[NETFILTER]: nfnetlink_log: flush queue early
[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 timer_list timer;
62         int peer_pid;                   /* PID of the peer process */
63
64         /* configurable parameters */
65         unsigned int flushtimeout;      /* timeout until queue flush */
66         unsigned int nlbufsiz;          /* netlink buffer allocation size */
67         unsigned int qthreshold;        /* threshold of the queue */
68         u_int32_t copy_range;
69         u_int32_t seq;                  /* instance-local sequential counter */
70         u_int16_t group_num;            /* number of this queue */
71         u_int16_t flags;
72         u_int8_t copy_mode;
73 };
74
75 static DEFINE_RWLOCK(instances_lock);
76 static atomic_t global_seq;
77
78 #define INSTANCE_BUCKETS        16
79 static struct hlist_head instance_table[INSTANCE_BUCKETS];
80 static unsigned int hash_init;
81
82 static inline u_int8_t instance_hashfn(u_int16_t group_num)
83 {
84         return ((group_num & 0xff) % INSTANCE_BUCKETS);
85 }
86
87 static struct nfulnl_instance *
88 __instance_lookup(u_int16_t group_num)
89 {
90         struct hlist_head *head;
91         struct hlist_node *pos;
92         struct nfulnl_instance *inst;
93
94         UDEBUG("entering (group_num=%u)\n", group_num);
95
96         head = &instance_table[instance_hashfn(group_num)];
97         hlist_for_each_entry(inst, pos, head, hlist) {
98                 if (inst->group_num == group_num)
99                         return inst;
100         }
101         return NULL;
102 }
103
104 static inline void
105 instance_get(struct nfulnl_instance *inst)
106 {
107         atomic_inc(&inst->use);
108 }
109
110 static struct nfulnl_instance *
111 instance_lookup_get(u_int16_t group_num)
112 {
113         struct nfulnl_instance *inst;
114
115         read_lock_bh(&instances_lock);
116         inst = __instance_lookup(group_num);
117         if (inst)
118                 instance_get(inst);
119         read_unlock_bh(&instances_lock);
120
121         return inst;
122 }
123
124 static void
125 instance_put(struct nfulnl_instance *inst)
126 {
127         if (inst && atomic_dec_and_test(&inst->use)) {
128                 UDEBUG("kfree(inst=%p)\n", inst);
129                 kfree(inst);
130                 module_put(THIS_MODULE);
131         }
132 }
133
134 static void nfulnl_timer(unsigned long data);
135
136 static struct nfulnl_instance *
137 instance_create(u_int16_t group_num, int pid)
138 {
139         struct nfulnl_instance *inst;
140
141         UDEBUG("entering (group_num=%u, pid=%d)\n", group_num,
142                 pid);
143
144         write_lock_bh(&instances_lock);
145         if (__instance_lookup(group_num)) {
146                 inst = NULL;
147                 UDEBUG("aborting, instance already exists\n");
148                 goto out_unlock;
149         }
150
151         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
152         if (!inst)
153                 goto out_unlock;
154
155         INIT_HLIST_NODE(&inst->hlist);
156         spin_lock_init(&inst->lock);
157         /* needs to be two, since we _put() after creation */
158         atomic_set(&inst->use, 2);
159
160         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
161
162         inst->peer_pid = pid;
163         inst->group_num = group_num;
164
165         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
166         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
167         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
168         inst->copy_mode         = NFULNL_COPY_PACKET;
169         inst->copy_range        = 0xffff;
170
171         if (!try_module_get(THIS_MODULE))
172                 goto out_free;
173
174         hlist_add_head(&inst->hlist,
175                        &instance_table[instance_hashfn(group_num)]);
176
177         UDEBUG("newly added node: %p, next=%p\n", &inst->hlist,
178                 inst->hlist.next);
179
180         write_unlock_bh(&instances_lock);
181
182         return inst;
183
184 out_free:
185         instance_put(inst);
186 out_unlock:
187         write_unlock_bh(&instances_lock);
188         return NULL;
189 }
190
191 static void __nfulnl_flush(struct nfulnl_instance *inst);
192
193 static void
194 __instance_destroy(struct nfulnl_instance *inst)
195 {
196         /* first pull it out of the global list */
197         UDEBUG("removing instance %p (queuenum=%u) from hash\n",
198                 inst, inst->group_num);
199
200         hlist_del(&inst->hlist);
201
202         /* then flush all pending packets from skb */
