Revert "net: Make accesses to ->br_port safe for sparse RCU"
[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                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
407                                      htonl(indev->br_port->br->dev->ifindex));
408                 } else {
409                         /* Case 2: indev is bridge group, we need to look for
410                          * physical device (when called from ipv4) */
411                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
412                                      htonl(indev->ifindex));
413                         if (skb->nf_bridge && skb->nf_bridge->physindev)
414                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
415                                              htonl(skb->nf_bridge->physindev->ifindex));
416                 }
417 #endif
418         }
419
420         if (outdev) {
421                 tmp_uint = htonl(outdev->ifindex);
422 #ifndef CONFIG_BRIDGE_NETFILTER
423                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
424                              htonl(outdev->ifindex));
425 #else
426                 if (pf == PF_BRIDGE) {
427                         /* Case 1: outdev is physical output device, we need to
428                          * look for bridge group (when called from
429                          * netfilter_bridge) */
430                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
431                                      htonl(outdev->ifindex));
432                         /* this is the bridge group "brX" */
433                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
434                                      htonl(outdev->br_port->br->dev->ifindex));
435                 } else {
436                         /* Case 2: indev is a bridge group, we need to look
437                          * for physical device (when called from ipv4) */
438                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
439                                      htonl(outdev->ifindex));
440                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
441                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
442                                              htonl(skb->nf_bridge->physoutdev->ifindex));
443                 }
444 #endif
445         }
446
447         if (skb->mark)
448                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
449
450         if (indev && skb->dev) {
451                 struct nfulnl_msg_packet_hw phw;
452                 int len = dev_parse_header(skb, phw.hw_addr);
453                 if (len > 0) {
454                         phw.hw_addrlen = htons(len);
455                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
456                 }
457         }
458
459         if (indev && skb_mac_header_was_set(skb)) {
460                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
461                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
462                              htons(skb->dev->hard_header_len));
463                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
464                         skb_mac_header(skb));
465         }
466
467         if (skb->tstamp.tv64) {
468                 struct nfulnl_msg_packet_timestamp ts;
469                 struct timeval tv = ktime_to_timeval(skb->tstamp);
470                 ts.sec = cpu_to_be64(tv.tv_sec);
471                 ts.usec = cpu_to_be64(tv.tv_usec);
472
473                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
474         }
475
476         /* UID */
477         if (skb->sk) {
478                 read_lock_bh(&skb->sk->sk_callback_lock);
479                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
480                         struct file *file = skb->sk->sk_socket->file;
481                         __be32 uid = htonl(file->f_cred->fsuid);
482                         __be32 gid = htonl(file->f_cred->fsgid);
483                         /* need to unlock here since NLA_PUT may goto */
484                         read_unlock_bh(&skb->sk->sk_callback_lock);
485                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
486                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
487                 } else
488                         read_unlock_bh(&skb->sk->sk_callback_lock);
489         }
490
491         /* local sequence number */
492         if (inst->flags & NFULNL_CFG_F_SEQ)
493                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
494
495         /* global sequence number */
496         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
497                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
498                              htonl(atomic_inc_return(&global_seq)));
499
500         if (data_len) {
501                 struct nlattr *nla;
502                 int size = nla_attr_size(data_len);
503
504                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
505                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
506                         goto nlmsg_failure;
507                 }
508
509                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
510                 nla->nla_type = NFULA_PAYLOAD;
511                 nla->nla_len = size;
512
513                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
514                         BUG();
515         }
516
517         nlh->nlmsg_len = inst->skb->tail - old_tail;
518         return 0;
519
520 nlmsg_failure:
521 nla_put_failure:
522         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
523         return -1;
524 }
525
