Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / net / netfilter / nfnetlink_log.c
1 /*
2  * This is a module which is used for logging packets to userspace via
3  * nfetlink.
4  *
5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
6  *
7  * Based on the old ipv4-only ipt_ULOG.c:
8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License version 2 as
12  * published by the Free Software Foundation.
13  */
14 #include <linux/module.h>
15 #include <linux/skbuff.h>
16 #include <linux/init.h>
17 #include <linux/ip.h>
18 #include <linux/ipv6.h>
19 #include <linux/netdevice.h>
20 #include <linux/netfilter.h>
21 #include <linux/netlink.h>
22 #include <linux/netfilter/nfnetlink.h>
23 #include <linux/netfilter/nfnetlink_log.h>
24 #include <linux/spinlock.h>
25 #include <linux/sysctl.h>
26 #include <linux/proc_fs.h>
27 #include <linux/security.h>
28 #include <linux/list.h>
29 #include <linux/jhash.h>
30 #include <linux/random.h>
31 #include <net/sock.h>
32 #include <net/netfilter/nf_log.h>
33
34 #include <asm/atomic.h>
35
36 #ifdef CONFIG_BRIDGE_NETFILTER
37 #include "../bridge/br_private.h"
38 #endif
39
40 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
41 #define NFULNL_TIMEOUT_DEFAULT  HZ      /* every second */
42 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
43 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
44
45 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
46                                      printk(x, ## args); } while (0);
47
48 struct nfulnl_instance {
49         struct hlist_node hlist;        /* global list of instances */
50         spinlock_t lock;
51         atomic_t use;                   /* use count */
52
53         unsigned int qlen;              /* number of nlmsgs in skb */
54         struct sk_buff *skb;            /* pre-allocatd skb */
55         struct timer_list timer;
56         int peer_pid;                   /* PID of the peer process */
57
58         /* configurable parameters */
59         unsigned int flushtimeout;      /* timeout until queue flush */
60         unsigned int nlbufsiz;          /* netlink buffer allocation size */
61         unsigned int qthreshold;        /* threshold of the queue */
62         u_int32_t copy_range;
63         u_int32_t seq;                  /* instance-local sequential counter */
64         u_int16_t group_num;            /* number of this queue */
65         u_int16_t flags;
66         u_int8_t copy_mode;
67 };
68
69 static DEFINE_RWLOCK(instances_lock);
70 static atomic_t global_seq;
71
72 #define INSTANCE_BUCKETS        16
73 static struct hlist_head instance_table[INSTANCE_BUCKETS];
74 static unsigned int hash_init;
75
76 static inline u_int8_t instance_hashfn(u_int16_t group_num)
77 {
78         return ((group_num & 0xff) % INSTANCE_BUCKETS);
79 }
80
81 static struct nfulnl_instance *
82 __instance_lookup(u_int16_t group_num)
83 {
84         struct hlist_head *head;
85         struct hlist_node *pos;
86         struct nfulnl_instance *inst;
87
88         head = &instance_table[instance_hashfn(group_num)];
89         hlist_for_each_entry(inst, pos, head, hlist) {
90                 if (inst->group_num == group_num)
91                         return inst;
92         }
93         return NULL;
94 }
95
96 static inline void
97 instance_get(struct nfulnl_instance *inst)
98 {
99         atomic_inc(&inst->use);
100 }
101
102 static struct nfulnl_instance *
103 instance_lookup_get(u_int16_t group_num)
104 {
105         struct nfulnl_instance *inst;
106
107         read_lock_bh(&instances_lock);
108         inst = __instance_lookup(group_num);
109         if (inst)
110                 instance_get(inst);
111         read_unlock_bh(&instances_lock);
112
113         return inst;
114 }
115
116 static void
117 instance_put(struct nfulnl_instance *inst)
118 {
119         if (inst && atomic_dec_and_test(&inst->use)) {
120                 kfree(inst);
121                 module_put(THIS_MODULE);
122         }
123 }
124
125 static void nfulnl_timer(unsigned long data);
126
127 static struct nfulnl_instance *
128 instance_create(u_int16_t group_num, int pid)
129 {
130         struct nfulnl_instance *inst;
131         int err;
132
133         write_lock_bh(&instances_lock);
134         if (__instance_lookup(group_num)) {
135                 err = -EEXIST;
136                 goto out_unlock;
137         }
138
139         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
140         if (!inst) {
141                 err = -ENOMEM;
142                 goto out_unlock;
143         }
144
145         if (!try_module_get(THIS_MODULE)) {
146                 kfree(inst);
147                 err = -EAGAIN;
148                 goto out_unlock;
149         }
150
151         INIT_HLIST_NODE(&inst->hlist);
152         spin_lock_init(&inst->lock);
153         /* needs to be two, since we _put() after creation */
154         atomic_set(&inst->use, 2);
