bridge: use rx_handler_data pointer to store net_bridge_port pointer
[linux-2.6.git] / 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 <linux/slab.h>
32 #include <net/sock.h>
33 #include <net/netfilter/nf_log.h>
34 #include <net/netfilter/nfnetlink_log.h>
35
36 #include <asm/atomic.h>
37
38 #ifdef CONFIG_BRIDGE_NETFILTER
39 #include "../bridge/br_private.h"
40 #endif
41
42 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
43 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
44 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
45 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
46
47 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
48                                      printk(x, ## args); } while (0);
49
50 struct nfulnl_instance {
51         struct hlist_node hlist;        /* global list of instances */
52         spinlock_t lock;
53         atomic_t use;                   /* use count */
54
55         unsigned int qlen;              /* number of nlmsgs in skb */
56         struct sk_buff *skb;            /* pre-allocatd skb */
57         struct timer_list timer;
58         int peer_pid;                   /* PID of the peer process */
59
60         /* configurable parameters */
61         unsigned int flushtimeout;      /* timeout until queue flush */
62         unsigned int nlbufsiz;          /* netlink buffer allocation size */
63         unsigned int qthreshold;        /* threshold of the queue */
64         u_int32_t copy_range;
65         u_int32_t seq;                  /* instance-local sequential counter */
66         u_int16_t group_num;            /* number of this queue */
67         u_int16_t flags;
68         u_int8_t copy_mode;
69 };
70
71 static DEFINE_RWLOCK(instances_lock);
72 static atomic_t global_seq;
73
74 #define INSTANCE_BUCKETS        16
75 static struct hlist_head instance_table[INSTANCE_BUCKETS];
76 static unsigned int hash_init;
77
78 static inline u_int8_t instance_hashfn(u_int16_t group_num)
79 {
80         return ((group_num & 0xff) % INSTANCE_BUCKETS);
81 }
82
83 static struct nfulnl_instance *
84 __instance_lookup(u_int16_t group_num)
85 {
86         struct hlist_head *head;
87         struct hlist_node *pos;
88         struct nfulnl_instance *inst;
89
90         head = &instance_table[instance_hashfn(group_num)];
91         hlist_for_each_entry(inst, pos, head, hlist) {
92                 if (inst->group_num == group_num)
93                         return inst;
94         }
95         return NULL;
96 }
97
98 static inline void
99 instance_get(struct nfulnl_instance *inst)
100 {
101         atomic_inc(&inst->use);
102 }
103
104 static struct nfulnl_instance *
105 instance_lookup_get(u_int16_t group_num)
106 {
107         struct nfulnl_instance *inst;
108
109         read_lock_bh(&instances_lock);
110         inst = __instance_lookup(group_num);
111         if (inst)
112                 instance_get(inst);
113         read_unlock_bh(&instances_lock);
114
115         return inst;
116 }
117
118 static void
119 instance_put(struct nfulnl_instance *inst)
120 {
121         if (inst && atomic_dec_and_test(&inst->use)) {
122                 kfree(inst);
123                 module_put(THIS_MODULE);
124         }
125 }
126
127 static void nfulnl_timer(unsigned long data);
128
129 static struct nfulnl_instance *
130 instance_create(u_int16_t group_num, int pid)
131 {
132         struct nfulnl_instance *inst;
133         int err;
134
135         write_lock_bh(&instances_lock);
136         if (__instance_lookup(group_num)) {
137                 err = -EEXIST;
138                 goto out_unlock;
139         }
140
141         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
142         if (!inst) {
143                 err = -ENOMEM;
144                 goto out_unlock;
145         }
146
147         if (!try_module_get(THIS_MODULE)) {
148                 kfree(inst);
149                 err = -EAGAIN;
150                 goto out_unlock;
151         }
152
153         INIT_HLIST_NODE(&inst->hlist);
154         spin_lock_init(&inst->lock);
155         /* needs to be two, since we _put() after creation */
156         atomic_set(&inst->use, 2);
157
158         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
159
160         inst->peer_pid = pid;
161         inst->group_num = group_num;
162
163         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
164         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
165         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
166         inst->copy_mode         = NFULNL_COPY_PACKET;
167         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
168
169         hlist_add_head(&inst->hlist,
170                        &instance_table[instance_hashfn(group_num)]);
171
172         write_unlock_bh(&instances_lock);
173
174         return inst;
175
176 out_unlock:
177         write_unlock_bh(&instances_lock);
178         return ERR_PTR(err);
179 }
180
181 static void __nfulnl_flush(struct nfulnl_instance *inst);
