[NETFILTER]: nfnetlink_log: micro-optimization: don't modify destroyed instance
[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 <net/sock.h>
32
33 #include <asm/atomic.h>
34
35 #ifdef CONFIG_BRIDGE_NETFILTER
36 #include "../bridge/br_private.h"
37 #endif
38
39 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
40 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
41 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
42
43 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
44                                      printk(x, ## args); } while (0);
45
46 #if 0
47 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
48                                         __FILE__, __LINE__, __FUNCTION__,  \
49                                         ## args)
50 #else
51 #define UDEBUG(x, ...)
52 #endif
53
54 struct nfulnl_instance {
55         struct hlist_node hlist;        /* global list of instances */
56         spinlock_t lock;
57         atomic_t use;                   /* use count */
58
59         unsigned int qlen;              /* number of nlmsgs in skb */
60         struct sk_buff *skb;            /* pre-allocatd skb */
61         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
62         struct timer_list timer;
63         int peer_pid;                   /* PID of the peer process */
64
65         /* configurable parameters */
66         unsigned int flushtimeout;      /* timeout until queue flush */
67         unsigned int nlbufsiz;          /* netlink buffer allocation size */
68         unsigned int qthreshold;        /* threshold of the queue */
69         u_int32_t copy_range;
70         u_int32_t seq;                  /* instance-local sequential counter */
71         u_int16_t group_num;            /* number of this queue */
72         u_int16_t flags;
73         u_int8_t copy_mode;
74 };
75
76 static DEFINE_RWLOCK(instances_lock);
77 static atomic_t global_seq;
78
79 #define INSTANCE_BUCKETS        16
80 static struct hlist_head instance_table[INSTANCE_BUCKETS];
81 static unsigned int hash_init;
82
83 static inline u_int8_t instance_hashfn(u_int16_t group_num)
84 {
85         return ((group_num & 0xff) % INSTANCE_BUCKETS);
86 }
87
88 static struct nfulnl_instance *
89 __instance_lookup(u_int16_t group_num)
90 {
91         struct hlist_head *head;
92         struct hlist_node *pos;
93         struct nfulnl_instance *inst;
94
95         UDEBUG("entering (group_num=%u)\n", group_num);
96
97         head = &instance_table[instance_hashfn(group_num)];
98         hlist_for_each_entry(inst, pos, head, hlist) {
99                 if (inst->group_num == group_num)
100                         return inst;
101         }
102         return NULL;
103 }
104
105 static inline void
106 instance_get(struct nfulnl_instance *inst)
107 {
108         atomic_inc(&inst->use);
109 }
110
111 static struct nfulnl_instance *
112 instance_lookup_get(u_int16_t group_num)
113 {
114         struct nfulnl_instance *inst;
115
116         read_lock_bh(&instances_lock);
117         inst = __instance_lookup(group_num);
118         if (inst)
119                 instance_get(inst);
120         read_unlock_bh(&instances_lock);
121
122         return inst;
123 }
124
125 static void
126 instance_put(struct nfulnl_instance *inst)
127 {
128         if (inst && atomic_dec_and_test(&inst->use)) {
129                 UDEBUG("kfree(inst=%p)\n", inst);
130                 kfree(inst);
131                 module_put(THIS_MODULE);
132         }
133 }
134
135 static void nfulnl_timer(unsigned long data);
136
137 static struct nfulnl_instance *
138 instance_create(u_int16_t group_num, int pid)
139 {
140         struct nfulnl_instance *inst;
141
142         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
143                 pid);
144
145         write_lock_bh(&instances_lock);
146         if (__instance_lookup(group_num)) {
147                 inst = NULL;
148                 UDEBUG("aborting, instance already exists\n");
149                 goto out_unlock;
150         }
151
152         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
153         if (!inst)
154                 goto out_unlock;
155
156         INIT_HLIST_NODE(&inst->hlist);
157         spin_lock_init(&inst->lock);
158         /* needs to be two, since we _put() after creation */
159         atomic_set(&inst->use, 2);
160
161         init_timer(&inst->timer);
162         inst->timer.function = nfulnl_timer;
163         inst->timer.data = (unsigned long)inst;
164         /* don't start timer yet. (re)start it  with every packet */
165
166         inst->peer_pid = pid;
167         inst->group_num = group_num;
168
169         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
170         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
171         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
172         inst->copy_mode         = NFULNL_COPY_PACKET;
173         inst->copy_range        = 0xffff;
174
175         if (!try_module_get(THIS_MODULE))
176                 goto out_free;
177
178         hlist_add_head(&inst->hlist,
179                        &instance_table[instance_hashfn(group_num)]);
180
181         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
182                 inst->hlist.next);
183
184         write_unlock_bh(&instances_lock);
185
186         return inst;
187
188 out_free:
189         instance_put(inst);
190 out_unlock:
191         write_unlock_bh(&instances_lock);
