[NETFILTER]: nfnetlink_log: remove conditional locking
[linux-2.6.git] / net / netfilter / nfnetlink_log.c
1 /*
2  * This is a module which is used for logging packets to userspace via
3  * nfetlink.
4  *
5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
6  *
7  * Based on the old ipv4-only ipt_ULOG.c:
8  * (C) 2000-2004 by Harald Welte <laforge@netfilter.org>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License version 2 as
12  * published by the Free Software Foundation.
13  */
14 #include <linux/module.h>
15 #include <linux/skbuff.h>
16 #include <linux/init.h>
17 #include <linux/ip.h>
18 #include <linux/ipv6.h>
19 #include <linux/netdevice.h>
20 #include <linux/netfilter.h>
21 #include <linux/netlink.h>
22 #include <linux/netfilter/nfnetlink.h>
23 #include <linux/netfilter/nfnetlink_log.h>
24 #include <linux/spinlock.h>
25 #include <linux/sysctl.h>
26 #include <linux/proc_fs.h>
27 #include <linux/security.h>
28 #include <linux/list.h>
29 #include <linux/jhash.h>
30 #include <linux/random.h>
31 #include <net/sock.h>
32
33 #include <asm/atomic.h>
34
35 #ifdef CONFIG_BRIDGE_NETFILTER
36 #include "../bridge/br_private.h"
37 #endif
38
39 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
40 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
41 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
42
43 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
44                                      printk(x, ## args); } while (0);
45
46 #if 0
47 #define UDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
48                                         __FILE__, __LINE__, __FUNCTION__,  \
49                                         ## args)
50 #else
51 #define UDEBUG(x, ...)
52 #endif
53
54 struct nfulnl_instance {
55         struct hlist_node hlist;        /* global list of instances */
56         spinlock_t lock;
57         atomic_t use;                   /* use count */
58
59         unsigned int qlen;              /* number of nlmsgs in skb */
60         struct sk_buff *skb;            /* pre-allocatd skb */
61         struct nlmsghdr *lastnlh;       /* netlink header of last msg in skb */
62         struct timer_list timer;
63         int peer_pid;                   /* PID of the peer process */
64
65         /* configurable parameters */
66         unsigned int flushtimeout;      /* timeout until queue flush */
67         unsigned int nlbufsiz;          /* netlink buffer allocation size */
68         unsigned int qthreshold;        /* threshold of the queue */
69         u_int32_t copy_range;
70         u_int32_t seq;                  /* instance-local sequential counter */
71         u_int16_t group_num;            /* number of this queue */
72         u_int16_t flags;
73         u_int8_t copy_mode;
74 };
75
76 static DEFINE_RWLOCK(instances_lock);
77 static atomic_t global_seq;
78
79 #define INSTANCE_BUCKETS        16
80 static struct hlist_head instance_table[INSTANCE_BUCKETS];
81 static unsigned int hash_init;
82
83 static inline u_int8_t instance_hashfn(u_int16_t group_num)
84 {
85         return ((group_num & 0xff) % INSTANCE_BUCKETS);
86 }
87
88 static struct nfulnl_instance *
89 __instance_lookup(u_int16_t group_num)
90 {
91         struct hlist_head *head;
92         struct hlist_node *pos;
93         struct nfulnl_instance *inst;
94
95         UDEBUG("entering (group_num=%u)\n", group_num);
96
97         head = &instance_table[instance_hashfn(group_num)];
98         hlist_for_each_entry(inst, pos, head, hlist) {
99                 if (inst->group_num == group_num)
100                         return inst;
101         }
102         return NULL;
103 }
104
105 static inline void
106 instance_get(struct nfulnl_instance *inst)
107 {
108         atomic_inc(&inst->use);
109 }
110
111 static struct nfulnl_instance *
112 instance_lookup_get(u_int16_t group_num)
113 {
114         struct nfulnl_instance *inst;
115
116         read_lock_bh(&instances_lock);
117         inst = __instance_lookup(group_num);
118         if (inst)
119                 instance_get(inst);
120         read_unlock_bh(&instances_lock);
121
122         return inst;
123 }
124
125 static void
126 instance_put(struct nfulnl_instance *inst)
127 {
128         if (inst && atomic_dec_and_test(&inst->use)) {
129                 UDEBUG("kfree(inst=%p)\n", inst);
130                 kfree(inst);
131                 module_put(THIS_MODULE);
132         }
133 }
134
135 static void nfulnl_timer(unsigned long data);
136
137 static struct nfulnl_instance *
138 instance_create(u_int16_t group_num, int pid)
139 {
140         struct nfulnl_instance *inst;
141
142         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
143                 pid);
144
145         write_lock_bh(&instances_lock);
146         if (__instance_lookup(group_num)) {
147                 inst = NULL;
