Merge branch 'upstream-fixes' of git://lost.foo-projects.org/~ahkok/git/netdev-2...
[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->skb)
370                 return 0;
371
372         if (inst->qlen > 1)
373                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
374
375         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
376         if (status < 0) {
377                 UDEBUG("netlink_unicast() failed\n");
378                 /* FIXME: statistics */
379         }
380
381         inst->qlen = 0;
382         inst->skb = NULL;
383         inst->lastnlh = NULL;
384
385         return status;
386 }
387
388 static void nfulnl_timer(unsigned long data)
389 {
390         struct nfulnl_instance *inst = (struct nfulnl_instance *)data; 
391
392         UDEBUG("timer function called, flushing buffer\n");
393
394         spin_lock_bh(&inst->lock);
395         __nfulnl_send(inst);
396         instance_put(inst);
397         spin_unlock_bh(&inst->lock);
398 }
399
400 /* This is an inline function, we don't really care about a long
401  * list of arguments */
402 static inline int 
403 __build_packet_message(struct nfulnl_instance *inst,
404                         const struct sk_buff *skb, 
405                         unsigned int data_len,
406                         unsigned int pf,
407                         unsigned int hooknum,
408                         const struct net_device *indev,
409                         const struct net_device *outdev,
410                         const struct nf_loginfo *li,
411                         const char *prefix)
412 {
413         unsigned char *old_tail;
414         struct nfulnl_msg_packet_hdr pmsg;
415         struct nlmsghdr *nlh;
416         struct nfgenmsg *nfmsg;
417         u_int32_t tmp_uint;
418
419         UDEBUG("entered\n");
420                 
421         old_tail = inst->skb->tail;
422         nlh = NLMSG_PUT(inst->skb, 0, 0, 
423                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
424                         sizeof(struct nfgenmsg));
425         nfmsg = NLMSG_DATA(nlh);
426         nfmsg->nfgen_family = pf;
427         nfmsg->version = NFNETLINK_V0;
428         nfmsg->res_id = htons(inst->group_num);
429
430         pmsg.hw_protocol        = htons(skb->protocol);
431         pmsg.hook               = hooknum;
432
433         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
434
435         if (prefix) {
436                 int slen = strlen(prefix);
437                 if (slen > NFULNL_PREFIXLEN)
438                         slen = NFULNL_PREFIXLEN;
439                 NFA_PUT(inst->skb, NFULA_PREFIX, slen, prefix);
440         }
441
442         if (indev) {
443                 tmp_uint = htonl(indev->ifindex);
444 #ifndef CONFIG_BRIDGE_NETFILTER
445                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
446                         &tmp_uint);
447 #else
448                 if (pf == PF_BRIDGE) {
449                         /* Case 1: outdev is physical input device, we need to
450                          * look for bridge group (when called from
451                          * netfilter_bridge) */
452                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
453                                 sizeof(tmp_uint), &tmp_uint);
454                         /* this is the bridge group "brX" */
455                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
456                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
457                                 sizeof(tmp_uint), &tmp_uint);
458                 } else {
459                         /* Case 2: indev is bridge group, we need to look for
460                          * physical device (when called from ipv4) */
461                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
462                                 sizeof(tmp_uint), &tmp_uint);
463                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
464                                 tmp_uint = 
465                                     htonl(skb->nf_bridge->physindev->ifindex);
466                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
467                                         sizeof(tmp_uint), &tmp_uint);
468                         }
469                 }
470 #endif
471         }
472
473         if (outdev) {
474                 tmp_uint = htonl(outdev->ifindex);
475 #ifndef CONFIG_BRIDGE_NETFILTER
476                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
477                         &tmp_uint);
478 #else
479                 if (pf == PF_BRIDGE) {
480                         /* Case 1: outdev is physical output device, we need to
481                          * look for bridge group (when called from
482                          * netfilter_bridge) */
483                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
484                                 sizeof(tmp_uint), &tmp_uint);
485                         /* this is the bridge group "brX" */
486                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
487                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
488                                 sizeof(tmp_uint), &tmp_uint);
489                 } else {
490                         /* Case 2: indev is a bridge group, we need to look
491                          * for physical device (when called from ipv4) */
492                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
493                                 sizeof(tmp_uint), &tmp_uint);
494                         if (skb->nf_bridge) {
495                                 tmp_uint = 
