Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6] / net / netfilter / nfnetlink_log.c
1 /*
2  * This is a module which is used for logging packets to userspace via
3  * nfetlink.
4  *
5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
6  *
7  * Based on the old ipv4-only ipt_ULOG.c:
8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License version 2 as
12  * published by the Free Software Foundation.
13  *
14  * 2006-01-26 Harald Welte <laforge@netfilter.org>
15  *      - Add optional local and global sequence number to detect lost
16  *        events from userspace
17  *
18  */
19 #include <linux/module.h>
20 #include <linux/skbuff.h>
21 #include <linux/init.h>
22 #include <linux/ip.h>
23 #include <linux/ipv6.h>
24 #include <linux/netdevice.h>
25 #include <linux/netfilter.h>
26 #include <linux/netlink.h>
27 #include <linux/netfilter/nfnetlink.h>
28 #include <linux/netfilter/nfnetlink_log.h>
29 #include <linux/spinlock.h>
30 #include <linux/sysctl.h>
31 #include <linux/proc_fs.h>
32 #include <linux/security.h>
33 #include <linux/list.h>
34 #include <linux/jhash.h>
35 #include <linux/random.h>
36 #include <net/sock.h>
37
38 #include <asm/atomic.h>
39
40 #ifdef CONFIG_BRIDGE_NETFILTER
41 #include "../bridge/br_private.h"
42 #endif
43
44 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
45 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
46 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
47
48 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
49                                      printk(x, ## args); } while (0);
50
51 #if 0
52 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
53                                         __FILE__, __LINE__, __FUNCTION__,  \
54                                         ## args)
55 #else
56 #define UDEBUG(x, ...)
57 #endif
58
59 struct nfulnl_instance {
60         struct hlist_node hlist;        /* global list of instances */
61         spinlock_t lock;
62         atomic_t use;                   /* use count */
63
64         unsigned int qlen;              /* number of nlmsgs in skb */
65         struct sk_buff *skb;            /* pre-allocatd skb */
66         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
67         struct timer_list timer;
68         int peer_pid;                   /* PID of the peer process */
69
70         /* configurable parameters */
71         unsigned int flushtimeout;      /* timeout until queue flush */
72         unsigned int nlbufsiz;          /* netlink buffer allocation size */
73         unsigned int qthreshold;        /* threshold of the queue */
74         u_int32_t copy_range;
75         u_int32_t seq;                  /* instance-local sequential counter */
76         u_int16_t group_num;            /* number of this queue */
77         u_int16_t flags;
78         u_int8_t copy_mode;
79 };
80
81 static DEFINE_RWLOCK(instances_lock);
82 static atomic_t global_seq;
83
84 #define INSTANCE_BUCKETS        16
85 static struct hlist_head instance_table[INSTANCE_BUCKETS];
86 static unsigned int hash_init;
87
88 static inline u_int8_t instance_hashfn(u_int16_t group_num)
89 {
90         return ((group_num & 0xff) % INSTANCE_BUCKETS);
91 }
92
93 static struct nfulnl_instance *
94 __instance_lookup(u_int16_t group_num)
95 {
96         struct hlist_head *head;
97         struct hlist_node *pos;
98         struct nfulnl_instance *inst;
99
100         UDEBUG("entering (group_num=%u)\n", group_num);
101
102         head = &instance_table[instance_hashfn(group_num)];
103         hlist_for_each_entry(inst, pos, head, hlist) {
104                 if (inst->group_num == group_num)
105                         return inst;
106         }
107         return NULL;
108 }
109
110 static inline void
111 instance_get(struct nfulnl_instance *inst)
112 {
113         atomic_inc(&inst->use);
114 }
115
116 static struct nfulnl_instance *
117 instance_lookup_get(u_int16_t group_num)
118 {
119         struct nfulnl_instance *inst;
120
121         read_lock_bh(&instances_lock);
122         inst = __instance_lookup(group_num);
123         if (inst)
124                 instance_get(inst);
125         read_unlock_bh(&instances_lock);
126
127         return inst;
128 }
129
130 static void
131 instance_put(struct nfulnl_instance *inst)
132 {
133         if (inst && atomic_dec_and_test(&inst->use)) {
134                 UDEBUG("kfree(inst=%p)\n", inst);
135                 kfree(inst);
136                 module_put(THIS_MODULE);
137         }
138 }
139
140 static void nfulnl_timer(unsigned long data);
141
142 static struct nfulnl_instance *
143 instance_create(u_int16_t group_num, int pid)
144 {
145         struct nfulnl_instance *inst;
146
147         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
148                 pid);
149
150         write_lock_bh(&instances_lock);
151         if (__instance_lookup(group_num)) {
152                 inst = NULL;
153                 UDEBUG("aborting, instance already exists\n");