526 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
527
528 static struct nf_loginfo default_loginfo = {
529         .type =         NF_LOG_TYPE_ULOG,
530         .u = {
531                 .ulog = {
532                         .copy_len       = 0xffff,
533                         .group          = 0,
534                         .qthreshold     = 1,
535                 },
536         },
537 };
538
539 /* log handler for internal netfilter logging api */
540 void
541 nfulnl_log_packet(u_int8_t pf,
542                   unsigned int hooknum,
543                   const struct sk_buff *skb,
544                   const struct net_device *in,
545                   const struct net_device *out,
546                   const struct nf_loginfo *li_user,
547                   const char *prefix)
548 {
549         unsigned int size, data_len;
550         struct nfulnl_instance *inst;
551         const struct nf_loginfo *li;
552         unsigned int qthreshold;
553         unsigned int plen;
554
555         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
556                 li = li_user;
557         else
558                 li = &default_loginfo;
559
560         inst = instance_lookup_get(li->u.ulog.group);
561         if (!inst)
562                 return;
563
564         plen = 0;
565         if (prefix)
566                 plen = strlen(prefix) + 1;
567
568         /* FIXME: do we want to make the size calculation conditional based on
569          * what is actually present?  way more branches and checks, but more
570          * memory efficient... */
571         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
572                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
573                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
574                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
575 #ifdef CONFIG_BRIDGE_NETFILTER
576                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
577                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
578 #endif
579                 + nla_total_size(sizeof(u_int32_t))     /* mark */
580                 + nla_total_size(sizeof(u_int32_t))     /* uid */
581                 + nla_total_size(sizeof(u_int32_t))     /* gid */
582                 + nla_total_size(plen)                  /* prefix */
583                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
584                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
585
586         if (in && skb_mac_header_was_set(skb)) {
587                 size +=   nla_total_size(skb->dev->hard_header_len)
588                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
589                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
590         }
591
592         spin_lock_bh(&inst->lock);
593
594         if (inst->flags & NFULNL_CFG_F_SEQ)
595                 size += nla_total_size(sizeof(u_int32_t));
596         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
597                 size += nla_total_size(sizeof(u_int32_t));
598
599         qthreshold = inst->qthreshold;
600         /* per-rule qthreshold overrides per-instance */
601         if (li->u.ulog.qthreshold)
602                 if (qthreshold > li->u.ulog.qthreshold)
603                         qthreshold = li->u.ulog.qthreshold;
604
605
606         switch (inst->copy_mode) {
607         case NFULNL_COPY_META:
608         case NFULNL_COPY_NONE:
609                 data_len = 0;
610                 break;
611
612         case NFULNL_COPY_PACKET:
613                 if (inst->copy_range == 0
614                     || inst->copy_range > skb->len)
615                         data_len = skb->len;
616                 else
617                         data_len = inst->copy_range;
618
619                 size += nla_total_size(data_len);
620                 break;
621
622         default:
623                 goto unlock_and_release;
624         }
625
626         if (inst->skb &&
627             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
628                 /* either the queue len is too high or we don't have
629                  * enough room in the skb left. flush to userspace. */
630                 __nfulnl_flush(inst);
631         }
632
633         if (!inst->skb) {
634                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
635                 if (!inst->skb)
636                         goto alloc_failure;
637         }
638
639         inst->qlen++;
640
641         __build_packet_message(inst, skb, data_len, pf,
642                                 hooknum, in, out, li, prefix, plen);
643
644         if (inst->qlen >= qthreshold)
645                 __nfulnl_flush(inst);
646         /* timer_pending always called within inst->lock, so there
647          * is no chance of a race here */
648         else if (!timer_pending(&inst->timer)) {
649                 instance_get(inst);
650                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
651                 add_timer(&inst->timer);
652         }
653
654 unlock_and_release:
655         spin_unlock_bh(&inst->lock);
656         instance_put(inst);
657         return;
658
659 alloc_failure:
660         /* FIXME: statistics */
661         goto unlock_and_release;
662 }
663 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
664
665 static int
666 nfulnl_rcv_nl_event(struct notifier_block *this,
667                    unsigned long event, void *ptr)
668 {
669         struct netlink_notify *n = ptr;
670
671         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
672                 int i;
673
674                 /* destroy all instances for this pid */
675                 write_lock_bh(&instances_lock);