182
183 static void
184 __instance_destroy(struct nfulnl_instance *inst)
185 {
186         /* first pull it out of the global list */
187         hlist_del(&inst->hlist);
188
189         /* then flush all pending packets from skb */
190
191         spin_lock_bh(&inst->lock);
192         if (inst->skb)
193                 __nfulnl_flush(inst);
194         spin_unlock_bh(&inst->lock);
195
196         /* and finally put the refcount */
197         instance_put(inst);
198 }
199
200 static inline void
201 instance_destroy(struct nfulnl_instance *inst)
202 {
203         write_lock_bh(&instances_lock);
204         __instance_destroy(inst);
205         write_unlock_bh(&instances_lock);
206 }
207
208 static int
209 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
210                   unsigned int range)
211 {
212         int status = 0;
213
214         spin_lock_bh(&inst->lock);
215
216         switch (mode) {
217         case NFULNL_COPY_NONE:
218         case NFULNL_COPY_META:
219                 inst->copy_mode = mode;
220                 inst->copy_range = 0;
221                 break;
222
223         case NFULNL_COPY_PACKET:
224                 inst->copy_mode = mode;
225                 inst->copy_range = min_t(unsigned int,
226                                          range, NFULNL_COPY_RANGE_MAX);
227                 break;
228
229         default:
230                 status = -EINVAL;
231                 break;
232         }
233
234         spin_unlock_bh(&inst->lock);
235
236         return status;
237 }
238
239 static int
240 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
241 {
242         int status;
243
244         spin_lock_bh(&inst->lock);
245         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
246                 status = -ERANGE;
247         else if (nlbufsiz > 131072)
248                 status = -ERANGE;
249         else {
250                 inst->nlbufsiz = nlbufsiz;
251                 status = 0;
252         }
253         spin_unlock_bh(&inst->lock);
254
255         return status;
256 }
257
258 static int
259 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
260 {
261         spin_lock_bh(&inst->lock);
262         inst->flushtimeout = timeout;
263         spin_unlock_bh(&inst->lock);
264
265         return 0;
266 }
267
268 static int
269 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
270 {
271         spin_lock_bh(&inst->lock);
272         inst->qthreshold = qthresh;
273         spin_unlock_bh(&inst->lock);
274
275         return 0;
276 }
277
278 static int
279 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
280 {
281         spin_lock_bh(&inst->lock);
282         inst->flags = flags;
283         spin_unlock_bh(&inst->lock);
284
285         return 0;
286 }
287
288 static struct sk_buff *
289 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
290 {
291         struct sk_buff *skb;
292         unsigned int n;
293
294         /* alloc skb which should be big enough for a whole multipart
295          * message.  WARNING: has to be <= 128k due to slab restrictions */
296
297         n = max(inst_size, pkt_size);
298         skb = alloc_skb(n, GFP_ATOMIC);
299         if (!skb) {
300                 pr_notice("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
301                         inst_size);
302
303                 if (n > pkt_size) {
304                         /* try to allocate only as much as we need for current
305                          * packet */
306
307                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
308                         if (!skb)
309                                 pr_err("nfnetlink_log: can't even alloc %u "
310                                        "bytes\n", pkt_size);
311                 }
312         }
313
314         return skb;
315 }
316
317 static int
318 __nfulnl_send(struct nfulnl_instance *inst)
319 {
320         int status = -1;
321
322         if (inst->qlen > 1)
323                 NLMSG_PUT(inst->skb, 0, 0,
324                           NLMSG_DONE,
325                           sizeof(struct nfgenmsg));
326
327         status = nfnetlink_unicast(inst->skb, &init_net, inst->peer_pid,
328                                    MSG_DONTWAIT);
329
330         inst->qlen = 0;
331         inst->skb = NULL;
332
333 nlmsg_failure:
334         return status;
335 }
336
337 static void
338 __nfulnl_flush(struct nfulnl_instance *inst)
339 {
340         /* timer holds a reference */
341         if (del_timer(&inst->timer))
342                 instance_put(inst);
343         if (inst->skb)
344                 __nfulnl_send(inst);
345 }
346
347 static void
348 nfulnl_timer(unsigned long data)
349 {
350         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
351
352         spin_lock_bh(&inst->lock);
353         if (inst->skb)
354                 __nfulnl_send(inst);
355         spin_unlock_bh(&inst->lock);
356         instance_put(inst);
357 }
358
359 /* This is an inline function, we don't really care about a long