154                 goto out_unlock;
155         }
156
157         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
158         if (!inst)
159                 goto out_unlock;
160
161         INIT_HLIST_NODE(&inst->hlist);
162         spin_lock_init(&inst->lock);
163         /* needs to be two, since we _put() after creation */
164         atomic_set(&inst->use, 2);
165
166         init_timer(&inst->timer);
167         inst->timer.function = nfulnl_timer;
168         inst->timer.data = (unsigned long)inst;
169         /* don't start timer yet. (re)start it  with every packet */
170
171         inst->peer_pid = pid;
172         inst->group_num = group_num;
173
174         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
175         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
176         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
177         inst->copy_mode         = NFULNL_COPY_PACKET;
178         inst->copy_range        = 0xffff;
179
180         if (!try_module_get(THIS_MODULE))
181                 goto out_free;
182
183         hlist_add_head(&inst->hlist,
184                        &instance_table[instance_hashfn(group_num)]);
185
186         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
187                 inst->hlist.next);
188
189         write_unlock_bh(&instances_lock);
190
191         return inst;
192
193 out_free:
194         instance_put(inst);
195 out_unlock:
196         write_unlock_bh(&instances_lock);
197         return NULL;
198 }
199
200 static int __nfulnl_send(struct nfulnl_instance *inst);
201
202 static void
203 _instance_destroy2(struct nfulnl_instance *inst, int lock)
204 {
205         /* first pull it out of the global list */
206         if (lock)
207                 write_lock_bh(&instances_lock);
208
209         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
210                 inst, inst->group_num);
211
212         hlist_del(&inst->hlist);
213
214         if (lock)
215                 write_unlock_bh(&instances_lock);
216
217         /* then flush all pending packets from skb */
218
219         spin_lock_bh(&inst->lock);
220         if (inst->skb) {
221                 /* timer "holds" one reference (we have one more) */
222                 if (del_timer(&inst->timer))
223                         instance_put(inst);
224                 if (inst->qlen)
225                         __nfulnl_send(inst);
226                 if (inst->skb) {
227                         kfree_skb(inst->skb);
228                         inst->skb = NULL;
229                 }
230         }
231         spin_unlock_bh(&inst->lock);
232
233         /* and finally put the refcount */
234         instance_put(inst);
235 }
236
237 static inline void
238 __instance_destroy(struct nfulnl_instance *inst)
239 {
240         _instance_destroy2(inst, 0);
241 }
242
243 static inline void
244 instance_destroy(struct nfulnl_instance *inst)
245 {
246         _instance_destroy2(inst, 1);
247 }
248
249 static int
250 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
251                   unsigned int range)
252 {
253         int status = 0;
254
255         spin_lock_bh(&inst->lock);
256
257         switch (mode) {
258         case NFULNL_COPY_NONE:
259         case NFULNL_COPY_META:
260                 inst->copy_mode = mode;
261                 inst->copy_range = 0;
262                 break;
263
264         case NFULNL_COPY_PACKET:
265                 inst->copy_mode = mode;
266                 /* we're using struct nfattr which has 16bit nfa_len */
267                 if (range > 0xffff)
268                         inst->copy_range = 0xffff;
269                 else
270                         inst->copy_range = range;
271                 break;
272
273         default:
274                 status = -EINVAL;
275                 break;
276         }
277
278         spin_unlock_bh(&inst->lock);
279
280         return status;
281 }
282
283 static int
284 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
285 {
286         int status;
287
288         spin_lock_bh(&inst->lock);
289         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
290                 status = -ERANGE;
291         else if (nlbufsiz > 131072)
292                 status = -ERANGE;
293         else {
294                 inst->nlbufsiz = nlbufsiz;
295                 status = 0;
296         }
297         spin_unlock_bh(&inst->lock);
298
299         return status;
300 }
301
302 static int
303 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
304 {
305         spin_lock_bh(&inst->lock);
306         inst->flushtimeout = timeout;
307         spin_unlock_bh(&inst->lock);
308
309         return 0;
310 }
311
312 static int
313 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
314 {
315         spin_lock_bh(&inst->lock);
316         inst->qthreshold = qthresh;
317         spin_unlock_bh(&inst->lock);
318
319         return 0;
320 }
321
322 static int