496                                     htonl(skb->nf_bridge->physoutdev->ifindex);
497                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
498                                         sizeof(tmp_uint), &tmp_uint);
499                         }
500                 }
501 #endif
502         }
503
504         if (skb->nfmark) {
505                 tmp_uint = htonl(skb->nfmark);
506                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
507         }
508
509         if (indev && skb->dev && skb->dev->hard_header_parse) {
510                 struct nfulnl_msg_packet_hw phw;
511
512                 phw.hw_addrlen = 
513                         skb->dev->hard_header_parse((struct sk_buff *)skb, 
514                                                     phw.hw_addr);
515                 phw.hw_addrlen = htons(phw.hw_addrlen);
516                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
517         }
518
519         if (skb->tstamp.off_sec) {
520                 struct nfulnl_msg_packet_timestamp ts;
521
522                 ts.sec = cpu_to_be64(skb->tstamp.off_sec);
523                 ts.usec = cpu_to_be64(skb->tstamp.off_usec);
524
525                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
526         }
527
528         /* UID */
529         if (skb->sk) {
530                 read_lock_bh(&skb->sk->sk_callback_lock);
531                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
532                         u_int32_t uid = htonl(skb->sk->sk_socket->file->f_uid);
533                         /* need to unlock here since NFA_PUT may goto */
534                         read_unlock_bh(&skb->sk->sk_callback_lock);
535                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
536                 } else
537                         read_unlock_bh(&skb->sk->sk_callback_lock);
538         }
539
540         /* local sequence number */
541         if (inst->flags & NFULNL_CFG_F_SEQ) {
542                 tmp_uint = htonl(inst->seq++);
543                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
544         }
545         /* global sequence number */
546         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
547                 tmp_uint = atomic_inc_return(&global_seq);
548                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
549         }
550
551         if (data_len) {
552                 struct nfattr *nfa;
553                 int size = NFA_LENGTH(data_len);
554
555                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
556                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
557                         goto nlmsg_failure;
558                 }
559
560                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
561                 nfa->nfa_type = NFULA_PAYLOAD;
562                 nfa->nfa_len = size;
563
564                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
565                         BUG();
566         }
567                 
568         nlh->nlmsg_len = inst->skb->tail - old_tail;
569         return 0;
570
571 nlmsg_failure:
572         UDEBUG("nlmsg_failure\n");
573 nfattr_failure:
574         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
575         return -1;
576 }
577
578 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
579
580 static struct nf_loginfo default_loginfo = {
581         .type =         NF_LOG_TYPE_ULOG,
582         .u = {
583                 .ulog = {
584                         .copy_len       = 0xffff,
585                         .group          = 0,
586                         .qthreshold     = 1,
587                 },
588         },
589 };
590
591 /* log handler for internal netfilter logging api */
592 static void
593 nfulnl_log_packet(unsigned int pf,
594                   unsigned int hooknum,
595                   const struct sk_buff *skb,
596                   const struct net_device *in,
597                   const struct net_device *out,
598                   const struct nf_loginfo *li_user,
599                   const char *prefix)
600 {
601         unsigned int size, data_len;
602         struct nfulnl_instance *inst;
603         const struct nf_loginfo *li;
604         unsigned int qthreshold;
605         unsigned int nlbufsiz;
606
607         if (li_user && li_user->type == NF_LOG_TYPE_ULOG) 
608                 li = li_user;
609         else
610                 li = &default_loginfo;
611
612         inst = instance_lookup_get(li->u.ulog.group);
613         if (!inst)
614                 inst = instance_lookup_get(0);
615         if (!inst) {
616                 PRINTR("nfnetlink_log: trying to log packet, "
617                         "but no instance for group %u\n", li->u.ulog.group);
618                 return;
619         }
620
621         /* all macros expand to constant values at compile time */
622         /* FIXME: do we want to make the size calculation conditional based on
623          * what is actually present?  way more branches and checks, but more
624          * memory efficient... */
625         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
626                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
627                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
628                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
629 #ifdef CONFIG_BRIDGE_NETFILTER
630                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
631                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
632 #endif
633                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
634                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