360  * list of arguments */
361 static inline int
362 __build_packet_message(struct nfulnl_instance *inst,
363                         const struct sk_buff *skb,
364                         unsigned int data_len,
365                         u_int8_t pf,
366                         unsigned int hooknum,
367                         const struct net_device *indev,
368                         const struct net_device *outdev,
369                         const struct nf_loginfo *li,
370                         const char *prefix, unsigned int plen)
371 {
372         struct nfulnl_msg_packet_hdr pmsg;
373         struct nlmsghdr *nlh;
374         struct nfgenmsg *nfmsg;
375         __be32 tmp_uint;
376         sk_buff_data_t old_tail = inst->skb->tail;
377
378         nlh = NLMSG_PUT(inst->skb, 0, 0,
379                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
380                         sizeof(struct nfgenmsg));
381         nfmsg = NLMSG_DATA(nlh);
382         nfmsg->nfgen_family = pf;
383         nfmsg->version = NFNETLINK_V0;
384         nfmsg->res_id = htons(inst->group_num);
385
386         pmsg.hw_protocol        = skb->protocol;
387         pmsg.hook               = hooknum;
388
389         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
390
391         if (prefix)
392                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
393
394         if (indev) {
395 #ifndef CONFIG_BRIDGE_NETFILTER
396                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
397                              htonl(indev->ifindex));
398 #else
399                 if (pf == PF_BRIDGE) {
400                         /* Case 1: outdev is physical input device, we need to
401                          * look for bridge group (when called from
402                          * netfilter_bridge) */
403                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
404                                      htonl(indev->ifindex));
405                         /* this is the bridge group "brX" */
406                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
407                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
408                                      htonl(br_port_get_rcu(indev)->br->dev->ifindex));
409                 } else {
410                         /* Case 2: indev is bridge group, we need to look for
411                          * physical device (when called from ipv4) */
412                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
413                                      htonl(indev->ifindex));
414                         if (skb->nf_bridge && skb->nf_bridge->physindev)
415                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
416                                              htonl(skb->nf_bridge->physindev->ifindex));
417                 }
418 #endif
419         }
420
421         if (outdev) {
422                 tmp_uint = htonl(outdev->ifindex);
423 #ifndef CONFIG_BRIDGE_NETFILTER
424                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
425                              htonl(outdev->ifindex));
426 #else
427                 if (pf == PF_BRIDGE) {
428                         /* Case 1: outdev is physical output device, we need to
429                          * look for bridge group (when called from
430                          * netfilter_bridge) */
431                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
432                                      htonl(outdev->ifindex));
433                         /* this is the bridge group "brX" */
434                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
435                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
436                                      htonl(br_port_get_rcu(outdev)->br->dev->ifindex));
437                 } else {
438                         /* Case 2: indev is a bridge group, we need to look
439                          * for physical device (when called from ipv4) */
440                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
441                                      htonl(outdev->ifindex));
442                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
443                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
444                                              htonl(skb->nf_bridge->physoutdev->ifindex));
445                 }
446 #endif
447         }
448
449         if (skb->mark)
450                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
451
452         if (indev && skb->dev) {
453                 struct nfulnl_msg_packet_hw phw;
454                 int len = dev_parse_header(skb, phw.hw_addr);
455                 if (len > 0) {
456                         phw.hw_addrlen = htons(len);
457                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
458                 }
459         }
460
461         if (indev && skb_mac_header_was_set(skb)) {