155
156         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
157
158         inst->peer_pid = pid;
159         inst->group_num = group_num;
160
161         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
162         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
163         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
164         inst->copy_mode         = NFULNL_COPY_PACKET;
165         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
166
167         hlist_add_head(&inst->hlist,
168                        &instance_table[instance_hashfn(group_num)]);
169
170         write_unlock_bh(&instances_lock);
171
172         return inst;
173
174 out_unlock:
175         write_unlock_bh(&instances_lock);
176         return ERR_PTR(err);
177 }
178
179 static void __nfulnl_flush(struct nfulnl_instance *inst);
180
181 static void
182 __instance_destroy(struct nfulnl_instance *inst)
183 {
184         /* first pull it out of the global list */
185         hlist_del(&inst->hlist);
186
187         /* then flush all pending packets from skb */
188
189         spin_lock_bh(&inst->lock);
190         if (inst->skb)
191                 __nfulnl_flush(inst);
192         spin_unlock_bh(&inst->lock);
193
194         /* and finally put the refcount */
195         instance_put(inst);
196 }
197
198 static inline void
199 instance_destroy(struct nfulnl_instance *inst)
200 {
201         write_lock_bh(&instances_lock);
202         __instance_destroy(inst);
203         write_unlock_bh(&instances_lock);
204 }
205
206 static int
207 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
208                   unsigned int range)
209 {
210         int status = 0;
211
212         spin_lock_bh(&inst->lock);
213
214         switch (mode) {
215         case NFULNL_COPY_NONE:
216         case NFULNL_COPY_META:
217                 inst->copy_mode = mode;
218                 inst->copy_range = 0;
219                 break;
220
221         case NFULNL_COPY_PACKET:
222                 inst->copy_mode = mode;
223                 inst->copy_range = min_t(unsigned int,
224                                          range, NFULNL_COPY_RANGE_MAX);
225                 break;
226
227         default:
228                 status = -EINVAL;
229                 break;
230         }
231
232         spin_unlock_bh(&inst->lock);
233
234         return status;
235 }
236
237 static int
238 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
239 {
240         int status;
241
242         spin_lock_bh(&inst->lock);
243         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
244                 status = -ERANGE;
245         else if (nlbufsiz > 131072)
246                 status = -ERANGE;
247         else {
248                 inst->nlbufsiz = nlbufsiz;
249                 status = 0;
250         }
251         spin_unlock_bh(&inst->lock);
252
253         return status;
254 }
255
256 static int
257 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
258 {
259         spin_lock_bh(&inst->lock);
260         inst->flushtimeout = timeout;
261         spin_unlock_bh(&inst->lock);
262
263         return 0;
264 }
265
266 static int
267 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
268 {
269         spin_lock_bh(&inst->lock);
270         inst->qthreshold = qthresh;
271         spin_unlock_bh(&inst->lock);
272
273         return 0;
274 }
275
276 static int
277 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
278 {
279         spin_lock_bh(&inst->lock);
280         inst->flags = flags;
281         spin_unlock_bh(&inst->lock);
282
283         return 0;
284 }
285
286 static struct sk_buff *
287 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
288 {
289         struct sk_buff *skb;
290         unsigned int n;
291
292         /* alloc skb which should be big enough for a whole multipart
293          * message.  WARNING: has to be <= 128k due to slab restrictions */
294
295         n = max(inst_size, pkt_size);
296         skb = alloc_skb(n, GFP_ATOMIC);
297         if (!skb) {
298                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
299                         inst_size);
300
301                 if (n > pkt_size) {
302                         /* try to allocate only as much as we need for current
303                          * packet */
304
305                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
306                         if (!skb)
307                                 PRINTR("nfnetlink_log: can't even alloc %u "
308                                        "bytes\n", pkt_size);
309                 }
310         }
311
312         return skb;
313 }
314
315 static int
316 __nfulnl_send(struct nfulnl_instance *inst)
317 {
318         int status = -1;
319
320         if (inst->qlen > 1)
321                 NLMSG_PUT(inst->skb, 0, 0,
322                           NLMSG_DONE,