676                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
677                         struct hlist_node *tmp, *t2;
678                         struct nfulnl_instance *inst;
679                         struct hlist_head *head = &instance_table[i];
680
681                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
682                                 if ((net_eq(n->net, &init_net)) &&
683                                     (n->pid == inst->peer_pid))
684                                         __instance_destroy(inst);
685                         }
686                 }
687                 write_unlock_bh(&instances_lock);
688         }
689         return NOTIFY_DONE;
690 }
691
692 static struct notifier_block nfulnl_rtnl_notifier = {
693         .notifier_call  = nfulnl_rcv_nl_event,
694 };
695
696 static int
697 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
698                    const struct nlmsghdr *nlh,
699                    const struct nlattr * const nfqa[])
700 {
701         return -ENOTSUPP;
702 }
703
704 static struct nf_logger nfulnl_logger __read_mostly = {
705         .name   = "nfnetlink_log",
706         .logfn  = &nfulnl_log_packet,
707         .me     = THIS_MODULE,
708 };
709
710 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
711         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
712         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
713         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
714         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
715         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
716         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
717 };
718
719 static int
720 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
721                    const struct nlmsghdr *nlh,
722                    const struct nlattr * const nfula[])
723 {
724         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
725         u_int16_t group_num = ntohs(nfmsg->res_id);
726         struct nfulnl_instance *inst;
727         struct nfulnl_msg_config_cmd *cmd = NULL;
728         int ret = 0;
729
730         if (nfula[NFULA_CFG_CMD]) {
731                 u_int8_t pf = nfmsg->nfgen_family;
732                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
733
734                 /* Commands without queue context */
735                 switch (cmd->command) {
736                 case NFULNL_CFG_CMD_PF_BIND:
737                         return nf_log_bind_pf(pf, &nfulnl_logger);
738                 case NFULNL_CFG_CMD_PF_UNBIND:
739                         nf_log_unbind_pf(pf);
740                         return 0;
741                 }
742         }
743
744         inst = instance_lookup_get(group_num);
745         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
746                 ret = -EPERM;
747                 goto out_put;
748         }
749
750         if (cmd != NULL) {
751                 switch (cmd->command) {
752                 case NFULNL_CFG_CMD_BIND:
753                         if (inst) {
754                                 ret = -EBUSY;
755                                 goto out_put;
756                         }
757
758                         inst = instance_create(group_num,
759                                                NETLINK_CB(skb).pid);
760                         if (IS_ERR(inst)) {
761                                 ret = PTR_ERR(inst);
762                                 goto out;
763                         }
764                         break;
765                 case NFULNL_CFG_CMD_UNBIND:
766                         if (!inst) {
767                                 ret = -ENODEV;
768                                 goto out;
769                         }
770
771                         instance_destroy(inst);
772                         goto out_put;
773                 default:
774                         ret = -ENOTSUPP;
775                         break;
776                 }
777         }
778
779         if (nfula[NFULA_CFG_MODE]) {
780                 struct nfulnl_msg_config_mode *params;
781                 params = nla_data(nfula[NFULA_CFG_MODE]);
782
783                 if (!inst) {
784                         ret = -ENODEV;
785                         goto out;
786                 }
787                 nfulnl_set_mode(inst, params->copy_mode,
788                                 ntohl(params->copy_range));
789         }
790
791         if (nfula[NFULA_CFG_TIMEOUT]) {
792                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
793
794                 if (!inst) {
795                         ret = -ENODEV;
796                         goto out;
797                 }
798                 nfulnl_set_timeout(inst, ntohl(timeout));
799         }
800
801         if (nfula[NFULA_CFG_NLBUFSIZ]) {
802                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
803
804                 if (!inst) {
805                         ret = -ENODEV;
806                         goto out;
807                 }
808                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
809         }
810
811         if (nfula[NFULA_CFG_QTHRESH]) {
812                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
813
814                 if (!inst) {
815                         ret = -ENODEV;
816                         goto out;
817                 }
818                 nfulnl_set_qthresh(inst, ntohl(qthresh));
819         }
820
821         if (nfula[NFULA_CFG_FLAGS]) {
822                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
823
824                 if (!inst) {
825                         ret = -ENODEV;
826                         goto out;
