Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-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 nlattr which has 16bit nla_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         sk_buff_data_t 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         __be32 tmp_uint;
353
354         QDEBUG("entered\n");
355
356         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
357                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
358                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
359                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
360 #ifdef CONFIG_BRIDGE_NETFILTER
361                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
362                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
363 #endif
364                 + nla_total_size(sizeof(u_int32_t))     /* mark */
365                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
366                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
367
368         outdev = entinf->outdev;
369
370         spin_lock_bh(&queue->lock);
371
372         switch (queue->copy_mode) {
373         case NFQNL_COPY_META:
374         case NFQNL_COPY_NONE:
375                 data_len = 0;
376                 break;
377
378         case NFQNL_COPY_PACKET:
379                 if ((entskb->ip_summed == CHECKSUM_PARTIAL ||
380                      entskb->ip_summed == CHECKSUM_COMPLETE) &&
381                     (*errp = skb_checksum_help(entskb))) {
382                         spin_unlock_bh(&queue->lock);
383                         return NULL;
384                 }
385                 if (queue->copy_range == 0
386                     || queue->copy_range > entskb->len)
387                         data_len = entskb->len;
388                 else
389                         data_len = queue->copy_range;
390
391                 size += nla_total_size(data_len);
392                 break;
393
394         default:
395                 *errp = -EINVAL;
396                 spin_unlock_bh(&queue->lock);
397                 return NULL;
398         }
399
400         spin_unlock_bh(&queue->lock);
401
402         skb = alloc_skb(size, GFP_ATOMIC);
403         if (!skb)
404                 goto nlmsg_failure;
405
406         old_tail = skb->tail;
407         nlh = NLMSG_PUT(skb, 0, 0,
408                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
409                         sizeof(struct nfgenmsg));
410         nfmsg = NLMSG_DATA(nlh);
411         nfmsg->nfgen_family = entinf->pf;
412         nfmsg->version = NFNETLINK_V0;
413         nfmsg->res_id = htons(queue->queue_num);
414
415         pmsg.packet_id          = htonl(entry->id);
416         pmsg.hw_protocol        = entskb->protocol;
417         pmsg.hook               = entinf->hook;
418
419         NLA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
420
421         indev = entinf->indev;
422         if (indev) {
423                 tmp_uint = htonl(indev->ifindex);
424 #ifndef CONFIG_BRIDGE_NETFILTER
425                 NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
426 #else
427                 if (entinf->pf == PF_BRIDGE) {
428                         /* Case 1: indev is physical input device, we need to
429                          * look for bridge group (when called from
430                          * netfilter_bridge) */
431                         NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint),
432                                 &tmp_uint);
433                         /* this is the bridge group "brX" */
434                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
435                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
436                                 &tmp_uint);
437                 } else {
438                         /* Case 2: indev is bridge group, we need to look for
439                          * physical device (when called from ipv4) */
440                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
441                                 &tmp_uint);
442                         if (entskb->nf_bridge
443                             && entskb->nf_bridge->physindev) {
444                                 tmp_uint = htonl(entskb->nf_bridge->physindev->ifindex);
445                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
446                                         sizeof(tmp_uint), &tmp_uint);
447                         }
448                 }
449 #endif
450         }
451
452         if (outdev) {
453                 tmp_uint = htonl(outdev->ifindex);
454 #ifndef CONFIG_BRIDGE_NETFILTER
455                 NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
456 #else
457                 if (entinf->pf == PF_BRIDGE) {
458                         /* Case 1: outdev is physical output device, we need to
459                          * look for bridge group (when called from
460                          * netfilter_bridge) */
461                         NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
462                                 &tmp_uint);
463                         /* this is the bridge group "brX" */
464                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
465                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
466                                 &tmp_uint);
467                 } else {
468                         /* Case 2: outdev is bridge group, we need to look for