323                           sizeof(struct nfgenmsg));
324
325         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
326
327         inst->qlen = 0;
328         inst->skb = NULL;
329
330 nlmsg_failure:
331         return status;
332 }
333
334 static void
335 __nfulnl_flush(struct nfulnl_instance *inst)
336 {
337         /* timer holds a reference */
338         if (del_timer(&inst->timer))
339                 instance_put(inst);
340         if (inst->skb)
341                 __nfulnl_send(inst);
342 }
343
344 static void
345 nfulnl_timer(unsigned long data)
346 {
347         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
348
349         spin_lock_bh(&inst->lock);
350         if (inst->skb)
351                 __nfulnl_send(inst);
352         spin_unlock_bh(&inst->lock);
353         instance_put(inst);
354 }
355
356 /* This is an inline function, we don't really care about a long
357  * list of arguments */
358 static inline int
359 __build_packet_message(struct nfulnl_instance *inst,
360                         const struct sk_buff *skb,
361                         unsigned int data_len,
362                         unsigned int pf,
363                         unsigned int hooknum,
364                         const struct net_device *indev,
365                         const struct net_device *outdev,
366                         const struct nf_loginfo *li,
367                         const char *prefix, unsigned int plen)
368 {
369         struct nfulnl_msg_packet_hdr pmsg;
370         struct nlmsghdr *nlh;
371         struct nfgenmsg *nfmsg;
372         __be32 tmp_uint;
373         sk_buff_data_t old_tail = inst->skb->tail;
374
375         nlh = NLMSG_PUT(inst->skb, 0, 0,
376                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
377                         sizeof(struct nfgenmsg));
378         nfmsg = NLMSG_DATA(nlh);
379         nfmsg->nfgen_family = pf;
380         nfmsg->version = NFNETLINK_V0;
381         nfmsg->res_id = htons(inst->group_num);
382
383         pmsg.hw_protocol        = skb->protocol;
384         pmsg.hook               = hooknum;
385
386         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
387
388         if (prefix)
389                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
390
391         if (indev) {
392 #ifndef CONFIG_BRIDGE_NETFILTER
393                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
394                              htonl(indev->ifindex));
395 #else
396                 if (pf == PF_BRIDGE) {
397                         /* Case 1: outdev is physical input device, we need to
398                          * look for bridge group (when called from
399                          * netfilter_bridge) */
400                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
401                                      htonl(indev->ifindex));
402                         /* this is the bridge group "brX" */
403                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
404                                      htonl(indev->br_port->br->dev->ifindex));
405                 } else {
406                         /* Case 2: indev is bridge group, we need to look for
407                          * physical device (when called from ipv4) */
408                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
409                                      htonl(indev->ifindex));
410                         if (skb->nf_bridge && skb->nf_bridge->physindev)
411                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
412                                              htonl(skb->nf_bridge->physindev->ifindex));
413                 }
414 #endif
415         }
416
417         if (outdev) {
418                 tmp_uint = htonl(outdev->ifindex);
419 #ifndef CONFIG_BRIDGE_NETFILTER
420                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
421                              htonl(outdev->ifindex));
422 #else
423                 if (pf == PF_BRIDGE) {
424                         /* Case 1: outdev is physical output device, we need to
425                          * look for bridge group (when called from
426                          * netfilter_bridge) */
427                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
428                                      htonl(outdev->ifindex));
429                         /* this is the bridge group "brX" */
430                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
431                                      htonl(outdev->br_port->br->dev->ifindex));
432                 } else {
433                         /* Case 2: indev is a bridge group, we need to look
434                          * for physical device (when called from ipv4) */
435                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
436                                      htonl(outdev->ifindex));
