[NET]: Fix packet timestamping.
[linux-3.10.git] / net / netfilter / nfnetlink_queue.c
1 /*
2  * This is a module which is used for queueing packets and communicating with
3  * userspace via nfetlink.
4  *
5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
6  *
7  * Based on the old ipv4-only ip_queue.c:
8  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
9  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of the GNU General Public License version 2 as
13  * published by the Free Software Foundation.
14  *
15  */
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/init.h>
19 #include <linux/spinlock.h>
20 #include <linux/notifier.h>
21 #include <linux/netdevice.h>
22 #include <linux/netfilter.h>
23 #include <linux/proc_fs.h>
24 #include <linux/netfilter_ipv4.h>
25 #include <linux/netfilter_ipv6.h>
26 #include <linux/netfilter/nfnetlink.h>
27 #include <linux/netfilter/nfnetlink_queue.h>
28 #include <linux/list.h>
29 #include <net/sock.h>
30
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 = kmalloc(sizeof(*inst), GFP_ATOMIC);
140         if (!inst)
141                 goto out_unlock;
142
143         memset(inst, 0, sizeof(*inst));
144         inst->queue_num = queue_num;
145         inst->peer_pid = pid;
146         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
147         inst->copy_range = 0xfffff;
148         inst->copy_mode = NFQNL_COPY_NONE;
149         atomic_set(&inst->id_sequence, 0);
150         /* needs to be two, since we _put() after creation */
151         atomic_set(&inst->use, 2);
152         inst->lock = SPIN_LOCK_UNLOCKED;
153         INIT_LIST_HEAD(&inst->queue_list);
154
155         if (!try_module_get(THIS_MODULE))
156                 goto out_free;
157
158         hlist_add_head(&inst->hlist, 
159                        &instance_table[instance_hashfn(queue_num)]);
160
161         write_unlock_bh(&instances_lock);
162
163         QDEBUG("successfully created new instance\n");
164
165         return inst;
166
167 out_free:
168         kfree(inst);
169 out_unlock:
170         write_unlock_bh(&instances_lock);
171         return NULL;
172 }
173
174 static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
175
176 static void
177 _instance_destroy2(struct nfqnl_instance *inst, int lock)
178 {
179         /* first pull it out of the global list */
180         if (lock)
181                 write_lock_bh(&instances_lock);
182
183         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
184                 inst, inst->queue_num);
185         hlist_del(&inst->hlist);
186
187         if (lock)
188                 write_unlock_bh(&instances_lock);
189
190         /* then flush all pending skbs from the queue */
191         nfqnl_flush(inst, NF_DROP);
192
193         /* and finally put the refcount */
194         instance_put(inst);
195
196         module_put(THIS_MODULE);
197 }
198
199 static inline void
200 __instance_destroy(struct nfqnl_instance *inst)
201 {
202         _instance_destroy2(inst, 0);
203 }
204
205 static inline void
206 instance_destroy(struct nfqnl_instance *inst)
207 {
208         _instance_destroy2(inst, 1);
209 }
210
211
212
213 static void
214 issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
215 {
216         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
217
218         /* TCP input path (and probably other bits) assume to be called
219          * from softirq context, not from syscall, like issue_verdict is
220          * called.  TCP input path deadlocks with locks taken from timer
221          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
222
223         local_bh_disable();
224         nf_reinject(entry->skb, entry->info, verdict);
225         local_bh_enable();
226
227         kfree(entry);
228 }
229
230 static inline void
231 __enqueue_entry(struct nfqnl_instance *queue,
232                       struct nfqnl_queue_entry *entry)
233 {
234        list_add(&entry->list, &queue->queue_list);
235        queue->queue_total++;
236 }
237
238 /*
239  * Find and return a queued entry matched by cmpfn, or return the last
240  * entry if cmpfn is NULL.