323 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
324 {
325         spin_lock_bh(&inst->lock);
326         inst->flags = flags;
327         spin_unlock_bh(&inst->lock);
328
329         return 0;
330 }
331
332 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
333                                         unsigned int pkt_size)
334 {
335         struct sk_buff *skb;
336         unsigned int n;
337
338         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
339
340         /* alloc skb which should be big enough for a whole multipart
341          * message.  WARNING: has to be <= 128k due to slab restrictions */
342
343         n = max(inst_size, pkt_size);
344         skb = alloc_skb(n, GFP_ATOMIC);
345         if (!skb) {
346                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
347                         inst_size);
348
349                 if (n > pkt_size) {
350                         /* try to allocate only as much as we need for current
351                          * packet */
352
353                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
354                         if (!skb)
355                                 PRINTR("nfnetlink_log: can't even alloc %u "
356                                        "bytes\n", pkt_size);
357                 }
358         }
359
360         return skb;
361 }
362
363 static int
364 __nfulnl_send(struct nfulnl_instance *inst)
365 {
366         int status;
367
368         if (!inst->skb)
369                 return 0;
370
371         if (inst->qlen > 1)
372                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
373
374         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
375         if (status < 0) {
376                 UDEBUG("netlink_unicast() failed\n");
377                 /* FIXME: statistics */
378         }
379
380         inst->qlen = 0;
381         inst->skb = NULL;
382         inst->lastnlh = NULL;
383
384         return status;
385 }
386
387 static void nfulnl_timer(unsigned long data)
388 {
389         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
390
391         UDEBUG("timer function called, flushing buffer\n");
392
393         spin_lock_bh(&inst->lock);
394         __nfulnl_send(inst);
395         spin_unlock_bh(&inst->lock);
396         instance_put(inst);
397 }
398
399 /* This is an inline function, we don't really care about a long
400  * list of arguments */
401 static inline int
402 __build_packet_message(struct nfulnl_instance *inst,
403                         const struct sk_buff *skb,
404                         unsigned int data_len,
405                         unsigned int pf,
406                         unsigned int hooknum,
407                         const struct net_device *indev,
408                         const struct net_device *outdev,
409                         const struct nf_loginfo *li,
410                         const char *prefix, unsigned int plen)
411 {
412         unsigned char *old_tail;
413         struct nfulnl_msg_packet_hdr pmsg;
414         struct nlmsghdr *nlh;
415         struct nfgenmsg *nfmsg;
416         __be32 tmp_uint;
417
418         UDEBUG("entered\n");
419
420         old_tail = inst->skb->tail;
421         nlh = NLMSG_PUT(inst->skb, 0, 0,
422                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
423                         sizeof(struct nfgenmsg));
424         nfmsg = NLMSG_DATA(nlh);
425         nfmsg->nfgen_family = pf;
426         nfmsg->version = NFNETLINK_V0;
427         nfmsg->res_id = htons(inst->group_num);
428
429         pmsg.hw_protocol        = skb->protocol;
430         pmsg.hook               = hooknum;
431
432         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
433
434         if (prefix)
435                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
436
437         if (indev) {
438                 tmp_uint = htonl(indev->ifindex);
439 #ifndef CONFIG_BRIDGE_NETFILTER
440                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
441                         &tmp_uint);
442 #else
443                 if (pf == PF_BRIDGE) {
444                         /* Case 1: outdev is physical input device, we need to
445                          * look for bridge group (when called from
446                          * netfilter_bridge) */
447                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
448                                 sizeof(tmp_uint), &tmp_uint);
449                         /* this is the bridge group "brX" */
450                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
451                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
452                                 sizeof(tmp_uint), &tmp_uint);
453                 } else {
454                         /* Case 2: indev is bridge group, we need to look for
455                          * physical device (when called from ipv4) */
456                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