437                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
438                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
439                                              htonl(skb->nf_bridge->physoutdev->ifindex));
440                 }
441 #endif
442         }
443
444         if (skb->mark)
445                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
446
447         if (indev && skb->dev) {
448                 struct nfulnl_msg_packet_hw phw;
449                 int len = dev_parse_header(skb, phw.hw_addr);
450                 if (len > 0) {
451                         phw.hw_addrlen = htons(len);
452                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
453                 }
454         }
455
456         if (skb->tstamp.tv64) {
457                 struct nfulnl_msg_packet_timestamp ts;
458                 struct timeval tv = ktime_to_timeval(skb->tstamp);
459                 ts.sec = cpu_to_be64(tv.tv_sec);
460                 ts.usec = cpu_to_be64(tv.tv_usec);
461
462                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
463         }
464
465         /* UID */
466         if (skb->sk) {
467                 read_lock_bh(&skb->sk->sk_callback_lock);
468                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
469                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
470                         __be32 gid = htonl(skb->sk->sk_socket->file->f_gid);
471                         /* need to unlock here since NLA_PUT may goto */
472                         read_unlock_bh(&skb->sk->sk_callback_lock);
473                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
474                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
475                 } else
476                         read_unlock_bh(&skb->sk->sk_callback_lock);
477         }
478
479         /* local sequence number */
480         if (inst->flags & NFULNL_CFG_F_SEQ)
481                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
482
483         /* global sequence number */
484         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
485                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
486                              htonl(atomic_inc_return(&global_seq)));
487
488         if (data_len) {
489                 struct nlattr *nla;
490                 int size = nla_attr_size(data_len);
491
492                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
493                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
494                         goto nlmsg_failure;
495                 }
496
497                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
498                 nla->nla_type = NFULA_PAYLOAD;
499                 nla->nla_len = size;
500
501                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
502                         BUG();
503         }
504
505         nlh->nlmsg_len = inst->skb->tail - old_tail;
506         return 0;
507
508 nlmsg_failure:
509 nla_put_failure:
510         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
511         return -1;
512 }
513
514 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
515
516 static struct nf_loginfo default_loginfo = {
517         .type =         NF_LOG_TYPE_ULOG,
518         .u = {
519                 .ulog = {
520                         .copy_len       = 0xffff,
521                         .group          = 0,
522                         .qthreshold     = 1,
523                 },
524         },
525 };
526
527 /* log handler for internal netfilter logging api */
528 static void
529 nfulnl_log_packet(unsigned int pf,
530                   unsigned int hooknum,
531                   const struct sk_buff *skb,
532                   const struct net_device *in,
533                   const struct net_device *out,
534                   const struct nf_loginfo *li_user,
535                   const char *prefix)
536 {
537         unsigned int size, data_len;
538         struct nfulnl_instance *inst;
539         const struct nf_loginfo *li;
540         unsigned int qthreshold;
541         unsigned int plen;
542
543         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
544                 li = li_user;
545         else
546                 li = &default_loginfo;
547
548         inst = instance_lookup_get(li->u.ulog.group);
549         if (!inst)
550                 return;
551
552         plen = 0;
553         if (prefix)
554                 plen = strlen(prefix) + 1;
555
556         /* FIXME: do we want to make the size calculation conditional based on
557          * what is actually present?  way more branches and checks, but more
558          * memory efficient... */
559         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
560                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