241  */
242 static inline struct nfqnl_queue_entry *
243 __find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, 
244                    unsigned long data)
245 {
246         struct list_head *p;
247
248         list_for_each_prev(p, &queue->queue_list) {
249                 struct nfqnl_queue_entry *entry = (struct nfqnl_queue_entry *)p;
250                 
251                 if (!cmpfn || cmpfn(entry, data))
252                         return entry;
253         }
254         return NULL;
255 }
256
257 static inline void
258 __dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
259 {
260         list_del(&entry->list);
261         q->queue_total--;
262 }
263
264 static inline struct nfqnl_queue_entry *
265 __find_dequeue_entry(struct nfqnl_instance *queue,
266                      nfqnl_cmpfn cmpfn, unsigned long data)
267 {
268         struct nfqnl_queue_entry *entry;
269
270         entry = __find_entry(queue, cmpfn, data);
271         if (entry == NULL)
272                 return NULL;
273
274         __dequeue_entry(queue, entry);
275         return entry;
276 }
277
278
279 static inline void
280 __nfqnl_flush(struct nfqnl_instance *queue, int verdict)
281 {
282         struct nfqnl_queue_entry *entry;
283         
284         while ((entry = __find_dequeue_entry(queue, NULL, 0)))
285                 issue_verdict(entry, verdict);
286 }
287
288 static inline int
289 __nfqnl_set_mode(struct nfqnl_instance *queue,
290                  unsigned char mode, unsigned int range)
291 {
292         int status = 0;
293         
294         switch (mode) {
295         case NFQNL_COPY_NONE:
296         case NFQNL_COPY_META:
297                 queue->copy_mode = mode;
298                 queue->copy_range = 0;
299                 break;
300                 
301         case NFQNL_COPY_PACKET:
302                 queue->copy_mode = mode;
303                 /* we're using struct nfattr which has 16bit nfa_len */
304                 if (range > 0xffff)
305                         queue->copy_range = 0xffff;
306                 else
307                         queue->copy_range = range;
308                 break;
309                 
310         default:
311                 status = -EINVAL;
312
313         }
314         return status;
315 }
316
317 static struct nfqnl_queue_entry *
318 find_dequeue_entry(struct nfqnl_instance *queue,
319                          nfqnl_cmpfn cmpfn, unsigned long data)
320 {
321         struct nfqnl_queue_entry *entry;
322         
323         spin_lock_bh(&queue->lock);
324         entry = __find_dequeue_entry(queue, cmpfn, data);
325         spin_unlock_bh(&queue->lock);
326
327         return entry;
328 }
329
330 static void
331 nfqnl_flush(struct nfqnl_instance *queue, int verdict)
332 {
333         spin_lock_bh(&queue->lock);
334         __nfqnl_flush(queue, verdict);
335         spin_unlock_bh(&queue->lock);
336 }
337
338 static struct sk_buff *
339 nfqnl_build_packet_message(struct nfqnl_instance *queue,
340                            struct nfqnl_queue_entry *entry, int *errp)
341 {
342         unsigned char *old_tail;
343         size_t size;
344         size_t data_len = 0;
345         struct sk_buff *skb;
346         struct nfqnl_msg_packet_hdr pmsg;
347         struct nlmsghdr *nlh;
348         struct nfgenmsg *nfmsg;
349         unsigned int tmp_uint;
350
351         QDEBUG("entered\n");
352
353         /* all macros expand to constant values at compile time */
354         size =    NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
355                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
356                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
357 #ifdef CONFIG_BRIDGE_NETFILTER
358                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
359                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
360 #endif
361                 + NLMSG_SPACE(sizeof(u_int32_t))        /* mark */
362                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hw))
363                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_timestamp));
364
365         spin_lock_bh(&queue->lock);
366         
367         switch (queue->copy_mode) {
368         case NFQNL_COPY_META:
369         case NFQNL_COPY_NONE:
370                 data_len = 0;
371                 break;
372         
373         case NFQNL_COPY_PACKET:
374                 if (entry->skb->ip_summed == CHECKSUM_HW &&
375                     (*errp = skb_checksum_help(entry->skb,
376                                                entry->info->outdev == NULL))) {