469                          * physical output device (when called from ipv4) */
470                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
471                                 &tmp_uint);
472                         if (entskb->nf_bridge
473                             && entskb->nf_bridge->physoutdev) {
474                                 tmp_uint = htonl(entskb->nf_bridge->physoutdev->ifindex);
475                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
476                                         sizeof(tmp_uint), &tmp_uint);
477                         }
478                 }
479 #endif
480         }
481
482         if (entskb->mark) {
483                 tmp_uint = htonl(entskb->mark);
484                 NLA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
485         }
486
487         if (indev && entskb->dev) {
488                 struct nfqnl_msg_packet_hw phw;
489                 int len = dev_parse_header(entskb, phw.hw_addr);
490                 if (len) {
491                         phw.hw_addrlen = htons(len);
492                         NLA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
493                 }
494         }
495
496         if (entskb->tstamp.tv64) {
497                 struct nfqnl_msg_packet_timestamp ts;
498                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
499                 ts.sec = cpu_to_be64(tv.tv_sec);
500                 ts.usec = cpu_to_be64(tv.tv_usec);
501
502                 NLA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
503         }
504
505         if (data_len) {
506                 struct nlattr *nla;
507                 int size = nla_attr_size(data_len);
508
509                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
510                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
511                         goto nlmsg_failure;
512                 }
513
514                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
515                 nla->nla_type = NFQA_PAYLOAD;
516                 nla->nla_len = size;
517
518                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
519                         BUG();
520         }
521
522         nlh->nlmsg_len = skb->tail - old_tail;
523         return skb;
524
525 nlmsg_failure:
526 nla_put_failure:
527         if (skb)
528                 kfree_skb(skb);
529         *errp = -EINVAL;
530         if (net_ratelimit())
531                 printk(KERN_ERR "nf_queue: error creating packet message\n");
532         return NULL;
533 }
534
535 static int
536 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info,
537                      unsigned int queuenum, void *data)
538 {
539         int status = -EINVAL;
540         struct sk_buff *nskb;
541         struct nfqnl_instance *queue;
542         struct nfqnl_queue_entry *entry;
543
544         QDEBUG("entered\n");
545
546         queue = instance_lookup_get(queuenum);
547         if (!queue) {
548                 QDEBUG("no queue instance matching\n");
549                 return -EINVAL;
550         }
551
552         if (queue->copy_mode == NFQNL_COPY_NONE) {
553                 QDEBUG("mode COPY_NONE, aborting\n");
554                 status = -EAGAIN;
555                 goto err_out_put;
556         }
557
558         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
559         if (entry == NULL) {
560                 if (net_ratelimit())
561                         printk(KERN_ERR
562                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
563                 status = -ENOMEM;
564                 goto err_out_put;
565         }
566
567         entry->info = info;
568         entry->skb = skb;
569         entry->id = atomic_inc_return(&queue->id_sequence);
570
571         nskb = nfqnl_build_packet_message(queue, entry, &status);
572         if (nskb == NULL)
573                 goto err_out_free;
574
575         spin_lock_bh(&queue->lock);
576
577         if (!queue->peer_pid)
578                 goto err_out_free_nskb;
579
580         if (queue->queue_total >= queue->queue_maxlen) {
581                 queue->queue_dropped++;
582                 status = -ENOSPC;
583                 if (net_ratelimit())
584                           printk(KERN_WARNING "nf_queue: full at %d entries, "
585                                  "dropping packets(s). Dropped: %d\n",
586                                  queue->queue_total, queue->queue_dropped);
587                 goto err_out_free_nskb;
588         }
589
590         /* nfnetlink_unicast will either free the nskb or add it to a socket */
591         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
592         if (status < 0) {
593                 queue->queue_user_dropped++;
594                 goto err_out_unlock;
595         }
596
597         __enqueue_entry(queue, entry);
598
599         spin_unlock_bh(&queue->lock);
600         instance_put(queue);
601         return status;
602
603 err_out_free_nskb:
604         kfree_skb(nskb);
605
606 err_out_unlock:
607         spin_unlock_bh(&queue->lock);
608
609 err_out_free:
610         kfree(entry);
611 err_out_put:
612         instance_put(queue);
613         return status;
614 }
615
616 static int
617 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
618 {
619         int diff;
620         int err;
621
622         diff = data_len - e->skb->len;
623         if (diff < 0) {
624                 if (pskb_trim(e->skb, data_len))
625                         return -ENOMEM;
626         } else if (diff > 0) {