561                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
562                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
563 #ifdef CONFIG_BRIDGE_NETFILTER
564                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
565                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
566 #endif
567                 + nla_total_size(sizeof(u_int32_t))     /* mark */
568                 + nla_total_size(sizeof(u_int32_t))     /* uid */
569                 + nla_total_size(sizeof(u_int32_t))     /* gid */
570                 + nla_total_size(plen)                  /* prefix */
571                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
572                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
573
574         spin_lock_bh(&inst->lock);
575
576         if (inst->flags & NFULNL_CFG_F_SEQ)
577                 size += nla_total_size(sizeof(u_int32_t));
578         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
579                 size += nla_total_size(sizeof(u_int32_t));
580
581         qthreshold = inst->qthreshold;
582         /* per-rule qthreshold overrides per-instance */
583         if (qthreshold > li->u.ulog.qthreshold)
584                 qthreshold = li->u.ulog.qthreshold;
585
586         switch (inst->copy_mode) {
587         case NFULNL_COPY_META:
588         case NFULNL_COPY_NONE:
589                 data_len = 0;
590                 break;
591
592         case NFULNL_COPY_PACKET:
593                 if (inst->copy_range == 0
594                     || inst->copy_range > skb->len)
595                         data_len = skb->len;
596                 else
597                         data_len = inst->copy_range;
598
599                 size += nla_total_size(data_len);
600                 break;
601
602         default:
603                 goto unlock_and_release;
604         }
605
606         if (inst->skb &&
607             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
608                 /* either the queue len is too high or we don't have
609                  * enough room in the skb left. flush to userspace. */
610                 __nfulnl_flush(inst);
611         }
612
613         if (!inst->skb) {
614                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
615                 if (!inst->skb)
616                         goto alloc_failure;
617         }
618
619         inst->qlen++;
620
621         __build_packet_message(inst, skb, data_len, pf,
622                                 hooknum, in, out, li, prefix, plen);
623
624         if (inst->qlen >= qthreshold)
625                 __nfulnl_flush(inst);
626         /* timer_pending always called within inst->lock, so there
627          * is no chance of a race here */
628         else if (!timer_pending(&inst->timer)) {
629                 instance_get(inst);
630                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
631                 add_timer(&inst->timer);
632         }
633
634 unlock_and_release:
635         spin_unlock_bh(&inst->lock);
636         instance_put(inst);
637         return;
638
639 alloc_failure:
640         /* FIXME: statistics */
641         goto unlock_and_release;
642 }
643
644 static int
645 nfulnl_rcv_nl_event(struct notifier_block *this,
646                    unsigned long event, void *ptr)
647 {
648         struct netlink_notify *n = ptr;
649
650         if (event == NETLINK_URELEASE &&
651             n->protocol == NETLINK_NETFILTER && n->pid) {
652                 int i;
653
654                 /* destroy all instances for this pid */
655                 write_lock_bh(&instances_lock);
656                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
657                         struct hlist_node *tmp, *t2;
658                         struct nfulnl_instance *inst;
659                         struct hlist_head *head = &instance_table[i];
660
661                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
662                                 if ((n->net == &init_net) &&
663                                     (n->pid == inst->peer_pid))
664                                         __instance_destroy(inst);
665                         }
666                 }
667                 write_unlock_bh(&instances_lock);
668         }
669         return NOTIFY_DONE;
670 }
671
672 static struct notifier_block nfulnl_rtnl_notifier = {
673         .notifier_call  = nfulnl_rcv_nl_event,
674 };
675
676 static int
677 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
678                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
679 {
680         return -ENOTSUPP;
681 }
682
683 static const struct nf_logger nfulnl_logger = {
684         .name   = "nfnetlink_log",
685         .logfn  = &nfulnl_log_packet,
686         .me     = THIS_MODULE,
687 };
688
689 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