377                         spin_unlock_bh(&queue->lock);
378                         return NULL;
379                 }
380                 if (queue->copy_range == 0 
381                     || queue->copy_range > entry->skb->len)
382                         data_len = entry->skb->len;
383                 else
384                         data_len = queue->copy_range;
385                 
386                 size += NLMSG_SPACE(data_len);
387                 break;
388         
389         default:
390                 *errp = -EINVAL;
391                 spin_unlock_bh(&queue->lock);
392                 return NULL;
393         }
394
395         spin_unlock_bh(&queue->lock);
396
397         skb = alloc_skb(size, GFP_ATOMIC);
398         if (!skb)
399                 goto nlmsg_failure;
400                 
401         old_tail= skb->tail;
402         nlh = NLMSG_PUT(skb, 0, 0, 
403                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
404                         sizeof(struct nfgenmsg));
405         nfmsg = NLMSG_DATA(nlh);
406         nfmsg->nfgen_family = entry->info->pf;
407         nfmsg->version = NFNETLINK_V0;
408         nfmsg->res_id = htons(queue->queue_num);
409
410         pmsg.packet_id          = htonl(entry->id);
411         pmsg.hw_protocol        = htons(entry->skb->protocol);
412         pmsg.hook               = entry->info->hook;
413
414         NFA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
415
416         if (entry->info->indev) {
417                 tmp_uint = htonl(entry->info->indev->ifindex);
418 #ifndef CONFIG_BRIDGE_NETFILTER
419                 NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
420 #else
421                 if (entry->info->pf == PF_BRIDGE) {
422                         /* Case 1: indev is physical input device, we need to
423                          * look for bridge group (when called from 
424                          * netfilter_bridge) */
425                         NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint), 
426                                 &tmp_uint);
427                         /* this is the bridge group "brX" */
428                         tmp_uint = htonl(entry->info->indev->br_port->br->dev->ifindex);
429                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
430                                 &tmp_uint);
431                 } else {
432                         /* Case 2: indev is bridge group, we need to look for
433                          * physical device (when called from ipv4) */
434                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
435                                 &tmp_uint);
436                         if (entry->skb->nf_bridge
437                             && entry->skb->nf_bridge->physindev) {
438                                 tmp_uint = htonl(entry->skb->nf_bridge->physindev->ifindex);
439                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
440                                         sizeof(tmp_uint), &tmp_uint);
441                         }
442                 }
443 #endif
444         }
445
446         if (entry->info->outdev) {
447                 tmp_uint = htonl(entry->info->outdev->ifindex);
448 #ifndef CONFIG_BRIDGE_NETFILTER
449                 NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
450 #else
451                 if (entry->info->pf == PF_BRIDGE) {
452                         /* Case 1: outdev is physical output device, we need to
453                          * look for bridge group (when called from 
454                          * netfilter_bridge) */
455                         NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
456                                 &tmp_uint);
457                         /* this is the bridge group "brX" */
458                         tmp_uint = htonl(entry->info->outdev->br_port->br->dev->ifindex);
459                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
460                                 &tmp_uint);
461                 } else {
462                         /* Case 2: outdev is bridge group, we need to look for
463                          * physical output device (when called from ipv4) */
464                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
465                                 &tmp_uint);
466                         if (entry->skb->nf_bridge
467                             && entry->skb->nf_bridge->physoutdev) {
468                                 tmp_uint = htonl(entry->skb->nf_bridge->physoutdev->ifindex);
