -ENODEV;
805
806         if (queue->peer_pid != NETLINK_CB(skb).pid) {
807                 err = -EPERM;
808                 goto err_out_put;
809         }
810
811         if (!nfqa[NFQA_VERDICT_HDR-1]) {
812                 err = -EINVAL;
813                 goto err_out_put;
814         }
815
816         vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
817         verdict = ntohl(vhdr->verdict);
818
819         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
820                 err = -EINVAL;
821                 goto err_out_put;
822         }
823
824         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
825         if (entry == NULL) {
826                 err = -ENOENT;
827                 goto err_out_put;
828         }
829
830         if (nfqa[NFQA_PAYLOAD-1]) {
831                 if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
832                                  NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
833                         verdict = NF_DROP;
834         }
835
836         if (nfqa[NFQA_MARK-1])
837                 entry->skb->nfmark = ntohl(*(u_int32_t *)
838                                            NFA_DATA(nfqa[NFQA_MARK-1]));
839                 
840         issue_verdict(entry, verdict);
841         instance_put(queue);
842         return 0;
843
844 err_out_put:
845         instance_put(queue);
846         return err;
847 }
848
849 static int
850 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
851                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
852 {
853         return -ENOTSUPP;
854 }
855
856 static const int nfqa_cfg_min[NFQA_CFG_MAX] = {
857         [NFQA_CFG_CMD-1]        = sizeof(struct nfqnl_msg_config_cmd),
858         [NFQA_CFG_PARAMS-1]     = sizeof(struct nfqnl_msg_config_params),
859 };
860
861 static struct nf_queue_handler nfqh = {
862         .name   = "nf_queue",
863         .outfn  = &nfqnl_enqueue_packet,
864 };
865
866 static int
867 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
868                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
869 {
870         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
871         u_int16_t queue_num = ntohs(nfmsg->res_id);
872         struct nfqnl_instance *queue;
873         int ret = 0;
874
875         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
876
877         if (nfattr_bad_size(nfqa, NFQA_CFG_MAX, nfqa_cfg_min)) {
878                 QDEBUG("bad attribute size\n");
879                 return -EINVAL;
880         }
881
882         queue = instance_lookup_get(queue_num);
883         if (nfqa[NFQA_CFG_CMD-1]) {
884                 struct nfqnl_msg_config_cmd *cmd;
885                 cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
886                 QDEBUG("found CFG_CMD\n");
887
888                 switch (cmd->command) {
889                 case NFQNL_CFG_CMD_BIND:
890                         if (queue)
891                                 return -EBUSY;
892
893                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
894                         if (!queue)
895                                 return -EINVAL;
896                         break;
897                 case NFQNL_CFG_CMD_UNBIND:
898                         if (!queue)
899                                 return -ENODEV;
900
901                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
902                                 ret = -EPERM;
903                                 goto out_put;
904                         }
905
906                         instance_destroy(queue);
907                         break;
908                 case NFQNL_CFG_CMD_PF_BIND:
909                         QDEBUG("registering queue handler for pf=%u\n",
910                                 ntohs(cmd->pf));
911                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
912                         break;
913                 case NFQNL_CFG_CMD_PF_UNBIND:
914                         QDEBUG("unregistering queue handler for pf=%u\n",
915                                 ntohs(cmd->pf));
916                         /* This is a bug and a feature.  We can unregister
917                          * other handlers(!) */
918                         ret = nf_unregister_queue_handler(ntohs(cmd->pf));
919                         break;
920                 default:
921                         ret = -EINVAL;
922                         break;
923                 }
924         } else {
925                 if (!queue) {
926                         QDEBUG("no config command, and no instance ENOENT\n");
927                         ret = -ENOENT;
928                         goto out_put;
929                 }
930
931                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
932                         QDEBUG("no config command, and wrong pid\n");
933                         ret = -EPERM;
934                         goto out_put;
935                 }
936         }
937
938         if (nfqa[NFQA_CFG_PARAMS-1]) {
939                 struct nfqnl_msg_config_params *params;
940
941                 if (!queue) {
942                         ret = -ENOENT;
943                         goto out_put;
944                 }
945                 params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
946                 nfqnl_set_mode(queue, params->copy_mode,
947                                 ntohl(params->copy_range));
948         }
949
950 out_put:
951         instance_put(queue);
952         return ret;
953 }
954
955 static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
956         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