462                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
463                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
464                              htons(skb->dev->hard_header_len));
465                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
466                         skb_mac_header(skb));
467         }
468
469         if (skb->tstamp.tv64) {
470                 struct nfulnl_msg_packet_timestamp ts;
471                 struct timeval tv = ktime_to_timeval(skb->tstamp);
472                 ts.sec = cpu_to_be64(tv.tv_sec);
473                 ts.usec = cpu_to_be64(tv.tv_usec);
474
475                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
476         }
477
478         /* UID */
479         if (skb->sk) {
480                 read_lock_bh(&skb->sk->sk_callback_lock);
481                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
482                         struct file *file = skb->sk->sk_socket->file;
483                         __be32 uid = htonl(file->f_cred->fsuid);
484                         __be32 gid = htonl(file->f_cred->fsgid);
485                         /* need to unlock here since NLA_PUT may goto */
486                         read_unlock_bh(&skb->sk->sk_callback_lock);
487                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
488                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
489                 } else
490                         read_unlock_bh(&skb->sk->sk_callback_lock);
491         }
492
493         /* local sequence number */
494         if (inst->flags & NFULNL_CFG_F_SEQ)
495                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
496
497         /* global sequence number */
498         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
499                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
500                              htonl(atomic_inc_return(&global_seq)));
501
502         if (data_len) {
503                 struct nlattr *nla;
504                 int size = nla_attr_size(data_len);
505
506                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
507                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
508                         goto nlmsg_failure;
509                 }
510
511                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
512                 nla->nla_type = NFULA_PAYLOAD;
513                 nla->nla_len = size;
514
515                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
516                         BUG();
517         }
518
519         nlh->nlmsg_len = inst->skb->tail - old_tail;
520         return 0;
521
522 nlmsg_failure:
523 nla_put_failure:
524         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
525         return -1;
526 }
527
528 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
529
530 static struct nf_loginfo default_loginfo = {
531         .type =         NF_LOG_TYPE_ULOG,
532         .u = {
533                 .ulog = {
534                         .copy_len       = 0xffff,
535                         .group          = 0,
536                         .qthreshold     = 1,
537                 },
538         },
539 };
540
541 /* log handler for internal netfilter logging api */
542 void
543 nfulnl_log_packet(u_int8_t pf,
544                   unsigned int hooknum,
545                   const struct sk_buff *skb,
546                   const struct net_device *in,
547                   const struct net_device *out,
548                   const struct nf_loginfo *li_user,
549                   const char *prefix)
550 {
551         unsigned int size, data_len;
552         struct nfulnl_instance *inst;
553         const struct nf_loginfo *li;
554         unsigned int qthreshold;
555         unsigned int plen;
556
557         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
558                 li = li_user;
559         else
560                 li = &default_loginfo;
561
562         inst = instance_lookup_get(li->u.ulog.group);
563         if (!inst)
564                 return;
565
566         plen = 0;
567         if (prefix)
568                 plen = strlen(prefix) + 1;
569
570         /* FIXME: do we want to make the size calculation conditional based on
571          * what is actually present?  way more branches and checks, but more
572          * memory efficient... */
573         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
574                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
575                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
576                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
577 #ifdef CONFIG_BRIDGE_NETFILTER
578                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
579                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
580 #endif
581                 + nla_total_size(sizeof(u_int32_t))     /* mark */
582                 + nla_total_size(sizeof(u_int32_t))     /* uid */