469                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
470                                         sizeof(tmp_uint), &tmp_uint);
471                         }
472                 }
473 #endif
474         }
475
476         if (entry->skb->nfmark) {
477                 tmp_uint = htonl(entry->skb->nfmark);
478                 NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
479         }
480
481         if (entry->info->indev && entry->skb->dev
482             && entry->skb->dev->hard_header_parse) {
483                 struct nfqnl_msg_packet_hw phw;
484
485                 phw.hw_addrlen =
486                         entry->skb->dev->hard_header_parse(entry->skb,
487                                                            phw.hw_addr);
488                 phw.hw_addrlen = htons(phw.hw_addrlen);
489                 NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
490         }
491
492         if (entry->skb->tstamp.off_sec) {
493                 struct nfqnl_msg_packet_timestamp ts;
494
495                 ts.sec = cpu_to_be64(entry->skb->tstamp.off_sec);
496                 ts.usec = cpu_to_be64(entry->skb->tstamp.off_usec);
497
498                 NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
499         }
500
501         if (data_len) {
502                 struct nfattr *nfa;
503                 int size = NFA_LENGTH(data_len);
504
505                 if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
506                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
507                         goto nlmsg_failure;
508                 }
509
510                 nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
511                 nfa->nfa_type = NFQA_PAYLOAD;
512                 nfa->nfa_len = size;
513
514                 if (skb_copy_bits(entry->skb, 0, NFA_DATA(nfa), data_len))
515                         BUG();
516         }
517                 
518         nlh->nlmsg_len = skb->tail - old_tail;
519         return skb;
520
521 nlmsg_failure:
522 nfattr_failure:
523         if (skb)
524                 kfree_skb(skb);
525         *errp = -EINVAL;
526         if (net_ratelimit())
527                 printk(KERN_ERR "nf_queue: error creating packet message\n");
528         return NULL;
529 }
530
531 static int
532 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info, 
533                      unsigned int queuenum, void *data)
534 {
535         int status = -EINVAL;
536         struct sk_buff *nskb;
537         struct nfqnl_instance *queue;
538         struct nfqnl_queue_entry *entry;
539
540         QDEBUG("entered\n");
541
542         queue = instance_lookup_get(queuenum);
543         if (!queue) {
544                 QDEBUG("no queue instance matching\n");
545                 return -EINVAL;
546         }
547
548         if (queue->copy_mode == NFQNL_COPY_NONE) {
549                 QDEBUG("mode COPY_NONE, aborting\n");
550                 status = -EAGAIN;
551                 goto err_out_put;
552         }
553
554         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
555         if (entry == NULL) {
556                 if (net_ratelimit())
557                         printk(KERN_ERR 
558                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
559                 status = -ENOMEM;
560                 goto err_out_put;
561         }
562
563         entry->info = info;
564         entry->skb = skb;
565         entry->id = atomic_inc_return(&queue->id_sequence);
566
567         nskb = nfqnl_build_packet_message(queue, entry, &status);
568         if (nskb == NULL)
569                 goto err_out_free;
570                 
571         spin_lock_bh(&queue->lock);
572         
573         if (!queue->peer_pid)
574                 goto err_out_free_nskb; 
575
576         if (queue->queue_total >= queue->queue_maxlen) {
577                 queue->queue_dropped++;
578                 status = -ENOSPC;
579                 if (net_ratelimit())
580                           printk(KERN_WARNING "ip_queue: full at %d entries, "
581                                  "dropping packets(s). Dropped: %d\n", 
582                                  queue->queue_total, queue->queue_dropped);
583                 goto err_out_free_nskb;
584         }
585
586         /* nfnetlink_unicast will either free the nskb or add it to a socket */
587         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
588         if (status < 0) {
589                 queue->queue_user_dropped++;
590                 goto err_out_unlock;
591         }
592