148                 UDEBUG("aborting, instance already exists\n");
149                 goto out_unlock;
150         }
151
152         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
153         if (!inst)
154                 goto out_unlock;
155
156         INIT_HLIST_NODE(&inst->hlist);
157         spin_lock_init(&inst->lock);
158         /* needs to be two, since we _put() after creation */
159         atomic_set(&inst->use, 2);
160
161         init_timer(&inst->timer);
162         inst->timer.function = nfulnl_timer;
163         inst->timer.data = (unsigned long)inst;
164         /* don't start timer yet. (re)start it  with every packet */
165
166         inst->peer_pid = pid;
167         inst->group_num = group_num;
168
169         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
170         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
171         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
172         inst->copy_mode         = NFULNL_COPY_PACKET;
173         inst->copy_range        = 0xffff;
174
175         if (!try_module_get(THIS_MODULE))
176                 goto out_free;
177
178         hlist_add_head(&inst->hlist,
179                        &instance_table[instance_hashfn(group_num)]);
180
181         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
182                 inst->hlist.next);
183
184         write_unlock_bh(&instances_lock);
185
186         return inst;
187
188 out_free:
189         instance_put(inst);
190 out_unlock:
191         write_unlock_bh(&instances_lock);
192         return NULL;
193 }
194
195 static int __nfulnl_send(struct nfulnl_instance *inst);
196
197 static void
198 __instance_destroy(struct nfulnl_instance *inst)
199 {
200         /* first pull it out of the global list */
201         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
202                 inst, inst->group_num);
203
204         hlist_del(&inst->hlist);
205
206         /* then flush all pending packets from skb */
207
208         spin_lock_bh(&inst->lock);
209         if (inst->skb) {
210                 /* timer "holds" one reference (we have one more) */
211                 if (del_timer(&inst->timer))
212                         instance_put(inst);
213                 if (inst->qlen)
214                         __nfulnl_send(inst);
215                 if (inst->skb) {
216                         kfree_skb(inst->skb);
217                         inst->skb = NULL;
218                 }
219         }
220         spin_unlock_bh(&inst->lock);
221
222         /* and finally put the refcount */
223         instance_put(inst);
224 }
225
226 static inline void
227 instance_destroy(struct nfulnl_instance *inst)
228 {
229         write_lock_bh(&instances_lock);
230         __instance_destroy(inst);
231         write_unlock_bh(&instances_lock);
232 }
233
234 static int
235 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
236                   unsigned int range)
237 {
238         int status = 0;
239
240         spin_lock_bh(&inst->lock);
241
242         switch (mode) {
243         case NFULNL_COPY_NONE:
244         case NFULNL_COPY_META:
245                 inst->copy_mode = mode;
246                 inst->copy_range = 0;
247                 break;
248
249         case NFULNL_COPY_PACKET:
250                 inst->copy_mode = mode;
251                 /* we're using struct nfattr which has 16bit nfa_len */
252                 if (range > 0xffff)
253                         inst->copy_range = 0xffff;
254                 else
255                         inst->copy_range = range;
256                 break;
257
258         default:
259                 status = -EINVAL;
260                 break;
261         }
262
263         spin_unlock_bh(&inst->lock);
264
265         return status;
266 }
267
268 static int
269 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
270 {
271         int status;
272
273         spin_lock_bh(&inst->lock);
274         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
275                 status = -ERANGE;
276         else if (nlbufsiz > 131072)
277                 status = -ERANGE;
278         else {
279                 inst->nlbufsiz = nlbufsiz;
280                 status = 0;
281         }
282         spin_unlock_bh(&inst->lock);
283
284         return status;
285 }
286
287 static int
288 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
289 {
290         spin_lock_bh(&inst->lock);
291         inst->flushtimeout = timeout;
292         spin_unlock_bh(&inst->lock);
293
294         return 0;
295 }
296
297 static int
298 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
299 {
300         spin_lock_bh(&inst->lock);
301         inst->qthreshold = qthresh;
302         spin_unlock_bh(&inst->lock);
303
304         return 0;
305 }
306
307 static int
308 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