192         return NULL;
193 }
194
195 static int __nfulnl_send(struct nfulnl_instance *inst);
196
197 static void
198 _instance_destroy2(struct nfulnl_instance *inst, int lock)
199 {
200         /* first pull it out of the global list */
201         if (lock)
202                 write_lock_bh(&instances_lock);
203
204         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
205                 inst, inst->group_num);
206
207         hlist_del(&inst->hlist);
208
209         if (lock)
210                 write_unlock_bh(&instances_lock);
211
212         /* then flush all pending packets from skb */
213
214         spin_lock_bh(&inst->lock);
215         if (inst->skb) {
216                 /* timer "holds" one reference (we have one more) */
217                 if (del_timer(&inst->timer))
218                         instance_put(inst);
219                 if (inst->qlen)
220                         __nfulnl_send(inst);
221                 if (inst->skb) {
222                         kfree_skb(inst->skb);
223                         inst->skb = NULL;
224                 }
225         }
226         spin_unlock_bh(&inst->lock);
227
228         /* and finally put the refcount */
229         instance_put(inst);
230 }
231
232 static inline void
233 __instance_destroy(struct nfulnl_instance *inst)
234 {
235         _instance_destroy2(inst, 0);
236 }
237
238 static inline void
239 instance_destroy(struct nfulnl_instance *inst)
240 {
241         _instance_destroy2(inst, 1);
242 }
243
244 static int
245 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
246                   unsigned int range)
247 {
248         int status = 0;
249
250         spin_lock_bh(&inst->lock);
251
252         switch (mode) {
253         case NFULNL_COPY_NONE:
254         case NFULNL_COPY_META:
255                 inst->copy_mode = mode;
256                 inst->copy_range = 0;
257                 break;
258
259         case NFULNL_COPY_PACKET:
260                 inst->copy_mode = mode;
261                 /* we're using struct nfattr which has 16bit nfa_len */
262                 if (range > 0xffff)
263                         inst->copy_range = 0xffff;
264                 else
265                         inst->copy_range = range;
266                 break;
267
268         default:
269                 status = -EINVAL;
270                 break;
271         }
272
273         spin_unlock_bh(&inst->lock);
274
275         return status;
276 }
277
278 static int
279 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
280 {
281         int status;
282
283         spin_lock_bh(&inst->lock);
284         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
285                 status = -ERANGE;
286         else if (nlbufsiz > 131072)
287                 status = -ERANGE;
288         else {
289                 inst->nlbufsiz = nlbufsiz;
290                 status = 0;
291         }
292         spin_unlock_bh(&inst->lock);
293
294         return status;
295 }
296
297 static int
298 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
299 {
300         spin_lock_bh(&inst->lock);
301         inst->flushtimeout = timeout;
302         spin_unlock_bh(&inst->lock);
303
304         return 0;
305 }
306
307 static int
308 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
309 {
310         spin_lock_bh(&inst->lock);
311         inst->qthreshold = qthresh;
312         spin_unlock_bh(&inst->lock);
313
314         return 0;
315 }
316
317 static int
318 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
319 {
320         spin_lock_bh(&inst->lock);
321         inst->flags = flags;
322         spin_unlock_bh(&inst->lock);
323
324         return 0;
325 }
326
327 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
328                                         unsigned int pkt_size)
329 {
330         struct sk_buff *skb;
331         unsigned int n;
332
333         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
334
335         /* alloc skb which should be big enough for a whole multipart
336          * message.  WARNING: has to be <= 128k due to slab restrictions */
337
338         n = max(inst_size, pkt_size);
339         skb = alloc_skb(n, GFP_ATOMIC);
340         if (!skb) {
341                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
342                         inst_size);
343
344                 if (n > pkt_size) {
345                         /* try to allocate only as much as we need for current
346                          * packet */
347
348                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
349                         if (!skb)
350                                 PRINTR("nfnetlink_log: can't even alloc %u "
351                                        "bytes\n", pkt_size);
352                 }
353         }
354
355         return skb;
356 }
357
358 static int
359 __nfulnl_send(struct nfulnl_instance *inst)
360 {
361         int status;
362
363         if (!inst->skb)
364                 return 0;
365
366         if (inst->qlen > 1)
367                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
368
369         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
370         if (status < 0) {
371                 UDEBUG("netlink_unicast() failed\n");
372                 /* FIXME: statistics */
373         }
374
375         inst->qlen = 0;
376         inst->skb = NULL;
377         inst->lastnlh = NULL;
378
379         return status;