593         __enqueue_entry(queue, entry);
594
595         spin_unlock_bh(&queue->lock);
596         instance_put(queue);
597         return status;
598
599 err_out_free_nskb:
600         kfree_skb(nskb); 
601         
602 err_out_unlock:
603         spin_unlock_bh(&queue->lock);
604
605 err_out_free:
606         kfree(entry);
607 err_out_put:
608         instance_put(queue);
609         return status;
610 }
611
612 static int
613 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
614 {
615         int diff;
616
617         diff = data_len - e->skb->len;
618         if (diff < 0)
619                 skb_trim(e->skb, data_len);
620         else if (diff > 0) {
621                 if (data_len > 0xFFFF)
622                         return -EINVAL;
623                 if (diff > skb_tailroom(e->skb)) {
624                         struct sk_buff *newskb;
625                         
626                         newskb = skb_copy_expand(e->skb,
627                                                  skb_headroom(e->skb),
628                                                  diff,
629                                                  GFP_ATOMIC);
630                         if (newskb == NULL) {
631                                 printk(KERN_WARNING "ip_queue: OOM "
632                                       "in mangle, dropping packet\n");
633                                 return -ENOMEM;
634                         }
635                         if (e->skb->sk)
636                                 skb_set_owner_w(newskb, e->skb->sk);
637                         kfree_skb(e->skb);
638                         e->skb = newskb;
639                 }
640                 skb_put(e->skb, diff);
641         }
642         if (!skb_make_writable(&e->skb, data_len))
643                 return -ENOMEM;
644         memcpy(e->skb->data, data, data_len);
645         e->skb->ip_summed = CHECKSUM_NONE;
646         return 0;
647 }
648
649 static inline int
650 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
651 {
652         return (id == e->id);
653 }
654
655 static int
656 nfqnl_set_mode(struct nfqnl_instance *queue,
657                unsigned char mode, unsigned int range)
658 {
659         int status;
660
661         spin_lock_bh(&queue->lock);
662         status = __nfqnl_set_mode(queue, mode, range);
663         spin_unlock_bh(&queue->lock);
664
665         return status;
666 }
667
668 static int
669 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
670 {
671         if (entry->info->indev)
672                 if (entry->info->indev->ifindex == ifindex)
673                         return 1;
674                         
675         if (entry->info->outdev)
676                 if (entry->info->outdev->ifindex == ifindex)
677                         return 1;
678
679         return 0;
680 }
681
682 /* drop all packets with either indev or outdev == ifindex from all queue
683  * instances */
684 static void
685 nfqnl_dev_drop(int ifindex)
686 {
687         int i;
688         
689         QDEBUG("entering for ifindex %u\n", ifindex);
690
691         /* this only looks like we have to hold the readlock for a way too long
692          * time, issue_verdict(),  nf_reinject(), ... - but we always only
693          * issue NF_DROP, which is processed directly in nf_reinject() */
694         read_lock_bh(&instances_lock);
695
696         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
697                 struct hlist_node *tmp;
698                 struct nfqnl_instance *inst;
699                 struct hlist_head *head = &instance_table[i];
700
701                 hlist_for_each_entry(inst, tmp, head, hlist) {
702                         struct nfqnl_queue_entry *entry;
703                         while ((entry = find_dequeue_entry(inst, dev_cmp, 
704                                                            ifindex)) != NULL)
705                                 issue_verdict(entry, NF_DROP);
706                 }
707         }
708
709         read_unlock_bh(&instances_lock);
710 }
711
712 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
713
714 static int
715 nfqnl_rcv_dev_event(struct notifier_block *this,
716                     unsigned long event, void *ptr)
717 {
718         struct net_device *dev = ptr;
719
720         /* Drop any packets associated with the downed device */
721         if (event == NETDEV_DOWN)
722                 nfqnl_dev_drop(dev->ifindex);
723         return NOTIFY_DONE;
724 }
725
726 static struct notifier_block nfqnl_dev_notifier = {
727         .notifier_call  = nfqnl_rcv_dev_event,