309 {
310         spin_lock_bh(&inst->lock);
311         inst->flags = flags;
312         spin_unlock_bh(&inst->lock);
313
314         return 0;
315 }
316
317 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
318                                         unsigned int pkt_size)
319 {
320         struct sk_buff *skb;
321         unsigned int n;
322
323         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
324
325         /* alloc skb which should be big enough for a whole multipart
326          * message.  WARNING: has to be <= 128k due to slab restrictions */
327
328         n = max(inst_size, pkt_size);
329         skb = alloc_skb(n, GFP_ATOMIC);
330         if (!skb) {
331                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
332                         inst_size);
333
334                 if (n > pkt_size) {
335                         /* try to allocate only as much as we need for current
336                          * packet */
337
338                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
339                         if (!skb)
340                                 PRINTR("nfnetlink_log: can't even alloc %u "
341                                        "bytes\n", pkt_size);
342                 }
343         }
344
345         return skb;
346 }
347
348 static int
349 __nfulnl_send(struct nfulnl_instance *inst)
350 {
351         int status;
352
353         if (inst->qlen > 1)
354                 inst->lastnlh->nlmsg_type = NLMSG_DONE;
355
356         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
357         if (status < 0) {
358                 UDEBUG("netlink_unicast() failed\n");
359                 /* FIXME: statistics */
360         }
361
362         inst->qlen = 0;
363         inst->skb = NULL;
364         inst->lastnlh = NULL;
365
366         return status;
367 }
368
369 static void nfulnl_timer(unsigned long data)
370 {
371         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
372
373         UDEBUG("timer function called, flushing buffer\n");
374
375         spin_lock_bh(&inst->lock);
376         if (inst->skb)
377                 __nfulnl_send(inst);
378         spin_unlock_bh(&inst->lock);
379         instance_put(inst);
380 }
381
382 /* This is an inline function, we don't really care about a long
383  * list of arguments */
384 static inline int
385 __build_packet_message(struct nfulnl_instance *inst,
386                         const struct sk_buff *skb,
387                         unsigned int data_len,
388                         unsigned int pf,
389                         unsigned int hooknum,
390                         const struct net_device *indev,
391                         const struct net_device *outdev,
392                         const struct nf_loginfo *li,
393                         const char *prefix, unsigned int plen)
394 {
395         struct nfulnl_msg_packet_hdr pmsg;
396         struct nlmsghdr *nlh;
397         struct nfgenmsg *nfmsg;
398         __be32 tmp_uint;
399         sk_buff_data_t old_tail = inst->skb->tail;
400
401         UDEBUG("entered\n");
402
403         nlh = NLMSG_PUT(inst->skb, 0, 0,
404                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
405                         sizeof(struct nfgenmsg));
406         nfmsg = NLMSG_DATA(nlh);
407         nfmsg->nfgen_family = pf;
408         nfmsg->version = NFNETLINK_V0;
409         nfmsg->res_id = htons(inst->group_num);
410
411         pmsg.hw_protocol        = skb->protocol;
412         pmsg.hook               = hooknum;
413
414         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
415
416         if (prefix)
417                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
418
419         if (indev) {
420                 tmp_uint = htonl(indev->ifindex);
421 #ifndef CONFIG_BRIDGE_NETFILTER
422                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
423                         &tmp_uint);
424 #else
425                 if (pf == PF_BRIDGE) {
426                         /* Case 1: outdev is physical input device, we need to
427                          * look for bridge group (when called from
428                          * netfilter_bridge) */
429                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
430                                 sizeof(tmp_uint), &tmp_uint);
431                         /* this is the bridge group "brX" */
432                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
433                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
434                                 sizeof(tmp_uint), &tmp_uint);
435                 } else {
436                         /* Case 2: indev is bridge group, we need to look for
437                          * physical device (when called from ipv4) */