457                                 sizeof(tmp_uint), &tmp_uint);
458                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
459                                 tmp_uint =
460                                     htonl(skb->nf_bridge->physindev->ifindex);
461                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
462                                         sizeof(tmp_uint), &tmp_uint);
463                         }
464                 }
465 #endif
466         }
467
468         if (outdev) {
469                 tmp_uint = htonl(outdev->ifindex);
470 #ifndef CONFIG_BRIDGE_NETFILTER
471                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
472                         &tmp_uint);
473 #else
474                 if (pf == PF_BRIDGE) {
475                         /* Case 1: outdev is physical output device, we need to
476                          * look for bridge group (when called from
477                          * netfilter_bridge) */
478                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
479                                 sizeof(tmp_uint), &tmp_uint);
480                         /* this is the bridge group "brX" */
481                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
482                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
483                                 sizeof(tmp_uint), &tmp_uint);
484                 } else {
485                         /* Case 2: indev is a bridge group, we need to look
486                          * for physical device (when called from ipv4) */
487                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
488                                 sizeof(tmp_uint), &tmp_uint);
489                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
490                                 tmp_uint =
491                                     htonl(skb->nf_bridge->physoutdev->ifindex);
492                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
493                                         sizeof(tmp_uint), &tmp_uint);
494                         }
495                 }
496 #endif
497         }
498
499         if (skb->mark) {
500                 tmp_uint = htonl(skb->mark);
501                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
502         }
503
504         if (indev && skb->dev && skb->dev->hard_header_parse) {
505                 struct nfulnl_msg_packet_hw phw;
506                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
507                                                     phw.hw_addr);
508                 phw.hw_addrlen = htons(len);
509                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
510         }
511
512         if (skb->tstamp.off_sec) {
513                 struct nfulnl_msg_packet_timestamp ts;
514
515                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
516                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
517
518                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
519         }
520
521         /* UID */
522         if (skb->sk) {
523                 read_lock_bh(&skb->sk->sk_callback_lock);
524                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
525                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
526                         /* need to unlock here since NFA_PUT may goto */
527                         read_unlock_bh(&skb->sk->sk_callback_lock);
528                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
529                 } else
530                         read_unlock_bh(&skb->sk->sk_callback_lock);
531         }
532
533         /* local sequence number */
534         if (inst->flags & NFULNL_CFG_F_SEQ) {
535                 tmp_uint = htonl(inst->seq++);
536                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
537         }
538         /* global sequence number */
539         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
540                 tmp_uint = htonl(atomic_inc_return(&global_seq));
541                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
542         }
543
544         if (data_len) {
545                 struct nfattr *nfa;
546                 int size = NFA_LENGTH(data_len);
547
548                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
549                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
550                         goto nlmsg_failure;
551                 }
552
553                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
554                 nfa->nfa_type = NFULA_PAYLOAD;
555                 nfa->nfa_len = size;
556
557                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
558                         BUG();
559         }
560
561         nlh->nlmsg_len = inst->skb->tail - old_tail;
562         inst->lastnlh = nlh;
563         return 0;
564
565 nlmsg_failure:
566         UDEBUG("nlmsg_failure\n");