627                 if (data_len > 0xFFFF)
628                         return -EINVAL;
629                 if (diff > skb_tailroom(e->skb)) {
630                         err = pskb_expand_head(e->skb, 0,
631                                                diff - skb_tailroom(e->skb),
632                                                GFP_ATOMIC);
633                         if (err) {
634                                 printk(KERN_WARNING "nf_queue: OOM "
635                                       "in mangle, dropping packet\n");
636                                 return err;
637                         }
638                 }
639                 skb_put(e->skb, diff);
640         }
641         if (!skb_make_writable(e->skb, data_len))
642                 return -ENOMEM;
643         skb_copy_to_linear_data(e->skb, 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         struct nf_info *entinf = entry->info;
671
672         if (entinf->indev)
673                 if (entinf->indev->ifindex == ifindex)
674                         return 1;
675         if (entinf->outdev)
676                 if (entinf->outdev->ifindex == ifindex)
677                         return 1;
678 #ifdef CONFIG_BRIDGE_NETFILTER
679         if (entry->skb->nf_bridge) {
680                 if (entry->skb->nf_bridge->physindev &&
681                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
682                         return 1;
683                 if (entry->skb->nf_bridge->physoutdev &&
684                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
685                         return 1;
686         }
687 #endif
688         return 0;
689 }
690
691 /* drop all packets with either indev or outdev == ifindex from all queue
692  * instances */
693 static void
694 nfqnl_dev_drop(int ifindex)
695 {
696         int i;
697
698         QDEBUG("entering for ifindex %u\n", ifindex);
699
700         /* this only looks like we have to hold the readlock for a way too long
701          * time, issue_verdict(),  nf_reinject(), ... - but we always only
702          * issue NF_DROP, which is processed directly in nf_reinject() */
703         read_lock_bh(&instances_lock);
704
705         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
706                 struct hlist_node *tmp;
707                 struct nfqnl_instance *inst;
708                 struct hlist_head *head = &instance_table[i];
709
710                 hlist_for_each_entry(inst, tmp, head, hlist) {
711                         struct nfqnl_queue_entry *entry;
712                         while ((entry = find_dequeue_entry(inst, dev_cmp,
713                                                            ifindex)) != NULL)
714                                 issue_verdict(entry, NF_DROP);
715                 }
716         }
717
718         read_unlock_bh(&instances_lock);
719 }
720
721 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
722
723 static int
724 nfqnl_rcv_dev_event(struct notifier_block *this,
725                     unsigned long event, void *ptr)
726 {
727         struct net_device *dev = ptr;
728
729         if (dev->nd_net != &init_net)
730                 return NOTIFY_DONE;
731
732         /* Drop any packets associated with the downed device */
733         if (event == NETDEV_DOWN)
734                 nfqnl_dev_drop(dev->ifindex);
735         return NOTIFY_DONE;
736 }
737
738 static struct notifier_block nfqnl_dev_notifier = {
739         .notifier_call  = nfqnl_rcv_dev_event,
740 };
741
742 static int
743 nfqnl_rcv_nl_event(struct notifier_block *this,
744                    unsigned long event, void *ptr)
745 {
746         struct netlink_notify *n = ptr;
747
748         if (event == NETLINK_URELEASE &&
749             n->protocol == NETLINK_NETFILTER && n->pid) {
750                 int i;
751
752                 /* destroy all instances for this pid */
753                 write_lock_bh(&instances_lock);
754                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
755                         struct hlist_node *tmp, *t2;
756                         struct nfqnl_instance *inst;
757                         struct hlist_head *head = &instance_table[i];
758
759                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
760                                 if ((n->net == &init_net) &&
761                                     (n->pid == inst->peer_pid))
762                                         __instance_destroy(inst);
763                         }
764                 }
765                 write_unlock_bh(&instances_lock);
766         }
767         return NOTIFY_DONE;
768 }
769
770 static struct notifier_block nfqnl_rtnl_notifier = {
771         .notifier_call  = nfqnl_rcv_nl_event,
772 };
773
774 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
775         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
776         [NFQA_MARK]             = { .type = NLA_U32 },
777         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
778 };
779
780 static int
781 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
782                    struct nlmsghdr *nlh, struct nlattr *nfqa[])
783 {
784         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
785         u_int16_t queue_num = ntohs(nfmsg->res_id);
786
787         struct nfqnl_msg_verdict_hdr *vhdr;
788         struct nfqnl_instance *queue;
789         unsigned int verdict;
790         struct nfqnl_queue_entry *entry;
791         int err;
792
793         queue = instance_lookup_get(queue_num);