635                 + NFA_SPACE(NFULNL_PREFIXLEN)   /* prefix */
636                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
637                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
638
639         UDEBUG("initial size=%u\n", size);
640
641         spin_lock_bh(&inst->lock);
642
643         if (inst->flags & NFULNL_CFG_F_SEQ)
644                 size += NFA_SPACE(sizeof(u_int32_t));
645         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
646                 size += NFA_SPACE(sizeof(u_int32_t));
647
648         qthreshold = inst->qthreshold;
649         /* per-rule qthreshold overrides per-instance */
650         if (qthreshold > li->u.ulog.qthreshold)
651                 qthreshold = li->u.ulog.qthreshold;
652         
653         switch (inst->copy_mode) {
654         case NFULNL_COPY_META:
655         case NFULNL_COPY_NONE:
656                 data_len = 0;
657                 break;
658         
659         case NFULNL_COPY_PACKET:
660                 if (inst->copy_range == 0 
661                     || inst->copy_range > skb->len)
662                         data_len = skb->len;
663                 else
664                         data_len = inst->copy_range;
665                 
666                 size += NFA_SPACE(data_len);
667                 UDEBUG("copy_packet, therefore size now %u\n", size);
668                 break;
669         
670         default:
671                 spin_unlock_bh(&inst->lock);
672                 instance_put(inst);
673                 return;
674         }
675
676         if (size > inst->nlbufsiz)
677                 nlbufsiz = size;
678         else
679                 nlbufsiz = inst->nlbufsiz;
680
681         if (!inst->skb) {
682                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
683                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
684                                 inst->nlbufsiz, size);
685                         goto alloc_failure;
686                 }
687         } else if (inst->qlen >= qthreshold ||
688                    size > skb_tailroom(inst->skb)) {
689                 /* either the queue len is too high or we don't have
690                  * enough room in the skb left. flush to userspace. */
691                 UDEBUG("flushing old skb\n");
692
693                 __nfulnl_send(inst);
694
695                 if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) {
696                         UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n",
697                                 inst->nlbufsiz, size);
698                         goto alloc_failure;
699                 }
700         }
701
702         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
703         inst->qlen++;
704
705         __build_packet_message(inst, skb, data_len, pf,
706                                 hooknum, in, out, li, prefix);
707
708         /* timer_pending always called within inst->lock, so there
709          * is no chance of a race here */
710         if (!timer_pending(&inst->timer)) {
711                 instance_get(inst);
712                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
713                 add_timer(&inst->timer);
714         }
715         spin_unlock_bh(&inst->lock);
716
717         return;
718
719 alloc_failure:
720         spin_unlock_bh(&inst->lock);
721         instance_put(inst);
722         UDEBUG("error allocating skb\n");
723         /* FIXME: statistics */
724 }
725
726 static int
727 nfulnl_rcv_nl_event(struct notifier_block *this,
728                    unsigned long event, void *ptr)
729 {
730         struct netlink_notify *n = ptr;
731
732         if (event == NETLINK_URELEASE &&
733             n->protocol == NETLINK_NETFILTER && n->pid) {
734                 int i;
735
736                 /* destroy all instances for this pid */
737                 write_lock_bh(&instances_lock);
738                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
739                         struct hlist_node *tmp, *t2;
740                         struct nfulnl_instance *inst;
741                         struct hlist_head *head = &instance_table[i];
742
743                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
744                                 UDEBUG("node = %p\n", inst);
745                                 if (n->pid == inst->peer_pid)
746                                         __instance_destroy(inst);
747                         }
748                 }
749                 write_unlock_bh(&instances_lock);
750         }
751         return NOTIFY_DONE;
752 }
753
754 static struct notifier_block nfulnl_rtnl_notifier = {
755         .notifier_call  = nfulnl_rcv_nl_event,
756 };
757
758 static int
759 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
760                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
761 {
762         return -ENOTSUPP;
763 }
764
765 static struct nf_logger nfulnl_logger = {
766         .name   = "nfnetlink_log",
767         .logfn  = &nfulnl_log_packet,
768         .me     = THIS_MODULE,
769 };
770
771 static const int nfula_min[NFULA_MAX] = {
772         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
773         [NFULA_MARK-1]          = sizeof(u_int32_t),
774         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
775         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
776         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
777         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
778         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
779         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