380 }
381
382 static void nfulnl_timer(unsigned long data)
383 {
384         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
385
386         UDEBUG("timer function called, flushing buffer\n");
387
388         spin_lock_bh(&inst->lock);
389         __nfulnl_send(inst);
390         spin_unlock_bh(&inst->lock);
391         instance_put(inst);
392 }
393
394 /* This is an inline function, we don't really care about a long
395  * list of arguments */
396 static inline int
397 __build_packet_message(struct nfulnl_instance *inst,
398                         const struct sk_buff *skb,
399                         unsigned int data_len,
400                         unsigned int pf,
401                         unsigned int hooknum,
402                         const struct net_device *indev,
403                         const struct net_device *outdev,
404                         const struct nf_loginfo *li,
405                         const char *prefix, unsigned int plen)
406 {
407         struct nfulnl_msg_packet_hdr pmsg;
408         struct nlmsghdr *nlh;
409         struct nfgenmsg *nfmsg;
410         __be32 tmp_uint;
411         sk_buff_data_t old_tail = inst->skb->tail;
412
413         UDEBUG("entered\n");
414
415         nlh = NLMSG_PUT(inst->skb, 0, 0,
416                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
417                         sizeof(struct nfgenmsg));
418         nfmsg = NLMSG_DATA(nlh);
419         nfmsg->nfgen_family = pf;
420         nfmsg->version = NFNETLINK_V0;
421         nfmsg->res_id = htons(inst->group_num);
422
423         pmsg.hw_protocol        = skb->protocol;
424         pmsg.hook               = hooknum;
425
426         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
427
428         if (prefix)
429                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
430
431         if (indev) {
432                 tmp_uint = htonl(indev->ifindex);
433 #ifndef CONFIG_BRIDGE_NETFILTER
434                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
435                         &tmp_uint);
436 #else
437                 if (pf == PF_BRIDGE) {
438                         /* Case 1: outdev is physical input device, we need to
439                          * look for bridge group (when called from
440                          * netfilter_bridge) */
441                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
442                                 sizeof(tmp_uint), &tmp_uint);
443                         /* this is the bridge group "brX" */
444                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
445                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
446                                 sizeof(tmp_uint), &tmp_uint);
447                 } else {
448                         /* Case 2: indev is bridge group, we need to look for
449                          * physical device (when called from ipv4) */
450                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
451                                 sizeof(tmp_uint), &tmp_uint);
452                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
453                                 tmp_uint =
454                                     htonl(skb->nf_bridge->physindev->ifindex);
455                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
456                                         sizeof(tmp_uint), &tmp_uint);
457                         }
458                 }
459 #endif
460         }
461
462         if (outdev) {
463                 tmp_uint = htonl(outdev->ifindex);
464 #ifndef CONFIG_BRIDGE_NETFILTER
465                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
466                         &tmp_uint);
467 #else
468                 if (pf == PF_BRIDGE) {
469                         /* Case 1: outdev is physical output device, we need to
470                          * look for bridge group (when called from
471                          * netfilter_bridge) */
472                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
473                                 sizeof(tmp_uint), &tmp_uint);
474                         /* this is the bridge group "brX" */
475                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
476                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
477                                 sizeof(tmp_uint), &tmp_uint);
478                 } else {
479                         /* Case 2: indev is a bridge group, we need to look
480                          * for physical device (when called from ipv4) */
481                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
482                                 sizeof(tmp_uint), &tmp_uint);
483                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
484                                 tmp_uint =
485                                     htonl(skb->nf_bridge->physoutdev->ifindex);
486                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
487                                         sizeof(tmp_uint), &tmp_uint);
488                         }
489                 }
490 #endif
491         }
492
493         if (skb->mark) {
494                 tmp_uint = htonl(skb->mark);
495                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
496         }
497
498         if (indev && skb->dev && skb->dev->hard_header_parse) {
499                 struct nfulnl_msg_packet_hw phw;