567 nfattr_failure:
568         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
569         return -1;
570 }
571
572 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
573
574 static struct nf_loginfo default_loginfo = {
575         .type =         NF_LOG_TYPE_ULOG,
576         .u = {
577                 .ulog = {
578                         .copy_len       = 0xffff,
579                         .group          = 0,
580                         .qthreshold     = 1,
581                 },
582         },
583 };
584
585 /* log handler for internal netfilter logging api */
586 static void
587 nfulnl_log_packet(unsigned int pf,
588                   unsigned int hooknum,
589                   const struct sk_buff *skb,
590                   const struct net_device *in,
591                   const struct net_device *out,
592                   const struct nf_loginfo *li_user,
593                   const char *prefix)
594 {
595         unsigned int size, data_len;
596         struct nfulnl_instance *inst;
597         const struct nf_loginfo *li;
598         unsigned int qthreshold;
599         unsigned int nlbufsiz;
600         unsigned int plen;
601
602         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
603                 li = li_user;
604         else
605                 li = &default_loginfo;
606
607         inst = instance_lookup_get(li->u.ulog.group);
608         if (!inst)
609                 inst = instance_lookup_get(0);
610         if (!inst) {
611                 PRINTR("nfnetlink_log: trying to log packet, "
612                         "but no instance for group %u\n", li->u.ulog.group);
613                 return;
614         }
615
616         plen = 0;
617         if (prefix)
618                 plen = strlen(prefix) + 1;
619
620         /* all macros expand to constant values at compile time */
621         /* FIXME: do we want to make the size calculation conditional based on
622          * what is actually present?  way more branches and checks, but more
623          * memory efficient... */
624         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
625                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
626                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
627                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
628 #ifdef CONFIG_BRIDGE_NETFILTER
629                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
630                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
631 #endif
632                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
633                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
634                 + NFA_SPACE(plen)               /* prefix */
635                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
636                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
637
638         UDEBUG("initial size=%u\n", size);
639
640         spin_lock_bh(&inst->lock);
641
642         if (inst->flags & NFULNL_CFG_F_SEQ)
643                 size += NFA_SPACE(sizeof(u_int32_t));
644         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
645                 size += NFA_SPACE(sizeof(u_int32_t));
646
647         qthreshold = inst->qthreshold;
648         /* per-rule qthreshold overrides per-instance */
649         if (qthreshold > li->u.ulog.qthreshold)
650                 qthreshold = li->u.ulog.qthreshold;
651
652         switch (inst->copy_mode) {
653         case NFULNL_COPY_META:
654         case NFULNL_COPY_NONE:
655                 data_len = 0;
656                 break;
657
658         case NFULNL_COPY_PACKET:
659                 if (inst->copy_range == 0
660                     || inst->copy_range > skb->len)
661                         data_len = skb->len;
662                 else
663                         data_len = inst->copy_range;
664
665                 size += NFA_SPACE(data_len);
666                 UDEBUG("copy_packet, therefore size now %u\n", size);
667                 break;
668
669         default:
670                 spin_unlock_bh(&inst->lock);
671                 instance_put(inst);
672                 return;
673         }
674
675         if (size > inst->nlbufsiz)
676                 nlbufsiz = size;
677         else
678                 nlbufsiz = inst->nlbufsiz;
679
680         if (!inst->skb) {
681                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
682                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
683                                 inst->nlbufsiz, size);
684                         goto alloc_failure;
685                 }
686         } else if (inst->qlen >= qthreshold ||
687                    size > skb_tailroom(inst->skb)) {
688                 /* either the queue len is too high or we don't have
689                  * enough room in the skb left. flush to userspace. */
690                 UDEBUG("flushing old skb\n");
691