728 };
729
730 static int
731 nfqnl_rcv_nl_event(struct notifier_block *this,
732                    unsigned long event, void *ptr)
733 {
734         struct netlink_notify *n = ptr;
735
736         if (event == NETLINK_URELEASE &&
737             n->protocol == NETLINK_NETFILTER && n->pid) {
738                 int i;
739
740                 /* destroy all instances for this pid */
741                 write_lock_bh(&instances_lock);
742                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
743                         struct hlist_node *tmp, *t2;
744                         struct nfqnl_instance *inst;
745                         struct hlist_head *head = &instance_table[i];
746
747                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
748                                 if (n->pid == inst->peer_pid)
749                                         __instance_destroy(inst);
750                         }
751                 }
752                 write_unlock_bh(&instances_lock);
753         }
754         return NOTIFY_DONE;
755 }
756
757 static struct notifier_block nfqnl_rtnl_notifier = {
758         .notifier_call  = nfqnl_rcv_nl_event,
759 };
760
761 static const int nfqa_verdict_min[NFQA_MAX] = {
762         [NFQA_VERDICT_HDR-1]    = sizeof(struct nfqnl_msg_verdict_hdr),
763         [NFQA_MARK-1]           = sizeof(u_int32_t),
764         [NFQA_PAYLOAD-1]        = 0,
765 };
766
767 static int
768 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
769                    struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
770 {
771         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
772         u_int16_t queue_num = ntohs(nfmsg->res_id);
773
774         struct nfqnl_msg_verdict_hdr *vhdr;
775         struct nfqnl_instance *queue;
776         unsigned int verdict;
777         struct nfqnl_queue_entry *entry;
778         int err;
779
780         if (nfattr_bad_size(nfqa, NFQA_MAX, nfqa_verdict_min)) {
781                 QDEBUG("bad attribute size\n");
782                 return -EINVAL;
783         }
784
785         queue = instance_lookup_get(queue_num);
786         if (!queue)
787                 return -ENODEV;
788
789         if (queue->peer_pid != NETLINK_CB(skb).pid) {
790                 err = -EPERM;
791                 goto err_out_put;
792         }
793
794         if (!nfqa[NFQA_VERDICT_HDR-1]) {
795                 err = -EINVAL;
796                 goto err_out_put;
797         }
798
799         vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
800         verdict = ntohl(vhdr->verdict);
801
802         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
803                 err = -EINVAL;
804                 goto err_out_put;
805         }
806
807         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
808         if (entry == NULL) {
809                 err = -ENOENT;
810                 goto err_out_put;
811         }
812
813         if (nfqa[NFQA_PAYLOAD-1]) {
814                 if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
815                                  NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
816                         verdict = NF_DROP;
817         }
818
819         if (nfqa[NFQA_MARK-1])
820                 skb->nfmark = ntohl(*(u_int32_t *)NFA_DATA(nfqa[NFQA_MARK-1]));
821                 
822         issue_verdict(entry, verdict);
823         instance_put(queue);
824         return 0;
825
826 err_out_put:
827         instance_put(queue);
828         return err;
829 }
830
831 static int
832 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
833                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
834 {
835         return -ENOTSUPP;
836 }
837
838 static const int nfqa_cfg_min[NFQA_CFG_MAX] = {
839         [NFQA_CFG_CMD-1]        = sizeof(struct nfqnl_msg_config_cmd),
840         [NFQA_CFG_PARAMS-1]     = sizeof(struct nfqnl_msg_config_params),
841 };
842
843 static struct nf_queue_handler nfqh = {
844         .name   = "nf_queue",
845         .outfn  = &nfqnl_enqueue_packet,
846 };
847
848 static int
849 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
850                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
851 {
852         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
853         u_int16_t queue_num = ntohs(nfmsg->res_id);
854         struct nfqnl_instance *queue;
855         int ret = 0;
856
857         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
858
859         if (nfattr_bad_size(nfqa, NFQA_CFG_MAX, nfqa_cfg_min)) {