500                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
501                                                     phw.hw_addr);
502                 phw.hw_addrlen = htons(len);
503                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
504         }
505
506         if (skb->tstamp.tv64) {
507                 struct nfulnl_msg_packet_timestamp ts;
508                 struct timeval tv = ktime_to_timeval(skb->tstamp);
509                 ts.sec = cpu_to_be64(tv.tv_sec);
510                 ts.usec = cpu_to_be64(tv.tv_usec);
511
512                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
513         }
514
515         /* UID */
516         if (skb->sk) {
517                 read_lock_bh(&skb->sk->sk_callback_lock);
518                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
519                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
520                         /* need to unlock here since NFA_PUT may goto */
521                         read_unlock_bh(&skb->sk->sk_callback_lock);
522                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
523                 } else
524                         read_unlock_bh(&skb->sk->sk_callback_lock);
525         }
526
527         /* local sequence number */
528         if (inst->flags & NFULNL_CFG_F_SEQ) {
529                 tmp_uint = htonl(inst->seq++);
530                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
531         }
532         /* global sequence number */
533         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
534                 tmp_uint = htonl(atomic_inc_return(&global_seq));
535                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
536         }
537
538         if (data_len) {
539                 struct nfattr *nfa;
540                 int size = NFA_LENGTH(data_len);
541
542                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
543                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
544                         goto nlmsg_failure;
545                 }
546
547                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
548                 nfa->nfa_type = NFULA_PAYLOAD;
549                 nfa->nfa_len = size;
550
551                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
552                         BUG();
553         }
554
555         nlh->nlmsg_len = inst->skb->tail - old_tail;
556         inst->lastnlh = nlh;
557         return 0;
558
559 nlmsg_failure:
560         UDEBUG("nlmsg_failure\n");
561 nfattr_failure:
562         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
563         return -1;
564 }
565
566 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
567
568 static struct nf_loginfo default_loginfo = {
569         .type =         NF_LOG_TYPE_ULOG,
570         .u = {
571                 .ulog = {
572                         .copy_len       = 0xffff,
573                         .group          = 0,
574                         .qthreshold     = 1,
575                 },
576         },
577 };
578
579 /* log handler for internal netfilter logging api */
580 static void
581 nfulnl_log_packet(unsigned int pf,
582                   unsigned int hooknum,
583                   const struct sk_buff *skb,
584                   const struct net_device *in,
585                   const struct net_device *out,
586                   const struct nf_loginfo *li_user,
587                   const char *prefix)
588 {
589         unsigned int size, data_len;
590         struct nfulnl_instance *inst;
591         const struct nf_loginfo *li;
592         unsigned int qthreshold;
593         unsigned int plen;
594
595         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
596                 li = li_user;
597         else
598                 li = &default_loginfo;
599
600         inst = instance_lookup_get(li->u.ulog.group);
601         if (!inst)
602                 inst = instance_lookup_get(0);
603         if (!inst) {
604                 PRINTR("nfnetlink_log: trying to log packet, "
605                         "but no instance for group %u\n", li->u.ulog.group);
606                 return;
607         }
608
609         plen = 0;
610         if (prefix)
611                 plen = strlen(prefix) + 1;
612
613         /* all macros expand to constant values at compile time */
614         /* FIXME: do we want to make the size calculation conditional based on
615          * what is actually present?  way more branches and checks, but more
616          * memory efficient... */
617         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
618                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
619                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
620                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
621 #ifdef CONFIG_BRIDGE_NETFILTER
622                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
623                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
624 #endif
625                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
626                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
627                 + NFA_SPACE(plen)               /* prefix */
628                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
629                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
630