957                                     .attr_count = NFQA_MAX, },
958         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
959                                     .attr_count = NFQA_MAX, },
960         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
961                                     .attr_count = NFQA_CFG_MAX, },
962 };
963
964 static struct nfnetlink_subsystem nfqnl_subsys = {
965         .name           = "nf_queue",
966         .subsys_id      = NFNL_SUBSYS_QUEUE,
967         .cb_count       = NFQNL_MSG_MAX,
968         .cb             = nfqnl_cb,
969 };
970
971 #ifdef CONFIG_PROC_FS
972 struct iter_state {
973         unsigned int bucket;
974 };
975
976 static struct hlist_node *get_first(struct seq_file *seq)
977 {
978         struct iter_state *st = seq->private;
979
980         if (!st)
981                 return NULL;
982
983         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
984                 if (!hlist_empty(&instance_table[st->bucket]))
985                         return instance_table[st->bucket].first;
986         }
987         return NULL;
988 }
989
990 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
991 {
992         struct iter_state *st = seq->private;
993
994         h = h->next;
995         while (!h) {
996                 if (++st->bucket >= INSTANCE_BUCKETS)
997                         return NULL;
998
999                 h = instance_table[st->bucket].first;
1000         }
1001         return h;
1002 }
1003
1004 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1005 {
1006         struct hlist_node *head;
1007         head = get_first(seq);
1008
1009         if (head)
1010                 while (pos && (head = get_next(seq, head)))
1011                         pos--;
1012         return pos ? NULL : head;
1013 }
1014
1015 static void *seq_start(struct seq_file *seq, loff_t *pos)
1016 {
1017         read_lock_bh(&instances_lock);
1018         return get_idx(seq, *pos);
1019 }
1020
1021 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1022 {
1023         (*pos)++;
1024         return get_next(s, v);
1025 }
1026
1027 static void seq_stop(struct seq_file *s, void *v)
1028 {
1029         read_unlock_bh(&instances_lock);
1030 }
1031
1032 static int seq_show(struct seq_file *s, void *v)
1033 {
1034         const struct nfqnl_instance *inst = v;
1035
1036         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1037                           inst->queue_num,
1038                           inst->peer_pid, inst->queue_total,
1039                           inst->copy_mode, inst->copy_range,
1040                           inst->queue_dropped, inst->queue_user_dropped,
1041                           atomic_read(&inst->id_sequence),
1042                           atomic_read(&inst->use));
1043 }
1044
1045 static struct seq_operations nfqnl_seq_ops = {
1046         .start  = seq_start,
1047         .next   = seq_next,
1048         .stop   = seq_stop,
1049         .show   = seq_show,
1050 };
1051
1052 static int nfqnl_open(struct inode *inode, struct file *file)
1053 {
1054         struct seq_file *seq;
1055         struct iter_state *is;
1056         int ret;
1057
1058         is = kzalloc(sizeof(*is), GFP_KERNEL);
1059         if (!is)
1060                 return -ENOMEM;
1061         ret = seq_open(file, &nfqnl_seq_ops);
1062         if (ret < 0)
1063                 goto out_free;
1064         seq = file->private_data;
1065         seq->private = is;
1066         return ret;
1067 out_free:
1068         kfree(is);
1069         return ret;
1070 }
1071
1072 static struct file_operations nfqnl_file_ops = {
1073         .owner   = THIS_MODULE,
1074         .open    = nfqnl_open,
1075         .read    = seq_read,
1076         .llseek  = seq_lseek,
1077         .release = seq_release_private,
1078 };
1079
1080 #endif /* PROC_FS */
1081
1082 static int __init nfnetlink_queue_init(void)
1083 {
1084         int i, status = -ENOMEM;
1085 #ifdef CONFIG_PROC_FS
1086         struct proc_dir_entry *proc_nfqueue;
1087 #endif
1088         
1089         for (i = 0; i < INSTANCE_BUCKETS; i++)
1090                 INIT_HLIST_HEAD(&instance_table[i]);
1091
1092         netlink_register_notifier(&nfqnl_rtnl_notifier);
1093         status = nfnetlink_subsys_register(&nfqnl_subsys);
1094         if (status < 0) {
1095                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1096                 goto cleanup_netlink_notifier;
1097         }
1098
1099 #ifdef CONFIG_PROC_FS
1100         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1101                                          proc_net_netfilter);
1102         if (!proc_nfqueue)
1103                 goto cleanup_subsys;
1104         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1105 #endif
1106
1107         register_netdevice_notifier(&nfqnl_dev_notifier);
1108         return status;
1109
1110 #ifdef CONFIG_PROC_FS
1111 cleanup_subsys:
1112         nfnetlink_subsys_unregister(&nfqnl_subsys);
1113 #endif
1114 cleanup_netlink_notifier:
1115         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1116         return status;
1117 }
1118
1119 static void __exit nfnetlink_queue_fini(void)
1120 {
1121         nf_unregister_queue_handlers(&nfqh);
1122         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1123 #ifdef CONFIG_PROC_FS
1124         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1125 #endif
1126         nfnetlink_subsys_unregister(&nfqnl_subsys);
1127         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
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(nfnetlink_queue_init);
1136 module_exit(nfnetlink_queue_fini);