438                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
439                                 sizeof(tmp_uint), &tmp_uint);
440                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
441                                 tmp_uint =
442                                     htonl(skb->nf_bridge->physindev->ifindex);
443                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
444                                         sizeof(tmp_uint), &tmp_uint);
445                         }
446                 }
447 #endif
448         }
449
450         if (outdev) {
451                 tmp_uint = htonl(outdev->ifindex);
452 #ifndef CONFIG_BRIDGE_NETFILTER
453                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
454                         &tmp_uint);
455 #else
456                 if (pf == PF_BRIDGE) {
457                         /* Case 1: outdev is physical output device, we need to
458                          * look for bridge group (when called from
459                          * netfilter_bridge) */
460                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
461                                 sizeof(tmp_uint), &tmp_uint);
462                         /* this is the bridge group "brX" */
463                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
464                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
465                                 sizeof(tmp_uint), &tmp_uint);
466                 } else {
467                         /* Case 2: indev is a bridge group, we need to look
468                          * for physical device (when called from ipv4) */
469                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
470                                 sizeof(tmp_uint), &tmp_uint);
471                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
472                                 tmp_uint =
473                                     htonl(skb->nf_bridge->physoutdev->ifindex);
474                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
475                                         sizeof(tmp_uint), &tmp_uint);
476                         }
477                 }
478 #endif
479         }
480
481         if (skb->mark) {
482                 tmp_uint = htonl(skb->mark);
483                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
484         }
485
486         if (indev && skb->dev && skb->dev->hard_header_parse) {
487                 struct nfulnl_msg_packet_hw phw;
488                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
489                                                     phw.hw_addr);
490                 phw.hw_addrlen = htons(len);
491                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
492         }
493
494         if (skb->tstamp.tv64) {
495                 struct nfulnl_msg_packet_timestamp ts;
496                 struct timeval tv = ktime_to_timeval(skb->tstamp);
497                 ts.sec = cpu_to_be64(tv.tv_sec);
498                 ts.usec = cpu_to_be64(tv.tv_usec);
499
500                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
501         }
502
503         /* UID */
504         if (skb->sk) {
505                 read_lock_bh(&skb->sk->sk_callback_lock);
506                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
507                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
508                         /* need to unlock here since NFA_PUT may goto */
509                         read_unlock_bh(&skb->sk->sk_callback_lock);
510                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
511                 } else
512                         read_unlock_bh(&skb->sk->sk_callback_lock);
513         }
514
515         /* local sequence number */
516         if (inst->flags & NFULNL_CFG_F_SEQ) {
517                 tmp_uint = htonl(inst->seq++);
518                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
519         }
520         /* global sequence number */
521         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
522                 tmp_uint = htonl(atomic_inc_return(&global_seq));
523                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
524         }
525
526         if (data_len) {
527                 struct nfattr *nfa;
528                 int size = NFA_LENGTH(data_len);
529
530                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
531                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
532                         goto nlmsg_failure;
533                 }
534
535                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
536                 nfa->nfa_type = NFULA_PAYLOAD;
537                 nfa->nfa_len = size;
538
539                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