780         [NFULA_PAYLOAD-1]       = 0,
781         [NFULA_PREFIX-1]        = 0,
782         [NFULA_UID-1]           = sizeof(u_int32_t),
783         [NFULA_SEQ-1]           = sizeof(u_int32_t),
784         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
785 };
786
787 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
788         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
789         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
790         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
791         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
792         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
793         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
794 };
795
796 static int
797 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
798                    struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp)
799 {
800         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
801         u_int16_t group_num = ntohs(nfmsg->res_id);
802         struct nfulnl_instance *inst;
803         int ret = 0;
804
805         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
806
807         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
808                 UDEBUG("bad attribute size\n");
809                 return -EINVAL;
810         }
811
812         inst = instance_lookup_get(group_num);
813         if (nfula[NFULA_CFG_CMD-1]) {
814                 u_int8_t pf = nfmsg->nfgen_family;
815                 struct nfulnl_msg_config_cmd *cmd;
816                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
817                 UDEBUG("found CFG_CMD for\n");
818
819                 switch (cmd->command) {
820                 case NFULNL_CFG_CMD_BIND:
821                         if (inst) {
822                                 ret = -EBUSY;
823                                 goto out_put;
824                         }
825
826                         inst = instance_create(group_num,
827                                                NETLINK_CB(skb).pid);
828                         if (!inst) {
829                                 ret = -EINVAL;
830                                 goto out_put;
831                         }
832                         break;
833                 case NFULNL_CFG_CMD_UNBIND:
834                         if (!inst) {
835                                 ret = -ENODEV;
836                                 goto out_put;
837                         }
838
839                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
840                                 ret = -EPERM;
841                                 goto out_put;
842                         }
843
844                         instance_destroy(inst);
845                         break;
846                 case NFULNL_CFG_CMD_PF_BIND:
847                         UDEBUG("registering log handler for pf=%u\n", pf);
848                         ret = nf_log_register(pf, &nfulnl_logger);
849                         break;
850                 case NFULNL_CFG_CMD_PF_UNBIND:
851                         UDEBUG("unregistering log handler for pf=%u\n", pf);
852                         /* This is a bug and a feature.  We cannot unregister
853                          * other handlers, like nfnetlink_inst can */
854                         nf_log_unregister_pf(pf);
855                         break;
856                 default:
857                         ret = -EINVAL;
858                         break;
859                 }
860         } else {
861                 if (!inst) {
862                         UDEBUG("no config command, and no instance for "
863                                 "group=%u pid=%u =>ENOENT\n",
864                                 group_num, NETLINK_CB(skb).pid);
865                         ret = -ENOENT;
866                         goto out_put;
867                 }
868
869                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
870                         UDEBUG("no config command, and wrong pid\n");
871                         ret = -EPERM;
872                         goto out_put;
873                 }
874         }
875
876         if (nfula[NFULA_CFG_MODE-1]) {
877                 struct nfulnl_msg_config_mode *params;
878                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
879
880                 nfulnl_set_mode(inst, params->copy_mode,
881                                 ntohs(params->copy_range));
882         }
883
884         if (nfula[NFULA_CFG_TIMEOUT-1]) {
885                 u_int32_t timeout = 
886                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
887
888                 nfulnl_set_timeout(inst, ntohl(timeout));
889         }
890
891         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
892                 u_int32_t nlbufsiz = 
893                         *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
894
895                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
896         }
897
898         if (nfula[NFULA_CFG_QTHRESH-1]) {
899                 u_int32_t qthresh = 
900                         *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
901
902                 nfulnl_set_qthresh(inst, ntohl(qthresh));
903         }
904
905         if (nfula[NFULA_CFG_FLAGS-1]) {
906                 u_int16_t flags =
907                         *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
908                 nfulnl_set_flags(inst, ntohs(flags));
909         }
910
911 out_put:
912         instance_put(inst);
913         return ret;
914 }
915
916 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
917         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