794         if (!queue)
795                 return -ENODEV;
796
797         if (queue->peer_pid != NETLINK_CB(skb).pid) {
798                 err = -EPERM;
799                 goto err_out_put;
800         }
801
802         if (!nfqa[NFQA_VERDICT_HDR]) {
803                 err = -EINVAL;
804                 goto err_out_put;
805         }
806
807         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
808         verdict = ntohl(vhdr->verdict);
809
810         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
811                 err = -EINVAL;
812                 goto err_out_put;
813         }
814
815         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
816         if (entry == NULL) {
817                 err = -ENOENT;
818                 goto err_out_put;
819         }
820
821         if (nfqa[NFQA_PAYLOAD]) {
822                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
823                                  nla_len(nfqa[NFQA_PAYLOAD]), entry) < 0)
824                         verdict = NF_DROP;
825         }
826
827         if (nfqa[NFQA_MARK])
828                 entry->skb->mark = ntohl(*(__be32 *)
829                                          nla_data(nfqa[NFQA_MARK]));
830
831         issue_verdict(entry, verdict);
832         instance_put(queue);
833         return 0;
834
835 err_out_put:
836         instance_put(queue);
837         return err;
838 }
839
840 static int
841 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
842                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
843 {
844         return -ENOTSUPP;
845 }
846
847 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
848         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
849         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
850 };
851
852 static struct nf_queue_handler nfqh = {
853         .name   = "nf_queue",
854         .outfn  = &nfqnl_enqueue_packet,
855 };
856
857 static int
858 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
859                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
860 {
861         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
862         u_int16_t queue_num = ntohs(nfmsg->res_id);
863         struct nfqnl_instance *queue;
864         int ret = 0;
865
866         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
867
868         queue = instance_lookup_get(queue_num);
869         if (nfqa[NFQA_CFG_CMD]) {
870                 struct nfqnl_msg_config_cmd *cmd;
871                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
872                 QDEBUG("found CFG_CMD\n");
873
874                 switch (cmd->command) {
875                 case NFQNL_CFG_CMD_BIND:
876                         if (queue)
877                                 return -EBUSY;
878
879                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
880                         if (!queue)
881                                 return -EINVAL;
882                         break;
883                 case NFQNL_CFG_CMD_UNBIND:
884                         if (!queue)
885                                 return -ENODEV;
886
887                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
888                                 ret = -EPERM;
889                                 goto out_put;
890                         }
891
892                         instance_destroy(queue);
893                         break;
894                 case NFQNL_CFG_CMD_PF_BIND:
895                         QDEBUG("registering queue handler for pf=%u\n",
896                                 ntohs(cmd->pf));
897                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
898                         break;
899                 case NFQNL_CFG_CMD_PF_UNBIND:
900                         QDEBUG("unregistering queue handler for pf=%u\n",
901                                 ntohs(cmd->pf));
902                         ret = nf_unregister_queue_handler(ntohs(cmd->pf), &nfqh);
903                         break;
904                 default:
905                         ret = -EINVAL;
906                         break;
907                 }
908         } else {
909                 if (!queue) {
910                         QDEBUG("no config command, and no instance ENOENT\n");
911                         ret = -ENOENT;
912                         goto out_put;
913                 }
914
915                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
916                         QDEBUG("no config command, and wrong pid\n");
917                         ret = -EPERM;
918                         goto out_put;
919                 }
920         }
921
922         if (nfqa[NFQA_CFG_PARAMS]) {
923                 struct nfqnl_msg_config_params *params;
924
925                 if (!queue) {
926                         ret = -ENOENT;
927                         goto out_put;
928                 }
929                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
930                 nfqnl_set_mode(queue, params->copy_mode,
931                                 ntohl(params->copy_range));
932         }
933
934         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
935                 __be32 *queue_maxlen;
936                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
937                 spin_lock_bh(&queue->lock);
938                 queue->queue_maxlen = ntohl(*queue_maxlen);
939                 spin_unlock_bh(&queue->lock);
940         }
941
942 out_put:
943         instance_put(queue);