540                         BUG();
541         }
542
543         nlh->nlmsg_len = inst->skb->tail - old_tail;
544         inst->lastnlh = nlh;
545         return 0;
546
547 nlmsg_failure:
548         UDEBUG("nlmsg_failure\n");
549 nfattr_failure:
550         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
551         return -1;
552 }
553
554 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
555
556 static struct nf_loginfo default_loginfo = {
557         .type =         NF_LOG_TYPE_ULOG,
558         .u = {
559                 .ulog = {
560                         .copy_len       = 0xffff,
561                         .group          = 0,
562                         .qthreshold     = 1,
563                 },
564         },
565 };
566
567 /* log handler for internal netfilter logging api */
568 static void
569 nfulnl_log_packet(unsigned int pf,
570                   unsigned int hooknum,
571                   const struct sk_buff *skb,
572                   const struct net_device *in,
573                   const struct net_device *out,
574                   const struct nf_loginfo *li_user,
575                   const char *prefix)
576 {
577         unsigned int size, data_len;
578         struct nfulnl_instance *inst;
579         const struct nf_loginfo *li;
580         unsigned int qthreshold;
581         unsigned int plen;
582
583         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
584                 li = li_user;
585         else
586                 li = &default_loginfo;
587
588         inst = instance_lookup_get(li->u.ulog.group);
589         if (!inst)
590                 inst = instance_lookup_get(0);
591         if (!inst) {
592                 PRINTR("nfnetlink_log: trying to log packet, "
593                         "but no instance for group %u\n", li->u.ulog.group);
594                 return;
595         }
596
597         plen = 0;
598         if (prefix)
599                 plen = strlen(prefix) + 1;
600
601         /* all macros expand to constant values at compile time */
602         /* FIXME: do we want to make the size calculation conditional based on
603          * what is actually present?  way more branches and checks, but more
604          * memory efficient... */
605         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
606                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
607                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
608                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
609 #ifdef CONFIG_BRIDGE_NETFILTER
610                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
611                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
612 #endif
613                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
614                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
615                 + NFA_SPACE(plen)               /* prefix */
616                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
617                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
618
619         UDEBUG("initial size=%u\n", size);
620
621         spin_lock_bh(&inst->lock);
622
623         if (inst->flags & NFULNL_CFG_F_SEQ)
624                 size += NFA_SPACE(sizeof(u_int32_t));
625         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
626                 size += NFA_SPACE(sizeof(u_int32_t));
627
628         qthreshold = inst->qthreshold;
629         /* per-rule qthreshold overrides per-instance */
630         if (qthreshold > li->u.ulog.qthreshold)
631                 qthreshold = li->u.ulog.qthreshold;
632
633         switch (inst->copy_mode) {
634         case NFULNL_COPY_META:
635         case NFULNL_COPY_NONE:
636                 data_len = 0;
637                 break;
638
639         case NFULNL_COPY_PACKET:
640                 if (inst->copy_range == 0
641                     || inst->copy_range > skb->len)
642                         data_len = skb->len;
643                 else
644                         data_len = inst->copy_range;
645
646                 size += NFA_SPACE(data_len);
647                 UDEBUG("copy_packet, therefore size now %u\n", size);
648                 break;
649
650         default:
651                 goto unlock_and_release;
652         }
653
654         if (inst->qlen >= qthreshold ||
655             (inst->skb && size > skb_tailroom(inst->skb))) {
656                 /* either the queue len is too high or we don't have
657                  * enough room in the skb left. flush to userspace. */
658                 UDEBUG("flushing old skb\n");
659
660                 /* timer "holds" one reference (we have another one) */
661                 if (del_timer(&inst->timer))
662                         instance_put(inst);
663                 __nfulnl_send(inst);