583                 + nla_total_size(sizeof(u_int32_t))     /* gid */
584                 + nla_total_size(plen)                  /* prefix */
585                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
586                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
587
588         if (in && skb_mac_header_was_set(skb)) {
589                 size +=   nla_total_size(skb->dev->hard_header_len)
590                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
591                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
592         }
593
594         spin_lock_bh(&inst->lock);
595
596         if (inst->flags & NFULNL_CFG_F_SEQ)
597                 size += nla_total_size(sizeof(u_int32_t));
598         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
599                 size += nla_total_size(sizeof(u_int32_t));
600
601         qthreshold = inst->qthreshold;
602         /* per-rule qthreshold overrides per-instance */
603         if (li->u.ulog.qthreshold)
604                 if (qthreshold > li->u.ulog.qthreshold)
605                         qthreshold = li->u.ulog.qthreshold;
606
607
608         switch (inst->copy_mode) {
609         case NFULNL_COPY_META:
610         case NFULNL_COPY_NONE:
611                 data_len = 0;
612                 break;
613
614         case NFULNL_COPY_PACKET:
615                 if (inst->copy_range == 0
616                     || inst->copy_range > skb->len)
617                         data_len = skb->len;
618                 else
619                         data_len = inst->copy_range;
620
621                 size += nla_total_size(data_len);
622                 break;
623
624         default:
625                 goto unlock_and_release;
626         }
627
628         if (inst->skb &&
629             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
630                 /* either the queue len is too high or we don't have
631                  * enough room in the skb left. flush to userspace. */
632                 __nfulnl_flush(inst);
633         }
634
635         if (!inst->skb) {
636                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
637                 if (!inst->skb)
638                         goto alloc_failure;
639         }
640
641         inst->qlen++;
642
643         __build_packet_message(inst, skb, data_len, pf,
644                                 hooknum, in, out, li, prefix, plen);
645
646         if (inst->qlen >= qthreshold)
647                 __nfulnl_flush(inst);
648         /* timer_pending always called within inst->lock, so there
649          * is no chance of a race here */
650         else if (!timer_pending(&inst->timer)) {
651                 instance_get(inst);
652                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
653                 add_timer(&inst->timer);
654         }
655
656 unlock_and_release:
657         spin_unlock_bh(&inst->lock);
658         instance_put(inst);
659         return;
660
661 alloc_failure:
662         /* FIXME: statistics */
663         goto unlock_and_release;
664 }
665 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
666
667 static int
668 nfulnl_rcv_nl_event(struct notifier_block *this,
669                    unsigned long event, void *ptr)
670 {
671         struct netlink_notify *n = ptr;
672
673         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
674                 int i;
675
676                 /* destroy all instances for this pid */
677                 write_lock_bh(&instances_lock);
678                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
679                         struct hlist_node *tmp, *t2;
680                         struct nfulnl_instance *inst;
681                         struct hlist_head *head = &instance_table[i];
682
683                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
684                                 if ((net_eq(n->net, &init_net)) &&
685                                     (n->pid == inst->peer_pid))
686                                         __instance_destroy(inst);
687                         }
688                 }
689                 write_unlock_bh(&instances_lock);
690         }
691         return NOTIFY_DONE;
692 }
693
694 static struct notifier_block nfulnl_rtnl_notifier = {
695         .notifier_call  = nfulnl_rcv_nl_event,
696 };
697
698 static int
699 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
700                    const struct nlmsghdr *nlh,
701                    const struct nlattr * const nfqa[])
702 {
703         return -ENOTSUPP;
704 }
705
706 static struct nf_logger nfulnl_logger __read_mostly = {
707         .name   = "nfnetlink_log",
708         .logfn  = &nfulnl_log_packet,
709         .me     = THIS_MODULE,
710 };
711
712 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
713         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
714         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