690         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
691         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
692         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
693         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
694         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
695         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
696 };
697
698 static int
699 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
700                    struct nlmsghdr *nlh, struct nlattr *nfula[])
701 {
702         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
703         u_int16_t group_num = ntohs(nfmsg->res_id);
704         struct nfulnl_instance *inst;
705         struct nfulnl_msg_config_cmd *cmd = NULL;
706         int ret = 0;
707
708         if (nfula[NFULA_CFG_CMD]) {
709                 u_int8_t pf = nfmsg->nfgen_family;
710                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
711
712                 /* Commands without queue context */
713                 switch (cmd->command) {
714                 case NFULNL_CFG_CMD_PF_BIND:
715                         return nf_log_register(pf, &nfulnl_logger);
716                 case NFULNL_CFG_CMD_PF_UNBIND:
717                         nf_log_unregister_pf(pf);
718                         return 0;
719                 }
720         }
721
722         inst = instance_lookup_get(group_num);
723         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
724                 ret = -EPERM;
725                 goto out_put;
726         }
727
728         if (cmd != NULL) {
729                 switch (cmd->command) {
730                 case NFULNL_CFG_CMD_BIND:
731                         if (inst) {
732                                 ret = -EBUSY;
733                                 goto out_put;
734                         }
735
736                         inst = instance_create(group_num,
737                                                NETLINK_CB(skb).pid);
738                         if (IS_ERR(inst)) {
739                                 ret = PTR_ERR(inst);
740                                 goto out;
741                         }
742                         break;
743                 case NFULNL_CFG_CMD_UNBIND:
744                         if (!inst) {
745                                 ret = -ENODEV;
746                                 goto out;
747                         }
748
749                         instance_destroy(inst);
750                         goto out;
751                 default:
752                         ret = -ENOTSUPP;
753                         break;
754                 }
755         }
756
757         if (nfula[NFULA_CFG_MODE]) {
758                 struct nfulnl_msg_config_mode *params;
759                 params = nla_data(nfula[NFULA_CFG_MODE]);
760
761                 if (!inst) {
762                         ret = -ENODEV;
763                         goto out;
764                 }
765                 nfulnl_set_mode(inst, params->copy_mode,
766                                 ntohl(params->copy_range));
767         }
768
769         if (nfula[NFULA_CFG_TIMEOUT]) {
770                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
771
772                 if (!inst) {
773                         ret = -ENODEV;
774                         goto out;
775                 }
776                 nfulnl_set_timeout(inst, ntohl(timeout));
777         }
778
779         if (nfula[NFULA_CFG_NLBUFSIZ]) {
780                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
781
782                 if (!inst) {
783                         ret = -ENODEV;
784                         goto out;
785                 }
786                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
787         }
788
789         if (nfula[NFULA_CFG_QTHRESH]) {
790                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
791
792                 if (!inst) {
793                         ret = -ENODEV;
794                         goto out;
795                 }
796                 nfulnl_set_qthresh(inst, ntohl(qthresh));
797         }
798
799         if (nfula[NFULA_CFG_FLAGS]) {
800                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
801
802                 if (!inst) {
803                         ret = -ENODEV;
804                         goto out;
805                 }
806                 nfulnl_set_flags(inst, ntohs(flags));
807         }
808
809 out_put:
810         instance_put(inst);
811 out:
812         return ret;
813 }
814
815 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
816         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
817                                     .attr_count = NFULA_MAX, },
818         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
819                                     .attr_count = NFULA_CFG_MAX,