860                 QDEBUG("bad attribute size\n");
861                 return -EINVAL;
862         }
863
864         queue = instance_lookup_get(queue_num);
865         if (nfqa[NFQA_CFG_CMD-1]) {
866                 struct nfqnl_msg_config_cmd *cmd;
867                 cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
868                 QDEBUG("found CFG_CMD\n");
869
870                 switch (cmd->command) {
871                 case NFQNL_CFG_CMD_BIND:
872                         if (queue)
873                                 return -EBUSY;
874
875                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
876                         if (!queue)
877                                 return -EINVAL;
878                         break;
879                 case NFQNL_CFG_CMD_UNBIND:
880                         if (!queue)
881                                 return -ENODEV;
882
883                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
884                                 ret = -EPERM;
885                                 goto out_put;
886                         }
887
888                         instance_destroy(queue);
889                         break;
890                 case NFQNL_CFG_CMD_PF_BIND:
891                         QDEBUG("registering queue handler for pf=%u\n",
892                                 ntohs(cmd->pf));
893                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
894                         break;
895                 case NFQNL_CFG_CMD_PF_UNBIND:
896                         QDEBUG("unregistering queue handler for pf=%u\n",
897                                 ntohs(cmd->pf));
898                         /* This is a bug and a feature.  We can unregister
899                          * other handlers(!) */
900                         ret = nf_unregister_queue_handler(ntohs(cmd->pf));
901                         break;
902                 default:
903                         ret = -EINVAL;
904                         break;
905                 }
906         } else {
907                 if (!queue) {
908                         QDEBUG("no config command, and no instance ENOENT\n");
909                         ret = -ENOENT;
910                         goto out_put;
911                 }
912
913                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
914                         QDEBUG("no config command, and wrong pid\n");
915                         ret = -EPERM;
916                         goto out_put;
917                 }
918         }
919
920         if (nfqa[NFQA_CFG_PARAMS-1]) {
921                 struct nfqnl_msg_config_params *params;
922                 params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
923
924                 nfqnl_set_mode(queue, params->copy_mode,
925                                 ntohl(params->copy_range));
926         }
927
928 out_put:
929         instance_put(queue);
930         return ret;
931 }
932
933 static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
934         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
935                                     .attr_count = NFQA_MAX,
936                                     .cap_required = CAP_NET_ADMIN },
937         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
938                                     .attr_count = NFQA_MAX,
939                                     .cap_required = CAP_NET_ADMIN },
940         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
941                                     .attr_count = NFQA_CFG_MAX,
942                                     .cap_required = CAP_NET_ADMIN },
943 };
944
945 static struct nfnetlink_subsystem nfqnl_subsys = {
946         .name           = "nf_queue",
947         .subsys_id      = NFNL_SUBSYS_QUEUE,
948         .cb_count       = NFQNL_MSG_MAX,
949         .cb             = nfqnl_cb,
950 };
951
952 #ifdef CONFIG_PROC_FS
953 struct iter_state {
954         unsigned int bucket;
955 };
956
957 static struct hlist_node *get_first(struct seq_file *seq)
958 {
959         struct iter_state *st = seq->private;
960
961         if (!st)
962                 return NULL;
963
964         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
965                 if (!hlist_empty(&instance_table[st->bucket]))
966                         return instance_table[st->bucket].first;
967         }
968         return NULL;
969 }
970
971 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
972 {
973         struct iter_state *st = seq->private;
974
975         h = h->next;
976         while (!h) {
