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