631         UDEBUG("initial size=%u\n", size);
632
633         spin_lock_bh(&inst->lock);
634
635         if (inst->flags & NFULNL_CFG_F_SEQ)
636                 size += NFA_SPACE(sizeof(u_int32_t));
637         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
638                 size += NFA_SPACE(sizeof(u_int32_t));
639
640         qthreshold = inst->qthreshold;
641         /* per-rule qthreshold overrides per-instance */
642         if (qthreshold > li->u.ulog.qthreshold)
643                 qthreshold = li->u.ulog.qthreshold;
644
645         switch (inst->copy_mode) {
646         case NFULNL_COPY_META:
647         case NFULNL_COPY_NONE:
648                 data_len = 0;
649                 break;
650
651         case NFULNL_COPY_PACKET:
652                 if (inst->copy_range == 0
653                     || inst->copy_range > skb->len)
654                         data_len = skb->len;
655                 else
656                         data_len = inst->copy_range;
657
658                 size += NFA_SPACE(data_len);
659                 UDEBUG("copy_packet, therefore size now %u\n", size);
660                 break;
661
662         default:
663                 goto unlock_and_release;
664         }
665
666         if (inst->qlen >= qthreshold ||
667             (inst->skb && size > skb_tailroom(inst->skb))) {
668                 /* either the queue len is too high or we don't have
669                  * enough room in the skb left. flush to userspace. */
670                 UDEBUG("flushing old skb\n");
671
672                 /* timer "holds" one reference (we have another one) */
673                 if (del_timer(&inst->timer))
674                         instance_put(inst);
675                 __nfulnl_send(inst);
676         }
677
678         if (!inst->skb) {
679                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
680                 if (!inst->skb)
681                         goto alloc_failure;
682         }
683
684         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
685         inst->qlen++;
686
687         __build_packet_message(inst, skb, data_len, pf,
688                                 hooknum, in, out, li, prefix, plen);
689
690         /* timer_pending always called within inst->lock, so there
691          * is no chance of a race here */
692         if (!timer_pending(&inst->timer)) {
693                 instance_get(inst);
694                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
695                 add_timer(&inst->timer);
696         }
697
698 unlock_and_release:
699         spin_unlock_bh(&inst->lock);
700         instance_put(inst);
701         return;
702
703 alloc_failure:
704         UDEBUG("error allocating skb\n");
705         /* FIXME: statistics */
706         goto unlock_and_release;
707 }
708
709 static int
710 nfulnl_rcv_nl_event(struct notifier_block *this,
711                    unsigned long event, void *ptr)
712 {
713         struct netlink_notify *n = ptr;
714
715         if (event == NETLINK_URELEASE &&
716             n->protocol == NETLINK_NETFILTER && n->pid) {
717                 int i;
718
719                 /* destroy all instances for this pid */
720                 write_lock_bh(&instances_lock);
721                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
722                         struct hlist_node *tmp, *t2;
723                         struct nfulnl_instance *inst;
724                         struct hlist_head *head = &instance_table[i];
725
726                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
727                                 UDEBUG("node = %p\n", inst);
728                                 if (n->pid == inst->peer_pid)
729                                         __instance_destroy(inst);
730                         }
731                 }
732                 write_unlock_bh(&instances_lock);
733         }
734         return NOTIFY_DONE;
735 }
736
737 static struct notifier_block nfulnl_rtnl_notifier = {
738         .notifier_call  = nfulnl_rcv_nl_event,
739 };
740
741 static int
742 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
743                   struct nlmsghdr *nlh, struct nfattr *nfqa[])
744 {
745         return -ENOTSUPP;
746 }
747
748 static struct nf_logger nfulnl_logger = {
749         .name   = "nfnetlink_log",
750         .logfn  = &nfulnl_log_packet,
751         .me     = THIS_MODULE,
752 };
753
754 static const int nfula_min[NFULA_MAX] = {
755         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
756         [NFULA_MARK-1]          = sizeof(u_int32_t),
757         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
758         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
759         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
760         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
761         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
762         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
763         [NFULA_PAYLOAD-1]       = 0,
764         [NFULA_PREFIX-1]        = 0,
765         [NFULA_UID-1]           = sizeof(u_int32_t),
766         [NFULA_SEQ-1]           = sizeof(u_int32_t),
767         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
768 };
769
770 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
771         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
772         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