664         }
665
666         if (!inst->skb) {
667                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
668                 if (!inst->skb)
669                         goto alloc_failure;
670         }
671
672         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
673         inst->qlen++;
674
675         __build_packet_message(inst, skb, data_len, pf,
676                                 hooknum, in, out, li, prefix, plen);
677
678         /* timer_pending always called within inst->lock, so there
679          * is no chance of a race here */
680         if (!timer_pending(&inst->timer)) {
681                 instance_get(inst);
682                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
683                 add_timer(&inst->timer);
684         }
685
686 unlock_and_release:
687         spin_unlock_bh(&inst->lock);
688         instance_put(inst);
689         return;
690
691 alloc_failure:
692         UDEBUG("error allocating skb\n");
693         /* FIXME: statistics */
694         goto unlock_and_release;
695 }
696
697 static int
698 nfulnl_rcv_nl_event(struct notifier_block *this,
699                    unsigned long event, void *ptr)
700 {
701         struct netlink_notify *n = ptr;
702
703         if (event == NETLINK_URELEASE &&
704             n->protocol == NETLINK_NETFILTER && n->pid) {
705                 int i;
706
707                 /* destroy all instances for this pid */
708                 write_lock_bh(&instances_lock);
709                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
710                         struct hlist_node *tmp, *t2;
711                         struct nfulnl_instance *inst;
712                         struct hlist_head *head = &instance_table[i];
713
714                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
715                                 UDEBUG("node = %p\n", inst);
716                                 if (n->pid == inst->peer_pid)
717                                         __instance_destroy(inst);
718                         }
719                 }
720                 write_unlock_bh(&instances_lock);
721         }
722         return NOTIFY_DONE;
723 }
724
725 static struct notifier_block nfulnl_rtnl_notifier = {
726         .notifier_call  = nfulnl_rcv_nl_event,
727 };
728
729 static int
730 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
731                   struct nlmsghdr *nlh, struct nfattr *nfqa[])
732 {
733         return -ENOTSUPP;
734 }
735
736 static struct nf_logger nfulnl_logger = {
737         .name   = "nfnetlink_log",
738         .logfn  = &nfulnl_log_packet,
739         .me     = THIS_MODULE,
740 };
741
742 static const int nfula_min[NFULA_MAX] = {
743         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
744         [NFULA_MARK-1]          = sizeof(u_int32_t),
745         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
746         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
747         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
748         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
749         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
750         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
751         [NFULA_PAYLOAD-1]       = 0,
752         [NFULA_PREFIX-1]        = 0,
753         [NFULA_UID-1]           = sizeof(u_int32_t),
754         [NFULA_SEQ-1]           = sizeof(u_int32_t),
755         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
756 };
757
758 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
759         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
760         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
761         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
762         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
763         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
764         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
765 };
766
767 static int
768 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
769                    struct nlmsghdr *nlh, struct nfattr *nfula[])
770 {
771         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
772         u_int16_t group_num = ntohs(nfmsg->res_id);
773         struct nfulnl_instance *inst;
774         int ret = 0;
775
776         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
777
778         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
779                 UDEBUG("bad attribute size\n");
780                 return -EINVAL;
781         }
782
783         inst = instance_lookup_get(group_num);
784         if (nfula[NFULA_CFG_CMD-1]) {
785                 u_int8_t pf = nfmsg->nfgen_family;
786                 struct nfulnl_msg_config_cmd *cmd;
787                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