203
204         spin_lock_bh(&inst->lock);
205         if (inst->skb)
206                 __nfulnl_flush(inst);
207         spin_unlock_bh(&inst->lock);
208
209         /* and finally put the refcount */
210         instance_put(inst);
211 }
212
213 static inline void
214 instance_destroy(struct nfulnl_instance *inst)
215 {
216         write_lock_bh(&instances_lock);
217         __instance_destroy(inst);
218         write_unlock_bh(&instances_lock);
219 }
220
221 static int
222 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
223                   unsigned int range)
224 {
225         int status = 0;
226
227         spin_lock_bh(&inst->lock);
228
229         switch (mode) {
230         case NFULNL_COPY_NONE:
231         case NFULNL_COPY_META:
232                 inst->copy_mode = mode;
233                 inst->copy_range = 0;
234                 break;
235
236         case NFULNL_COPY_PACKET:
237                 inst->copy_mode = mode;
238                 /* we're using struct nlattr which has 16bit nfa_len */
239                 if (range > 0xffff)
240                         inst->copy_range = 0xffff;
241                 else
242                         inst->copy_range = range;
243                 break;
244
245         default:
246                 status = -EINVAL;
247                 break;
248         }
249
250         spin_unlock_bh(&inst->lock);
251
252         return status;
253 }
254
255 static int
256 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
257 {
258         int status;
259
260         spin_lock_bh(&inst->lock);
261         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
262                 status = -ERANGE;
263         else if (nlbufsiz > 131072)
264                 status = -ERANGE;
265         else {
266                 inst->nlbufsiz = nlbufsiz;
267                 status = 0;
268         }
269         spin_unlock_bh(&inst->lock);
270
271         return status;
272 }
273
274 static int
275 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
276 {
277         spin_lock_bh(&inst->lock);
278         inst->flushtimeout = timeout;
279         spin_unlock_bh(&inst->lock);
280
281         return 0;
282 }
283
284 static int
285 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
286 {
287         spin_lock_bh(&inst->lock);
288         inst->qthreshold = qthresh;
289         spin_unlock_bh(&inst->lock);
290
291         return 0;
292 }
293
294 static int
295 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
296 {
297         spin_lock_bh(&inst->lock);
298         inst->flags = flags;
299         spin_unlock_bh(&inst->lock);
300
301         return 0;
302 }
303
304 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
305                                         unsigned int pkt_size)
306 {
307         struct sk_buff *skb;
308         unsigned int n;
309
310         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
311
312         /* alloc skb which should be big enough for a whole multipart
313          * message.  WARNING: has to be <= 128k due to slab restrictions */
314
315         n = max(inst_size, pkt_size);
316         skb = alloc_skb(n, GFP_ATOMIC);
317         if (!skb) {
318                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
319                         inst_size);
320
321                 if (n > pkt_size) {
322                         /* try to allocate only as much as we need for current
323                          * packet */
324
325                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
326                         if (!skb)
327                                 PRINTR("nfnetlink_log: can't even alloc %u "
328                                        "bytes\n", pkt_size);
329                 }
330         }
331
332         return skb;
333 }
334
335 static int
336 __nfulnl_send(struct nfulnl_instance *inst)
337 {
338         int status = -1;
339
340         if (inst->qlen > 1)
341                 NLMSG_PUT(inst->skb, 0, 0,
342                           NLMSG_DONE,
343                           sizeof(struct nfgenmsg));
344
345         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
346         if (status < 0) {
347                 UDEBUG("netlink_unicast() failed\n");
348                 /* FIXME: statistics */
349         }
350
351         inst->qlen = 0;
352         inst->skb = NULL;
353
354 nlmsg_failure:
355         return status;
356 }
357
358 static void
359 __nfulnl_flush(struct nfulnl_instance *inst)
360 {
361         /* timer holds a reference */
362         if (del_timer(&inst->timer))
363                 instance_put(inst);
364         if (inst->skb)