977                 if (++st->bucket >= INSTANCE_BUCKETS)
978                         return NULL;
979
980                 h = instance_table[st->bucket].first;
981         }
982         return h;
983 }
984
985 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
986 {
987         struct hlist_node *head;
988         head = get_first(seq);
989
990         if (head)
991                 while (pos && (head = get_next(seq, head)))
992                         pos--;
993         return pos ? NULL : head;
994 }
995
996 static void *seq_start(struct seq_file *seq, loff_t *pos)
997 {
998         read_lock_bh(&instances_lock);
999         return get_idx(seq, *pos);
1000 }
1001
1002 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1003 {
1004         (*pos)++;
1005         return get_next(s, v);
1006 }
1007
1008 static void seq_stop(struct seq_file *s, void *v)
1009 {
1010         read_unlock_bh(&instances_lock);
1011 }
1012
1013 static int seq_show(struct seq_file *s, void *v)
1014 {
1015         const struct nfqnl_instance *inst = v;
1016
1017         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1018                           inst->queue_num,
1019                           inst->peer_pid, inst->queue_total,
1020                           inst->copy_mode, inst->copy_range,
1021                           inst->queue_dropped, inst->queue_user_dropped,
1022                           atomic_read(&inst->id_sequence),
1023                           atomic_read(&inst->use));
1024 }
1025
1026 static struct seq_operations nfqnl_seq_ops = {
1027         .start  = seq_start,
1028         .next   = seq_next,
1029         .stop   = seq_stop,
1030         .show   = seq_show,
1031 };
1032
1033 static int nfqnl_open(struct inode *inode, struct file *file)
1034 {
1035         struct seq_file *seq;
1036         struct iter_state *is;
1037         int ret;
1038
1039         is = kmalloc(sizeof(*is), GFP_KERNEL);
1040         if (!is)
1041                 return -ENOMEM;
1042         memset(is, 0, sizeof(*is));
1043         ret = seq_open(file, &nfqnl_seq_ops);
1044         if (ret < 0)
1045                 goto out_free;
1046         seq = file->private_data;
1047         seq->private = is;
1048         return ret;
1049 out_free:
1050         kfree(is);
1051         return ret;
1052 }
1053
1054 static struct file_operations nfqnl_file_ops = {
1055         .owner   = THIS_MODULE,
1056         .open    = nfqnl_open,
1057         .read    = seq_read,
1058         .llseek  = seq_lseek,
1059         .release = seq_release_private,
1060 };
1061
1062 #endif /* PROC_FS */
1063
1064 static int
1065 init_or_cleanup(int init)
1066 {
1067         int i, status = -ENOMEM;
1068 #ifdef CONFIG_PROC_FS
1069         struct proc_dir_entry *proc_nfqueue;
1070 #endif
1071         
1072         if (!init)
1073                 goto cleanup;
1074
1075         for (i = 0; i < INSTANCE_BUCKETS; i++)
1076                 INIT_HLIST_HEAD(&instance_table[i]);
1077
1078         netlink_register_notifier(&nfqnl_rtnl_notifier);
1079         status = nfnetlink_subsys_register(&nfqnl_subsys);
1080         if (status < 0) {
1081                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1082                 goto cleanup_netlink_notifier;
1083         }
1084
1085 #ifdef CONFIG_PROC_FS
1086         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1087                                          proc_net_netfilter);
1088         if (!proc_nfqueue)
1089                 goto cleanup_subsys;
1090         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1091 #endif
1092
1093         register_netdevice_notifier(&nfqnl_dev_notifier);
1094
1095         return status;
1096
1097 cleanup:
1098         nf_unregister_queue_handlers(&nfqh);
1099         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1100 #ifdef CONFIG_PROC_FS
1101         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1102 cleanup_subsys:
1103 #endif  
1104         nfnetlink_subsys_unregister(&nfqnl_subsys);
1105 cleanup_netlink_notifier:
1106         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1107         return status;
1108 }
1109
1110 static int __init init(void)
1111 {
1112         
1113         return init_or_cleanup(1);
1114 }
1115
1116 static void __exit fini(void)
1117 {
1118         init_or_cleanup(0);
1119 }
1120
1121 MODULE_DESCRIPTION("netfilter packet queue handler");
1122 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1123 MODULE_LICENSE("GPL");
1124 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1125
1126 module_init(init);
1127 module_exit(fini);