788                 UDEBUG("found CFG_CMD for\n");
789
790                 switch (cmd->command) {
791                 case NFULNL_CFG_CMD_BIND:
792                         if (inst) {
793                                 ret = -EBUSY;
794                                 goto out_put;
795                         }
796
797                         inst = instance_create(group_num,
798                                                NETLINK_CB(skb).pid);
799                         if (!inst) {
800                                 ret = -EINVAL;
801                                 goto out;
802                         }
803                         break;
804                 case NFULNL_CFG_CMD_UNBIND:
805                         if (!inst) {
806                                 ret = -ENODEV;
807                                 goto out;
808                         }
809
810                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
811                                 ret = -EPERM;
812                                 goto out_put;
813                         }
814
815                         instance_destroy(inst);
816                         goto out;
817                 case NFULNL_CFG_CMD_PF_BIND:
818                         UDEBUG("registering log handler for pf=%u\n", pf);
819                         ret = nf_log_register(pf, &nfulnl_logger);
820                         break;
821                 case NFULNL_CFG_CMD_PF_UNBIND:
822                         UDEBUG("unregistering log handler for pf=%u\n", pf);
823                         /* This is a bug and a feature.  We cannot unregister
824                          * other handlers, like nfnetlink_inst can */
825                         nf_log_unregister_pf(pf);
826                         break;
827                 default:
828                         ret = -EINVAL;
829                         break;
830                 }
831
832                 if (!inst)
833                         goto out;
834         } else {
835                 if (!inst) {
836                         UDEBUG("no config command, and no instance for "
837                                 "group=%u pid=%u =>ENOENT\n",
838                                 group_num, NETLINK_CB(skb).pid);
839                         ret = -ENOENT;
840                         goto out;
841                 }
842
843                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
844                         UDEBUG("no config command, and wrong pid\n");
845                         ret = -EPERM;
846                         goto out_put;
847                 }
848         }
849
850         if (nfula[NFULA_CFG_MODE-1]) {
851                 struct nfulnl_msg_config_mode *params;
852                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
853
854                 nfulnl_set_mode(inst, params->copy_mode,
855                                 ntohl(params->copy_range));
856         }
857
858         if (nfula[NFULA_CFG_TIMEOUT-1]) {
859                 __be32 timeout =
860                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
861
862                 nfulnl_set_timeout(inst, ntohl(timeout));
863         }
864
865         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
866                 __be32 nlbufsiz =
867                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
868
869                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
870         }
871
872         if (nfula[NFULA_CFG_QTHRESH-1]) {
873                 __be32 qthresh =
874                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
875
876                 nfulnl_set_qthresh(inst, ntohl(qthresh));
877         }
878
879         if (nfula[NFULA_CFG_FLAGS-1]) {
880                 __be16 flags =
881                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
882                 nfulnl_set_flags(inst, ntohs(flags));
883         }
884
885 out_put:
886         instance_put(inst);
887 out:
888         return ret;
889 }
890
891 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
892         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
893                                     .attr_count = NFULA_MAX, },
894         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
895                                     .attr_count = NFULA_CFG_MAX, },
896 };
897
898 static struct nfnetlink_subsystem nfulnl_subsys = {
899         .name           = "log",
900         .subsys_id      = NFNL_SUBSYS_ULOG,
901         .cb_count       = NFULNL_MSG_MAX,
902         .cb             = nfulnl_cb,
903 };
904
905 #ifdef CONFIG_PROC_FS
906 struct iter_state {
907         unsigned int bucket;
908 };
909
910 static struct hlist_node *get_first(struct iter_state *st)
911 {
912         if (!st)
913                 return NULL;
914