944         return ret;
945 }
946
947 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
948         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
949                                     .attr_count = NFQA_MAX, },
950         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
951                                     .attr_count = NFQA_MAX,
952                                     .policy = nfqa_verdict_policy },
953         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
954                                     .attr_count = NFQA_CFG_MAX,
955                                     .policy = nfqa_cfg_policy },
956 };
957
958 static const struct nfnetlink_subsystem nfqnl_subsys = {
959         .name           = "nf_queue",
960         .subsys_id      = NFNL_SUBSYS_QUEUE,
961         .cb_count       = NFQNL_MSG_MAX,
962         .cb             = nfqnl_cb,
963 };
964
965 #ifdef CONFIG_PROC_FS
966 struct iter_state {
967         unsigned int bucket;
968 };
969
970 static struct hlist_node *get_first(struct seq_file *seq)
971 {
972         struct iter_state *st = seq->private;
973
974         if (!st)
975                 return NULL;
976
977         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
978                 if (!hlist_empty(&instance_table[st->bucket]))
979                         return instance_table[st->bucket].first;
980         }
981         return NULL;
982 }
983
984 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
985 {
986         struct iter_state *st = seq->private;
987
988         h = h->next;
989         while (!h) {
990                 if (++st->bucket >= INSTANCE_BUCKETS)
991                         return NULL;
992
993                 h = instance_table[st->bucket].first;
994         }
995         return h;
996 }
997
998 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
999 {
1000         struct hlist_node *head;
1001         head = get_first(seq);
1002
1003         if (head)
1004                 while (pos && (head = get_next(seq, head)))
1005                         pos--;
1006         return pos ? NULL : head;
1007 }
1008
1009 static void *seq_start(struct seq_file *seq, loff_t *pos)
1010 {
1011         read_lock_bh(&instances_lock);
1012         return get_idx(seq, *pos);
1013 }
1014
1015 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1016 {
1017         (*pos)++;
1018         return get_next(s, v);
1019 }
1020
1021 static void seq_stop(struct seq_file *s, void *v)
1022 {
1023         read_unlock_bh(&instances_lock);
1024 }
1025
1026 static int seq_show(struct seq_file *s, void *v)
1027 {
1028         const struct nfqnl_instance *inst = v;
1029
1030         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1031                           inst->queue_num,
1032                           inst->peer_pid, inst->queue_total,
1033                           inst->copy_mode, inst->copy_range,
1034                           inst->queue_dropped, inst->queue_user_dropped,
1035                           atomic_read(&inst->id_sequence),
1036                           atomic_read(&inst->use));
1037 }
1038
1039 static const struct seq_operations nfqnl_seq_ops = {
1040         .start  = seq_start,
1041         .next   = seq_next,
1042         .stop   = seq_stop,
1043         .show   = seq_show,
1044 };
1045
1046 static int nfqnl_open(struct inode *inode, struct file *file)
1047 {
1048         return seq_open_private(file, &nfqnl_seq_ops,
1049                         sizeof(struct iter_state));
1050 }
1051
1052 static const 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 __init nfnetlink_queue_init(void)
1063 {
1064         int i, status = -ENOMEM;
1065 #ifdef CONFIG_PROC_FS
1066         struct proc_dir_entry *proc_nfqueue;
1067 #endif
1068
1069         for (i = 0; i < INSTANCE_BUCKETS; i++)
1070                 INIT_HLIST_HEAD(&instance_table[i]);
1071
1072         netlink_register_notifier(&nfqnl_rtnl_notifier);
1073         status = nfnetlink_subsys_register(&nfqnl_subsys);
1074         if (status < 0) {
1075                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1076                 goto cleanup_netlink_notifier;
1077         }
1078
1079 #ifdef CONFIG_PROC_FS
1080         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1081                                          proc_net_netfilter);
1082         if (!proc_nfqueue)
1083                 goto cleanup_subsys;
1084         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1085 #endif
1086
1087         register_netdevice_notifier(&nfqnl_dev_notifier);
1088         return status;
1089
1090 #ifdef CONFIG_PROC_FS
1091 cleanup_subsys:
1092         nfnetlink_subsys_unregister(&nfqnl_subsys);
1093 #endif
1094 cleanup_netlink_notifier:
1095         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1096         return status;
1097 }
1098
1099 static void __exit nfnetlink_queue_fini(void)
1100 {
1101         nf_unregister_queue_handlers(&nfqh);
1102         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1103 #ifdef CONFIG_PROC_FS
1104         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1105 #endif
1106         nfnetlink_subsys_unregister(&nfqnl_subsys);
1107         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1108 }
1109
1110 MODULE_DESCRIPTION("netfilter packet queue handler");
1111 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1112 MODULE_LICENSE("GPL");
1113 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1114
1115 module_init(nfnetlink_queue_init);
1116 module_exit(nfnetlink_queue_fini);