715         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
716         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
717         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
718         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
719 };
720
721 static int
722 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
723                    const struct nlmsghdr *nlh,
724                    const struct nlattr * const nfula[])
725 {
726         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
727         u_int16_t group_num = ntohs(nfmsg->res_id);
728         struct nfulnl_instance *inst;
729         struct nfulnl_msg_config_cmd *cmd = NULL;
730         int ret = 0;
731
732         if (nfula[NFULA_CFG_CMD]) {
733                 u_int8_t pf = nfmsg->nfgen_family;
734                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
735
736                 /* Commands without queue context */
737                 switch (cmd->command) {
738                 case NFULNL_CFG_CMD_PF_BIND:
739                         return nf_log_bind_pf(pf, &nfulnl_logger);
740                 case NFULNL_CFG_CMD_PF_UNBIND:
741                         nf_log_unbind_pf(pf);
742                         return 0;
743                 }
744         }
745
746         inst = instance_lookup_get(group_num);
747         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
748                 ret = -EPERM;
749                 goto out_put;
750         }
751
752         if (cmd != NULL) {
753                 switch (cmd->command) {
754                 case NFULNL_CFG_CMD_BIND:
755                         if (inst) {
756                                 ret = -EBUSY;
757                                 goto out_put;
758                         }
759
760                         inst = instance_create(group_num,
761                                                NETLINK_CB(skb).pid);
762                         if (IS_ERR(inst)) {
763                                 ret = PTR_ERR(inst);
764                                 goto out;
765                         }
766                         break;
767                 case NFULNL_CFG_CMD_UNBIND:
768                         if (!inst) {
769                                 ret = -ENODEV;
770                                 goto out;
771                         }
772
773                         instance_destroy(inst);
774                         goto out_put;
775                 default:
776                         ret = -ENOTSUPP;
777                         break;
778                 }
779         }
780
781         if (nfula[NFULA_CFG_MODE]) {
782                 struct nfulnl_msg_config_mode *params;
783                 params = nla_data(nfula[NFULA_CFG_MODE]);
784
785                 if (!inst) {
786                         ret = -ENODEV;
787                         goto out;
788                 }
789                 nfulnl_set_mode(inst, params->copy_mode,
790                                 ntohl(params->copy_range));
791         }
792
793         if (nfula[NFULA_CFG_TIMEOUT]) {
794                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
795
796                 if (!inst) {
797                         ret = -ENODEV;
798                         goto out;
799                 }
800                 nfulnl_set_timeout(inst, ntohl(timeout));
801         }
802
803         if (nfula[NFULA_CFG_NLBUFSIZ]) {
804                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
805
806                 if (!inst) {
807                         ret = -ENODEV;
808                         goto out;
809                 }
810                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
811         }
812
813         if (nfula[NFULA_CFG_QTHRESH]) {
814                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
815
816                 if (!inst) {
817                         ret = -ENODEV;
818                         goto out;
819                 }
820                 nfulnl_set_qthresh(inst, ntohl(qthresh));
821         }
822
823         if (nfula[NFULA_CFG_FLAGS]) {
824                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
825
826                 if (!inst) {
827                         ret = -ENODEV;
828                         goto out;
829                 }
830                 nfulnl_set_flags(inst, ntohs(flags));
831         }
832
833 out_put:
834         instance_put(inst);
835 out:
836         return ret;
837 }
838
839 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
840         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
841                                     .attr_count = NFULA_MAX, },
842         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
843                                     .attr_count = NFULA_CFG_MAX,
844                                     .policy = nfula_cfg_policy },
845 };
846
847 static const struct nfnetlink_subsystem nfulnl_subsys = {
848         .name           = "log",