692                 /* timer "holds" one reference (we have another one) */
693                 if (del_timer(&inst->timer))
694                         instance_put(inst);
695                 __nfulnl_send(inst);
696
697                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
698                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
699                                 inst->nlbufsiz, size);
700                         goto alloc_failure;
701                 }
702         }
703
704         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
705         inst->qlen++;
706
707         __build_packet_message(inst, skb, data_len, pf,
708                                 hooknum, in, out, li, prefix, plen);
709
710         /* timer_pending always called within inst->lock, so there
711          * is no chance of a race here */
712         if (!timer_pending(&inst->timer)) {
713                 instance_get(inst);
714                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
715                 add_timer(&inst->timer);
716         }
717
718 unlock_and_release:
719         spin_unlock_bh(&inst->lock);
720         instance_put(inst);
721         return;
722
723 alloc_failure:
724         UDEBUG("error allocating skb\n");
725         /* FIXME: statistics */
726         goto unlock_and_release;
727 }
728
729 static int
730 nfulnl_rcv_nl_event(struct notifier_block *this,
731                    unsigned long event, void *ptr)
732 {
733         struct netlink_notify *n = ptr;
734
735         if (event == NETLINK_URELEASE &&
736             n->protocol == NETLINK_NETFILTER && n->pid) {
737                 int i;
738
739                 /* destroy all instances for this pid */
740                 write_lock_bh(&instances_lock);
741                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
742                         struct hlist_node *tmp, *t2;
743                         struct nfulnl_instance *inst;
744                         struct hlist_head *head = &instance_table[i];
745
746                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
747                                 UDEBUG("node = %p\n", inst);
748                                 if (n->pid == inst->peer_pid)
749                                         __instance_destroy(inst);
750                         }
751                 }
752                 write_unlock_bh(&instances_lock);
753         }
754         return NOTIFY_DONE;
755 }
756
757 static struct notifier_block nfulnl_rtnl_notifier = {
758         .notifier_call  = nfulnl_rcv_nl_event,
759 };
760
761 static int
762 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
763                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
764 {
765         return -ENOTSUPP;
766 }
767
768 static struct nf_logger nfulnl_logger = {
769         .name   = "nfnetlink_log",
770         .logfn  = &nfulnl_log_packet,
771         .me     = THIS_MODULE,
772 };
773
774 static const int nfula_min[NFULA_MAX] = {
775         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
776         [NFULA_MARK-1]          = sizeof(u_int32_t),
777         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
778         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
779         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
780         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
781         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
782         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
783         [NFULA_PAYLOAD-1]       = 0,
784         [NFULA_PREFIX-1]        = 0,
785         [NFULA_UID-1]           = sizeof(u_int32_t),
786         [NFULA_SEQ-1]           = sizeof(u_int32_t),
787         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
788 };
789
790 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
791         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
792         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
793         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
794         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
795         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
796         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
797 };
798
799 static int
800 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
801                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
802 {
803         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
804         u_int16_t group_num = ntohs(nfmsg->res_id);
805         struct nfulnl_instance *inst;
806         int ret = 0;
807
808         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
809
810         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
811                 UDEBUG("bad attribute size\n");
812                 return -EINVAL;
813         }
814
815         inst = instance_lookup_get(group_num);
816         if (nfula[NFULA_CFG_CMD-1]) {
817                 u_int8_t pf = nfmsg->nfgen_family;