918                                     .attr_count = NFULA_MAX, },
919         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
920                                     .attr_count = NFULA_CFG_MAX, },
921 };
922
923 static struct nfnetlink_subsystem nfulnl_subsys = {
924         .name           = "log",
925         .subsys_id      = NFNL_SUBSYS_ULOG,
926         .cb_count       = NFULNL_MSG_MAX,
927         .cb             = nfulnl_cb,
928 };
929
930 #ifdef CONFIG_PROC_FS
931 struct iter_state {
932         unsigned int bucket;
933 };
934
935 static struct hlist_node *get_first(struct seq_file *seq)
936 {
937         struct iter_state *st = seq->private;
938
939         if (!st)
940                 return NULL;
941
942         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
943                 if (!hlist_empty(&instance_table[st->bucket]))
944                         return instance_table[st->bucket].first;
945         }
946         return NULL;
947 }
948
949 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
950 {
951         struct iter_state *st = seq->private;
952
953         h = h->next;
954         while (!h) {
955                 if (++st->bucket >= INSTANCE_BUCKETS)
956                         return NULL;
957
958                 h = instance_table[st->bucket].first;
959         }
960         return h;
961 }
962
963 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
964 {
965         struct hlist_node *head;
966         head = get_first(seq);
967
968         if (head)
969                 while (pos && (head = get_next(seq, head)))
970                         pos--;
971         return pos ? NULL : head;
972 }
973
974 static void *seq_start(struct seq_file *seq, loff_t *pos)
975 {
976         read_lock_bh(&instances_lock);
977         return get_idx(seq, *pos);
978 }
979
980 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
981 {
982         (*pos)++;
983         return get_next(s, v);
984 }
985
986 static void seq_stop(struct seq_file *s, void *v)
987 {
988         read_unlock_bh(&instances_lock);
989 }
990
991 static int seq_show(struct seq_file *s, void *v)
992 {
993         const struct nfulnl_instance *inst = v;
994
995         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n", 
996                           inst->group_num,
997                           inst->peer_pid, inst->qlen, 
998                           inst->copy_mode, inst->copy_range,
999                           inst->flushtimeout, atomic_read(&inst->use));
1000 }
1001
1002 static struct seq_operations nful_seq_ops = {
1003         .start  = seq_start,
1004         .next   = seq_next,
1005         .stop   = seq_stop,
1006         .show   = seq_show,
1007 };
1008
1009 static int nful_open(struct inode *inode, struct file *file)
1010 {
1011         struct seq_file *seq;
1012         struct iter_state *is;
1013         int ret;
1014
1015         is = kzalloc(sizeof(*is), GFP_KERNEL);
1016         if (!is)
1017                 return -ENOMEM;
1018         ret = seq_open(file, &nful_seq_ops);
1019         if (ret < 0)
1020                 goto out_free;
1021         seq = file->private_data;
1022         seq->private = is;
1023         return ret;
1024 out_free:
1025         kfree(is);
1026         return ret;
1027 }
1028
1029 static struct file_operations nful_file_ops = {
1030         .owner   = THIS_MODULE,
1031         .open    = nful_open,
1032         .read    = seq_read,
1033         .llseek  = seq_lseek,
1034         .release = seq_release_private,
1035 };
1036
1037 #endif /* PROC_FS */
1038
1039 static int __init nfnetlink_log_init(void)
1040 {
1041         int i, status = -ENOMEM;
1042 #ifdef CONFIG_PROC_FS
1043         struct proc_dir_entry *proc_nful;
1044 #endif
1045         
1046         for (i = 0; i < INSTANCE_BUCKETS; i++)
1047                 INIT_HLIST_HEAD(&instance_table[i]);
1048         
1049         /* it's not really all that important to have a random value, so
1050          * we can do this from the init function, even if there hasn't
1051          * been that much entropy yet */
1052         get_random_bytes(&hash_init, sizeof(hash_init));
1053
1054         netlink_register_notifier(&nfulnl_rtnl_notifier);
1055         status = nfnetlink_subsys_register(&nfulnl_subsys);
1056         if (status < 0) {
1057                 printk(KERN_ERR "log: failed to create netlink socket\n");
1058                 goto cleanup_netlink_notifier;
1059         }
1060
1061 #ifdef CONFIG_PROC_FS
1062         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1063                                       proc_net_netfilter);
1064         if (!proc_nful)
1065                 goto cleanup_subsys;
1066         proc_nful->proc_fops = &nful_file_ops;
1067 #endif
1068         return status;
1069
1070 #ifdef CONFIG_PROC_FS
1071 cleanup_subsys:
1072         nfnetlink_subsys_unregister(&nfulnl_subsys);
1073 #endif
1074 cleanup_netlink_notifier:
1075         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1076         return status;
1077 }
1078
1079 static void __exit nfnetlink_log_fini(void)
1080 {
1081         nf_log_unregister_logger(&nfulnl_logger);
1082 #ifdef CONFIG_PROC_FS
1083         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1084 #endif
1085         nfnetlink_subsys_unregister(&nfulnl_subsys);
1086         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1087 }
1088
1089 MODULE_DESCRIPTION("netfilter userspace logging");
1090 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1091 MODULE_LICENSE("GPL");
1092 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1093
1094 module_init(nfnetlink_log_init);
1095 module_exit(nfnetlink_log_fini);