915         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
916                 if (!hlist_empty(&instance_table[st->bucket]))
917                         return instance_table[st->bucket].first;
918         }
919         return NULL;
920 }
921
922 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
923 {
924         h = h->next;
925         while (!h) {
926                 if (++st->bucket >= INSTANCE_BUCKETS)
927                         return NULL;
928
929                 h = instance_table[st->bucket].first;
930         }
931         return h;
932 }
933
934 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
935 {
936         struct hlist_node *head;
937         head = get_first(st);
938
939         if (head)
940                 while (pos && (head = get_next(st, head)))
941                         pos--;
942         return pos ? NULL : head;
943 }
944
945 static void *seq_start(struct seq_file *seq, loff_t *pos)
946 {
947         read_lock_bh(&instances_lock);
948         return get_idx(seq->private, *pos);
949 }
950
951 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
952 {
953         (*pos)++;
954         return get_next(s->private, v);
955 }
956
957 static void seq_stop(struct seq_file *s, void *v)
958 {
959         read_unlock_bh(&instances_lock);
960 }
961
962 static int seq_show(struct seq_file *s, void *v)
963 {
964         const struct nfulnl_instance *inst = v;
965
966         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
967                           inst->group_num,
968                           inst->peer_pid, inst->qlen,
969                           inst->copy_mode, inst->copy_range,
970                           inst->flushtimeout, atomic_read(&inst->use));
971 }
972
973 static struct seq_operations nful_seq_ops = {
974         .start  = seq_start,
975         .next   = seq_next,
976         .stop   = seq_stop,
977         .show   = seq_show,
978 };
979
980 static int nful_open(struct inode *inode, struct file *file)
981 {
982         struct seq_file *seq;
983         struct iter_state *is;
984         int ret;
985
986         is = kzalloc(sizeof(*is), GFP_KERNEL);
987         if (!is)
988                 return -ENOMEM;
989         ret = seq_open(file, &nful_seq_ops);
990         if (ret < 0)
991                 goto out_free;
992         seq = file->private_data;
993         seq->private = is;
994         return ret;
995 out_free:
996         kfree(is);
997         return ret;
998 }
999
1000 static const struct file_operations nful_file_ops = {
1001         .owner   = THIS_MODULE,
1002         .open    = nful_open,
1003         .read    = seq_read,
1004         .llseek  = seq_lseek,
1005         .release = seq_release_private,
1006 };
1007
1008 #endif /* PROC_FS */
1009
1010 static int __init nfnetlink_log_init(void)
1011 {
1012         int i, status = -ENOMEM;
1013 #ifdef CONFIG_PROC_FS
1014         struct proc_dir_entry *proc_nful;
1015 #endif
1016
1017         for (i = 0; i < INSTANCE_BUCKETS; i++)
1018                 INIT_HLIST_HEAD(&instance_table[i]);
1019
1020         /* it's not really all that important to have a random value, so
1021          * we can do this from the init function, even if there hasn't
1022          * been that much entropy yet */
1023         get_random_bytes(&hash_init, sizeof(hash_init));
1024
1025         netlink_register_notifier(&nfulnl_rtnl_notifier);
1026         status = nfnetlink_subsys_register(&nfulnl_subsys);
1027         if (status < 0) {
1028                 printk(KERN_ERR "log: failed to create netlink socket\n");
1029                 goto cleanup_netlink_notifier;
1030         }
1031
1032 #ifdef CONFIG_PROC_FS
1033         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1034                                       proc_net_netfilter);
1035         if (!proc_nful)
1036                 goto cleanup_subsys;
1037         proc_nful->proc_fops = &nful_file_ops;
1038 #endif
1039         return status;
1040
1041 #ifdef CONFIG_PROC_FS
1042 cleanup_subsys:
1043         nfnetlink_subsys_unregister(&nfulnl_subsys);
1044 #endif
1045 cleanup_netlink_notifier:
1046         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1047         return status;
1048 }
1049
1050 static void __exit nfnetlink_log_fini(void)
1051 {
1052         nf_log_unregister(&nfulnl_logger);
1053 #ifdef CONFIG_PROC_FS
1054         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1055 #endif
1056         nfnetlink_subsys_unregister(&nfulnl_subsys);
1057         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1058 }
1059
1060 MODULE_DESCRIPTION("netfilter userspace logging");
1061 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1062 MODULE_LICENSE("GPL");
1063 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1064
1065 module_init(nfnetlink_log_init);
1066 module_exit(nfnetlink_log_fini);