849         .subsys_id      = NFNL_SUBSYS_ULOG,
850         .cb_count       = NFULNL_MSG_MAX,
851         .cb             = nfulnl_cb,
852 };
853
854 #ifdef CONFIG_PROC_FS
855 struct iter_state {
856         unsigned int bucket;
857 };
858
859 static struct hlist_node *get_first(struct iter_state *st)
860 {
861         if (!st)
862                 return NULL;
863
864         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
865                 if (!hlist_empty(&instance_table[st->bucket]))
866                         return instance_table[st->bucket].first;
867         }
868         return NULL;
869 }
870
871 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
872 {
873         h = h->next;
874         while (!h) {
875                 if (++st->bucket >= INSTANCE_BUCKETS)
876                         return NULL;
877
878                 h = instance_table[st->bucket].first;
879         }
880         return h;
881 }
882
883 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
884 {
885         struct hlist_node *head;
886         head = get_first(st);
887
888         if (head)
889                 while (pos && (head = get_next(st, head)))
890                         pos--;
891         return pos ? NULL : head;
892 }
893
894 static void *seq_start(struct seq_file *seq, loff_t *pos)
895         __acquires(instances_lock)
896 {
897         read_lock_bh(&instances_lock);
898         return get_idx(seq->private, *pos);
899 }
900
901 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
902 {
903         (*pos)++;
904         return get_next(s->private, v);
905 }
906
907 static void seq_stop(struct seq_file *s, void *v)
908         __releases(instances_lock)
909 {
910         read_unlock_bh(&instances_lock);
911 }
912
913 static int seq_show(struct seq_file *s, void *v)
914 {
915         const struct nfulnl_instance *inst = v;
916
917         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
918                           inst->group_num,
919                           inst->peer_pid, inst->qlen,
920                           inst->copy_mode, inst->copy_range,
921                           inst->flushtimeout, atomic_read(&inst->use));
922 }
923
924 static const struct seq_operations nful_seq_ops = {
925         .start  = seq_start,
926         .next   = seq_next,
927         .stop   = seq_stop,
928         .show   = seq_show,
929 };
930
931 static int nful_open(struct inode *inode, struct file *file)
932 {
933         return seq_open_private(file, &nful_seq_ops,
934                         sizeof(struct iter_state));
935 }
936
937 static const struct file_operations nful_file_ops = {
938         .owner   = THIS_MODULE,
939         .open    = nful_open,
940         .read    = seq_read,
941         .llseek  = seq_lseek,
942         .release = seq_release_private,
943 };
944
945 #endif /* PROC_FS */
946
947 static int __init nfnetlink_log_init(void)
948 {
949         int i, status = -ENOMEM;
950
951         for (i = 0; i < INSTANCE_BUCKETS; i++)
952                 INIT_HLIST_HEAD(&instance_table[i]);
953
954         /* it's not really all that important to have a random value, so
955          * we can do this from the init function, even if there hasn't
956          * been that much entropy yet */
957         get_random_bytes(&hash_init, sizeof(hash_init));
958
959         netlink_register_notifier(&nfulnl_rtnl_notifier);
960         status = nfnetlink_subsys_register(&nfulnl_subsys);
961         if (status < 0) {
962                 printk(KERN_ERR "log: failed to create netlink socket\n");
963                 goto cleanup_netlink_notifier;
964         }
965
966         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
967         if (status < 0) {
968                 printk(KERN_ERR "log: failed to register logger\n");
969                 goto cleanup_subsys;
970         }
971
972 #ifdef CONFIG_PROC_FS
973         if (!proc_create("nfnetlink_log", 0440,
974                          proc_net_netfilter, &nful_file_ops))
975                 goto cleanup_logger;
976 #endif
977         return status;
978
979 #ifdef CONFIG_PROC_FS
980 cleanup_logger:
981         nf_log_unregister(&nfulnl_logger);
982 #endif
983 cleanup_subsys:
984         nfnetlink_subsys_unregister(&nfulnl_subsys);
985 cleanup_netlink_notifier:
986         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
987         return status;
988 }
989
990 static void __exit nfnetlink_log_fini(void)
991 {
992         nf_log_unregister(&nfulnl_logger);
993 #ifdef CONFIG_PROC_FS
994         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
995 #endif
996         nfnetlink_subsys_unregister(&nfulnl_subsys);
997         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
998 }
999
1000 MODULE_DESCRIPTION("netfilter userspace logging");
1001 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1002 MODULE_LICENSE("GPL");
1003 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1004
1005 module_init(nfnetlink_log_init);
1006 module_exit(nfnetlink_log_fini);