773         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
774         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
775         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
776         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
777 };
778
779 static int
780 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
781                    struct nlmsghdr *nlh, struct nfattr *nfula[])
782 {
783         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
784         u_int16_t group_num = ntohs(nfmsg->res_id);
785         struct nfulnl_instance *inst;
786         int ret = 0;
787
788         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
789
790         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
791                 UDEBUG("bad attribute size\n");
792                 return -EINVAL;
793         }
794
795         inst = instance_lookup_get(group_num);
796         if (nfula[NFULA_CFG_CMD-1]) {
797                 u_int8_t pf = nfmsg->nfgen_family;
798                 struct nfulnl_msg_config_cmd *cmd;
799                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
800                 UDEBUG("found CFG_CMD for\n");
801
802                 switch (cmd->command) {
803                 case NFULNL_CFG_CMD_BIND:
804                         if (inst) {
805                                 ret = -EBUSY;
806                                 goto out_put;
807                         }
808
809                         inst = instance_create(group_num,
810                                                NETLINK_CB(skb).pid);
811                         if (!inst) {
812                                 ret = -EINVAL;
813                                 goto out;
814                         }
815                         break;
816                 case NFULNL_CFG_CMD_UNBIND:
817                         if (!inst) {
818                                 ret = -ENODEV;
819                                 goto out;
820                         }
821
822                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
823                                 ret = -EPERM;
824                                 goto out_put;
825                         }
826
827                         instance_destroy(inst);
828                         goto out;
829                 case NFULNL_CFG_CMD_PF_BIND:
830                         UDEBUG("registering log handler for pf=%u\n", pf);
831                         ret = nf_log_register(pf, &nfulnl_logger);
832                         break;
833                 case NFULNL_CFG_CMD_PF_UNBIND:
834                         UDEBUG("unregistering log handler for pf=%u\n", pf);
835                         /* This is a bug and a feature.  We cannot unregister
836                          * other handlers, like nfnetlink_inst can */
837                         nf_log_unregister_pf(pf);
838                         break;
839                 default:
840                         ret = -EINVAL;
841                         break;
842                 }
843
844                 if (!inst)
845                         goto out;
846         } else {
847                 if (!inst) {
848                         UDEBUG("no config command, and no instance for "
849                                 "group=%u pid=%u =>ENOENT\n",
850                                 group_num, NETLINK_CB(skb).pid);
851                         ret = -ENOENT;
852                         goto out;
853                 }
854
855                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
856                         UDEBUG("no config command, and wrong pid\n");
857                         ret = -EPERM;
858                         goto out_put;
859                 }
860         }
861
862         if (nfula[NFULA_CFG_MODE-1]) {
863                 struct nfulnl_msg_config_mode *params;
864                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
865
866                 nfulnl_set_mode(inst, params->copy_mode,
867                                 ntohl(params->copy_range));
868         }
869
870         if (nfula[NFULA_CFG_TIMEOUT-1]) {
871                 __be32 timeout =
872                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
873
874                 nfulnl_set_timeout(inst, ntohl(timeout));
875         }
876
877         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
878                 __be32 nlbufsiz =
879                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
880
881                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
882         }
883
884         if (nfula[NFULA_CFG_QTHRESH-1]) {
885                 __be32 qthresh =
886                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
887
888                 nfulnl_set_qthresh(inst, ntohl(qthresh));
889         }
890
891         if (nfula[NFULA_CFG_FLAGS-1]) {
892                 __be16 flags =
893                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
894                 nfulnl_set_flags(inst, ntohs(flags));
895         }
896
897 out_put:
898         instance_put(inst);
899 out:
900         return ret;
901 }
902
903 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
904         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
905                                     .attr_count = NFULA_MAX, },
906         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
907                                     .attr_count = NFULA_CFG_MAX, },