818                 struct nfulnl_msg_config_cmd *cmd;
819                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
820                 UDEBUG("found CFG_CMD for\n");
821
822                 switch (cmd->command) {
823                 case NFULNL_CFG_CMD_BIND:
824                         if (inst) {
825                                 ret = -EBUSY;
826                                 goto out_put;
827                         }
828
829                         inst = instance_create(group_num,
830                                                NETLINK_CB(skb).pid);
831                         if (!inst) {
832                                 ret = -EINVAL;
833                                 goto out_put;
834                         }
835                         break;
836                 case NFULNL_CFG_CMD_UNBIND:
837                         if (!inst) {
838                                 ret = -ENODEV;
839                                 goto out_put;
840                         }
841
842                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
843                                 ret = -EPERM;
844                                 goto out_put;
845                         }
846
847                         instance_destroy(inst);
848                         break;
849                 case NFULNL_CFG_CMD_PF_BIND:
850                         UDEBUG("registering log handler for pf=%u\n", pf);
851                         ret = nf_log_register(pf, &nfulnl_logger);
852                         break;
853                 case NFULNL_CFG_CMD_PF_UNBIND:
854                         UDEBUG("unregistering log handler for pf=%u\n", pf);
855                         /* This is a bug and a feature.  We cannot unregister
856                          * other handlers, like nfnetlink_inst can */
857                         nf_log_unregister_pf(pf);
858                         break;
859                 default:
860                         ret = -EINVAL;
861                         break;
862                 }
863
864                 if (!inst)
865                         goto out;
866         } else {
867                 if (!inst) {
868                         UDEBUG("no config command, and no instance for "
869                                 "group=%u pid=%u =>ENOENT\n",
870                                 group_num, NETLINK_CB(skb).pid);
871                         ret = -ENOENT;
872                         goto out_put;
873                 }
874
875                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
876                         UDEBUG("no config command, and wrong pid\n");
877                         ret = -EPERM;
878                         goto out_put;
879                 }
880         }
881
882         if (nfula[NFULA_CFG_MODE-1]) {
883                 struct nfulnl_msg_config_mode *params;
884                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
885
886                 nfulnl_set_mode(inst, params->copy_mode,
887                                 ntohl(params->copy_range));
888         }
889
890         if (nfula[NFULA_CFG_TIMEOUT-1]) {
891                 __be32 timeout =
892                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
893
894                 nfulnl_set_timeout(inst, ntohl(timeout));
895         }
896
897         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
898                 __be32 nlbufsiz =
899                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
900
901                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
902         }
903
904         if (nfula[NFULA_CFG_QTHRESH-1]) {
905                 __be32 qthresh =
906                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
907
908                 nfulnl_set_qthresh(inst, ntohl(qthresh));
909         }
910
911         if (nfula[NFULA_CFG_FLAGS-1]) {
912                 __be16 flags =
913                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
914                 nfulnl_set_flags(inst, ntohs(flags));
915         }
916
917 out_put:
918         instance_put(inst);
919 out:
920         return ret;
921 }
922
923 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
924         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
925                                     .attr_count = NFULA_MAX, },
926         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
927                                     .attr_count = NFULA_CFG_MAX, },
928 };
929
930 static struct nfnetlink_subsystem nfulnl_subsys = {
931         .name           = "log",
932         .subsys_id      = NFNL_SUBSYS_ULOG,
933         .cb_count       = NFULNL_MSG_MAX,
934         .cb             = nfulnl_cb,
935 };
936
937 #ifdef CONFIG_PROC_FS
938 struct iter_state {
939         unsigned int bucket;
940 };
941
942 static struct hlist_node *get_first(struct seq_file *seq)
943 {
944         struct iter_state *st = seq->private;
945