827                 }
828                 nfulnl_set_flags(inst, ntohs(flags));
829         }
830
831 out_put:
832         instance_put(inst);
833 out:
834         return ret;
835 }
836
837 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
838         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
839                                     .attr_count = NFULA_MAX, },
840         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
841                                     .attr_count = NFULA_CFG_MAX,
842                                     .policy = nfula_cfg_policy },
843 };
844
845 static const struct nfnetlink_subsystem nfulnl_subsys = {
846         .name           = "log",
847         .subsys_id      = NFNL_SUBSYS_ULOG,
848         .cb_count       = NFULNL_MSG_MAX,
849         .cb             = nfulnl_cb,
850 };
851
852 #ifdef CONFIG_PROC_FS
853 struct iter_state {
854         unsigned int bucket;
855 };
856
857 static struct hlist_node *get_first(struct iter_state *st)
858 {
859         if (!st)
860                 return NULL;
861
862         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
863                 if (!hlist_empty(&instance_table[st->bucket]))
864                         return instance_table[st->bucket].first;
865         }
866         return NULL;
867 }
868
869 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
870 {
871         h = h->next;
872         while (!h) {
873                 if (++st->bucket >= INSTANCE_BUCKETS)
874                         return NULL;
875
876                 h = instance_table[st->bucket].first;
877         }
878         return h;
879 }
880
881 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
882 {
883         struct hlist_node *head;
884         head = get_first(st);
885
886         if (head)
887                 while (pos && (head = get_next(st, head)))
888                         pos--;
889         return pos ? NULL : head;
890 }
891
892 static void *seq_start(struct seq_file *seq, loff_t *pos)
893         __acquires(instances_lock)
894 {
895         read_lock_bh(&instances_lock);
896         return get_idx(seq->private, *pos);
897 }
898
899 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
900 {
901         (*pos)++;
902         return get_next(s->private, v);
903 }
904
905 static void seq_stop(struct seq_file *s, void *v)
906         __releases(instances_lock)
907 {
908         read_unlock_bh(&instances_lock);
909 }
910
911 static int seq_show(struct seq_file *s, void *v)
912 {
913         const struct nfulnl_instance *inst = v;
914
915         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
916                           inst->group_num,
917                           inst->peer_pid, inst->qlen,
918                           inst->copy_mode, inst->copy_range,
919                           inst->flushtimeout, atomic_read(&inst->use));
920 }
921
922 static const struct seq_operations nful_seq_ops = {
923         .start  = seq_start,
924         .next   = seq_next,
925         .stop   = seq_stop,
926         .show   = seq_show,
927 };
928
929 static int nful_open(struct inode *inode, struct file *file)
930 {
931         return seq_open_private(file, &nful_seq_ops,
932                         sizeof(struct iter_state));
933 }
934
935 static const struct file_operations nful_file_ops = {
936         .owner   = THIS_MODULE,
937         .open    = nful_open,
938         .read    = seq_read,
939         .llseek  = seq_lseek,
940         .release = seq_release_private,
941 };
942
943 #endif /* PROC_FS */
944
945 static int __init nfnetlink_log_init(void)
946 {
947         int i, status = -ENOMEM;
948
949         for (i = 0; i < INSTANCE_BUCKETS; i++)
950                 INIT_HLIST_HEAD(&instance_table[i]);
951
952         /* it's not really all that important to have a random value, so
953          * we can do this from the init function, even if there hasn't
954          * been that much entropy yet */
955         get_random_bytes(&hash_init, sizeof(hash_init));
956
957         netlink_register_notifier(&nfulnl_rtnl_notifier);
958         status = nfnetlink_subsys_register(&nfulnl_subsys);
959         if (status < 0) {
960                 printk(KERN_ERR "log: failed to create netlink socket\n");
961                 goto cleanup_netlink_notifier;
962         }
963
964         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
965         if (status < 0) {
966                 printk(KERN_ERR "log: failed to register logger\n");
967                 goto cleanup_subsys;
968         }
969
970 #ifdef CONFIG_PROC_FS
971         if (!proc_create("nfnetlink_log", 0440,
972                          proc_net_netfilter, &nful_file_ops))
973                 goto cleanup_logger;
974 #endif
975         return status;
976
977 #ifdef CONFIG_PROC_FS
978 cleanup_logger:
979         nf_log_unregister(&nfulnl_logger);
980 #endif
981 cleanup_subsys:
982         nfnetlink_subsys_unregister(&nfulnl_subsys);
983 cleanup_netlink_notifier:
984         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
985         return status;
986 }
987
988 static void __exit nfnetlink_log_fini(void)
989 {
990         nf_log_unregister(&nfulnl_logger);
991 #ifdef CONFIG_PROC_FS
992         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
993 #endif
994         nfnetlink_subsys_unregister(&nfulnl_subsys);
995         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
996 }
997
998 MODULE_DESCRIPTION("netfilter userspace logging");
999 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1000 MODULE_LICENSE("GPL");
1001 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1002
1003 module_init(nfnetlink_log_init);
1004 module_exit(nfnetlink_log_fini);