820                                     .policy = nfula_cfg_policy },
821 };
822
823 static const struct nfnetlink_subsystem nfulnl_subsys = {
824         .name           = "log",
825         .subsys_id      = NFNL_SUBSYS_ULOG,
826         .cb_count       = NFULNL_MSG_MAX,
827         .cb             = nfulnl_cb,
828 };
829
830 #ifdef CONFIG_PROC_FS
831 struct iter_state {
832         unsigned int bucket;
833 };
834
835 static struct hlist_node *get_first(struct iter_state *st)
836 {
837         if (!st)
838                 return NULL;
839
840         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
841                 if (!hlist_empty(&instance_table[st->bucket]))
842                         return instance_table[st->bucket].first;
843         }
844         return NULL;
845 }
846
847 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
848 {
849         h = h->next;
850         while (!h) {
851                 if (++st->bucket >= INSTANCE_BUCKETS)
852                         return NULL;
853
854                 h = instance_table[st->bucket].first;
855         }
856         return h;
857 }
858
859 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
860 {
861         struct hlist_node *head;
862         head = get_first(st);
863
864         if (head)
865                 while (pos && (head = get_next(st, head)))
866                         pos--;
867         return pos ? NULL : head;
868 }
869
870 static void *seq_start(struct seq_file *seq, loff_t *pos)
871         __acquires(instances_lock)
872 {
873         read_lock_bh(&instances_lock);
874         return get_idx(seq->private, *pos);
875 }
876
877 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
878 {
879         (*pos)++;
880         return get_next(s->private, v);
881 }
882
883 static void seq_stop(struct seq_file *s, void *v)
884         __releases(instances_lock)
885 {
886         read_unlock_bh(&instances_lock);
887 }
888
889 static int seq_show(struct seq_file *s, void *v)
890 {
891         const struct nfulnl_instance *inst = v;
892
893         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
894                           inst->group_num,
895                           inst->peer_pid, inst->qlen,
896                           inst->copy_mode, inst->copy_range,
897                           inst->flushtimeout, atomic_read(&inst->use));
898 }
899
900 static const struct seq_operations nful_seq_ops = {
901         .start  = seq_start,
902         .next   = seq_next,
903         .stop   = seq_stop,
904         .show   = seq_show,
905 };
906
907 static int nful_open(struct inode *inode, struct file *file)
908 {
909         return seq_open_private(file, &nful_seq_ops,
910                         sizeof(struct iter_state));
911 }
912
913 static const struct file_operations nful_file_ops = {
914         .owner   = THIS_MODULE,
915         .open    = nful_open,
916         .read    = seq_read,
917         .llseek  = seq_lseek,
918         .release = seq_release_private,
919 };
920
921 #endif /* PROC_FS */
922
923 static int __init nfnetlink_log_init(void)
924 {
925         int i, status = -ENOMEM;
926
927         for (i = 0; i < INSTANCE_BUCKETS; i++)
928                 INIT_HLIST_HEAD(&instance_table[i]);
929
930         /* it's not really all that important to have a random value, so
931          * we can do this from the init function, even if there hasn't
932          * been that much entropy yet */
933         get_random_bytes(&hash_init, sizeof(hash_init));
934
935         netlink_register_notifier(&nfulnl_rtnl_notifier);
936         status = nfnetlink_subsys_register(&nfulnl_subsys);
937         if (status < 0) {
938                 printk(KERN_ERR "log: failed to create netlink socket\n");
939                 goto cleanup_netlink_notifier;
940         }
941
942 #ifdef CONFIG_PROC_FS
943         if (!proc_create("nfnetlink_log", 0440,
944                          proc_net_netfilter, &nful_file_ops))
945                 goto cleanup_subsys;
946 #endif
947         return status;
948
949 #ifdef CONFIG_PROC_FS
950 cleanup_subsys:
951         nfnetlink_subsys_unregister(&nfulnl_subsys);
952 #endif
953 cleanup_netlink_notifier:
954         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
955         return status;
956 }
957
958 static void __exit nfnetlink_log_fini(void)
959 {
960         nf_log_unregister(&nfulnl_logger);
961 #ifdef CONFIG_PROC_FS
962         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
963 #endif
964         nfnetlink_subsys_unregister(&nfulnl_subsys);
965         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
966 }
967
968 MODULE_DESCRIPTION("netfilter userspace logging");
969 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
970 MODULE_LICENSE("GPL");
971 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
972
973 module_init(nfnetlink_log_init);
974 module_exit(nfnetlink_log_fini);