946         if (!st)
947                 return NULL;
948
949         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
950                 if (!hlist_empty(&instance_table[st->bucket]))
951                         return instance_table[st->bucket].first;
952         }
953         return NULL;
954 }
955
956 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
957 {
958         struct iter_state *st = seq->private;
959
960         h = h->next;
961         while (!h) {
962                 if (++st->bucket >= INSTANCE_BUCKETS)
963                         return NULL;
964
965                 h = instance_table[st->bucket].first;
966         }
967         return h;
968 }
969
970 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
971 {
972         struct hlist_node *head;
973         head = get_first(seq);
974
975         if (head)
976                 while (pos && (head = get_next(seq, head)))
977                         pos--;
978         return pos ? NULL : head;
979 }
980
981 static void *seq_start(struct seq_file *seq, loff_t *pos)
982 {
983         read_lock_bh(&instances_lock);
984         return get_idx(seq, *pos);
985 }
986
987 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
988 {
989         (*pos)++;
990         return get_next(s, v);
991 }
992
993 static void seq_stop(struct seq_file *s, void *v)
994 {
995         read_unlock_bh(&instances_lock);
996 }
997
998 static int seq_show(struct seq_file *s, void *v)
999 {
1000         const struct nfulnl_instance *inst = v;
1001
1002         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
1003                           inst->group_num,
1004                           inst->peer_pid, inst->qlen,
1005                           inst->copy_mode, inst->copy_range,
1006                           inst->flushtimeout, atomic_read(&inst->use));
1007 }
1008
1009 static struct seq_operations nful_seq_ops = {
1010         .start  = seq_start,
1011         .next   = seq_next,
1012         .stop   = seq_stop,
1013         .show   = seq_show,
1014 };
1015
1016 static int nful_open(struct inode *inode, struct file *file)
1017 {
1018         struct seq_file *seq;
1019         struct iter_state *is;
1020         int ret;
1021
1022         is = kzalloc(sizeof(*is), GFP_KERNEL);
1023         if (!is)
1024                 return -ENOMEM;
1025         ret = seq_open(file, &nful_seq_ops);
1026         if (ret < 0)
1027                 goto out_free;
1028         seq = file->private_data;
1029         seq->private = is;
1030         return ret;
1031 out_free:
1032         kfree(is);
1033         return ret;
1034 }
1035
1036 static const struct file_operations nful_file_ops = {
1037         .owner   = THIS_MODULE,
1038         .open    = nful_open,
1039         .read    = seq_read,
1040         .llseek  = seq_lseek,
1041         .release = seq_release_private,
1042 };
1043
1044 #endif /* PROC_FS */
1045
1046 static int __init nfnetlink_log_init(void)
1047 {
1048         int i, status = -ENOMEM;
1049 #ifdef CONFIG_PROC_FS
1050         struct proc_dir_entry *proc_nful;
1051 #endif
1052
1053         for (i = 0; i < INSTANCE_BUCKETS; i++)
1054                 INIT_HLIST_HEAD(&instance_table[i]);
1055
1056         /* it's not really all that important to have a random value, so
1057          * we can do this from the init function, even if there hasn't
1058          * been that much entropy yet */
1059         get_random_bytes(&hash_init, sizeof(hash_init));
1060
1061         netlink_register_notifier(&nfulnl_rtnl_notifier);
1062         status = nfnetlink_subsys_register(&nfulnl_subsys);
1063         if (status < 0) {
1064                 printk(KERN_ERR "log: failed to create netlink socket\n");
1065                 goto cleanup_netlink_notifier;
1066         }
1067
1068 #ifdef CONFIG_PROC_FS
1069         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1070                                       proc_net_netfilter);
1071         if (!proc_nful)
1072                 goto cleanup_subsys;
1073         proc_nful->proc_fops = &nful_file_ops;
1074 #endif
1075         return status;
1076
1077 #ifdef CONFIG_PROC_FS
1078 cleanup_subsys:
1079         nfnetlink_subsys_unregister(&nfulnl_subsys);
1080 #endif
1081 cleanup_netlink_notifier:
1082         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1083         return status;
1084 }
1085
1086 static void __exit nfnetlink_log_fini(void)
1087 {
1088         nf_log_unregister(&nfulnl_logger);
1089 #ifdef CONFIG_PROC_FS
1090         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1091 #endif
1092         nfnetlink_subsys_unregister(&nfulnl_subsys);
1093         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1094 }
1095
1096 MODULE_DESCRIPTION("netfilter userspace logging");
1097 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1098 MODULE_LICENSE("GPL");
1099 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1100
1101 module_init(nfnetlink_log_init);
1102 module_exit(nfnetlink_log_fini);