908 };
909
910 static struct nfnetlink_subsystem nfulnl_subsys = {
911         .name           = "log",
912         .subsys_id      = NFNL_SUBSYS_ULOG,
913         .cb_count       = NFULNL_MSG_MAX,
914         .cb             = nfulnl_cb,
915 };
916
917 #ifdef CONFIG_PROC_FS
918 struct iter_state {
919         unsigned int bucket;
920 };
921
922 static struct hlist_node *get_first(struct seq_file *seq)
923 {
924         struct iter_state *st = seq->private;
925
926         if (!st)
927                 return NULL;
928
929         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
930                 if (!hlist_empty(&instance_table[st->bucket]))
931                         return instance_table[st->bucket].first;
932         }
933         return NULL;
934 }
935
936 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
937 {
938         struct iter_state *st = seq->private;
939
940         h = h->next;
941         while (!h) {
942                 if (++st->bucket >= INSTANCE_BUCKETS)
943                         return NULL;
944
945                 h = instance_table[st->bucket].first;
946         }
947         return h;
948 }
949
950 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
951 {
952         struct hlist_node *head;
953         head = get_first(seq);
954
955         if (head)
956                 while (pos && (head = get_next(seq, head)))
957                         pos--;
958         return pos ? NULL : head;
959 }
960
961 static void *seq_start(struct seq_file *seq, loff_t *pos)
962 {
963         read_lock_bh(&instances_lock);
964         return get_idx(seq, *pos);
965 }
966
967 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
968 {
969         (*pos)++;
970         return get_next(s, v);
971 }
972
973 static void seq_stop(struct seq_file *s, void *v)
974 {
975         read_unlock_bh(&instances_lock);
976 }
977
978 static int seq_show(struct seq_file *s, void *v)
979 {
980         const struct nfulnl_instance *inst = v;
981
982         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
983                           inst->group_num,
984                           inst->peer_pid, inst->qlen,
985                           inst->copy_mode, inst->copy_range,
986                           inst->flushtimeout, atomic_read(&inst->use));
987 }
988
989 static struct seq_operations nful_seq_ops = {
990         .start  = seq_start,
991         .next   = seq_next,
992         .stop   = seq_stop,
993         .show   = seq_show,
994 };
995
996 static int nful_open(struct inode *inode, struct file *file)
997 {
998         struct seq_file *seq;
999         struct iter_state *is;
1000         int ret;
1001
1002         is = kzalloc(sizeof(*is), GFP_KERNEL);
1003         if (!is)
1004                 return -ENOMEM;
1005         ret = seq_open(file, &nful_seq_ops);
1006         if (ret < 0)
1007                 goto out_free;
1008         seq = file->private_data;
1009         seq->private = is;
1010         return ret;
1011 out_free:
1012         kfree(is);
1013         return ret;
1014 }
1015
1016 static const struct file_operations nful_file_ops = {
1017         .owner   = THIS_MODULE,
1018         .open    = nful_open,
1019         .read    = seq_read,
1020         .llseek  = seq_lseek,
1021         .release = seq_release_private,
1022 };
1023
1024 #endif /* PROC_FS */
1025
1026 static int __init nfnetlink_log_init(void)
1027 {
1028         int i, status = -ENOMEM;
1029 #ifdef CONFIG_PROC_FS
1030         struct proc_dir_entry *proc_nful;
1031 #endif
1032
1033         for (i = 0; i < INSTANCE_BUCKETS; i++)
1034                 INIT_HLIST_HEAD(&instance_table[i]);
1035
1036         /* it's not really all that important to have a random value, so
1037          * we can do this from the init function, even if there hasn't
1038          * been that much entropy yet */
1039         get_random_bytes(&hash_init, sizeof(hash_init));
1040
1041         netlink_register_notifier(&nfulnl_rtnl_notifier);
1042         status = nfnetlink_subsys_register(&nfulnl_subsys);
1043         if (status < 0) {
1044                 printk(KERN_ERR "log: failed to create netlink socket\n");
1045                 goto cleanup_netlink_notifier;
1046         }
1047
1048 #ifdef CONFIG_PROC_FS
1049         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1050                                       proc_net_netfilter);
1051         if (!proc_nful)
1052                 goto cleanup_subsys;
1053         proc_nful->proc_fops = &nful_file_ops;
1054 #endif
1055         return status;
1056
1057 #ifdef CONFIG_PROC_FS
1058 cleanup_subsys:
1059         nfnetlink_subsys_unregister(&nfulnl_subsys);
1060 #endif
1061 cleanup_netlink_notifier:
1062         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1063         return status;
1064 }
1065
1066 static void __exit nfnetlink_log_fini(void)
1067 {
1068         nf_log_unregister(&nfulnl_logger);
1069 #ifdef CONFIG_PROC_FS
1070         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1071 #endif
1072         nfnetlink_subsys_unregister(&nfulnl_subsys);
1073         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1074 }
1075
1076 MODULE_DESCRIPTION("netfilter userspace logging");
1077 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1078 MODULE_LICENSE("GPL");
1079 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1080
1081 module_init(nfnetlink_log_init);
1082 module_exit(nfnetlink_log_fini);