365                 __nfulnl_send(inst);
366 }
367
368 static void nfulnl_timer(unsigned long data)
369 {
370         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
371
372         UDEBUG("timer function called, flushing buffer\n");
373
374         spin_lock_bh(&inst->lock);
375         if (inst->skb)
376                 __nfulnl_send(inst);
377         spin_unlock_bh(&inst->lock);
378         instance_put(inst);
379 }
380
381 /* This is an inline function, we don't really care about a long
382  * list of arguments */
383 static inline int
384 __build_packet_message(struct nfulnl_instance *inst,
385                         const struct sk_buff *skb,
386                         unsigned int data_len,
387                         unsigned int pf,
388                         unsigned int hooknum,
389                         const struct net_device *indev,
390                         const struct net_device *outdev,
391                         const struct nf_loginfo *li,
392                         const char *prefix, unsigned int plen)
393 {
394         struct nfulnl_msg_packet_hdr pmsg;
395         struct nlmsghdr *nlh;
396         struct nfgenmsg *nfmsg;
397         __be32 tmp_uint;
398         sk_buff_data_t old_tail = inst->skb->tail;
399
400         UDEBUG("entered\n");
401
402         nlh = NLMSG_PUT(inst->skb, 0, 0,
403                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
404                         sizeof(struct nfgenmsg));
405         nfmsg = NLMSG_DATA(nlh);
406         nfmsg->nfgen_family = pf;
407         nfmsg->version = NFNETLINK_V0;
408         nfmsg->res_id = htons(inst->group_num);
409
410         pmsg.hw_protocol        = skb->protocol;
411         pmsg.hook               = hooknum;
412
413         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
414
415         if (prefix)
416                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
417
418         if (indev) {
419                 tmp_uint = htonl(indev->ifindex);
420 #ifndef CONFIG_BRIDGE_NETFILTER
421                 NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
422                         &tmp_uint);
423 #else
424                 if (pf == PF_BRIDGE) {
425                         /* Case 1: outdev is physical input device, we need to
426                          * look for bridge group (when called from
427                          * netfilter_bridge) */
428                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
429                                 sizeof(tmp_uint), &tmp_uint);
430                         /* this is the bridge group "brX" */
431                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
432                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
433                                 sizeof(tmp_uint), &tmp_uint);
434                 } else {
435                         /* Case 2: indev is bridge group, we need to look for
436                          * physical device (when called from ipv4) */
437                         NLA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
438                                 sizeof(tmp_uint), &tmp_uint);
439                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
440                                 tmp_uint =
441                                     htonl(skb->nf_bridge->physindev->ifindex);
442                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
443                                         sizeof(tmp_uint), &tmp_uint);
444                         }
445                 }
446 #endif
447         }
448
449         if (outdev) {
450                 tmp_uint = htonl(outdev->ifindex);
451 #ifndef CONFIG_BRIDGE_NETFILTER
452                 NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
453                         &tmp_uint);
454 #else
455                 if (pf == PF_BRIDGE) {
456                         /* Case 1: outdev is physical output device, we need to
457                          * look for bridge group (when called from
458                          * netfilter_bridge) */
459                         NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
460                                 sizeof(tmp_uint), &tmp_uint);
461                         /* this is the bridge group "brX" */
462                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
463                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
464                                 sizeof(tmp_uint), &tmp_uint);
465                 } else {
466                         /* Case 2: indev is a bridge group, we need to look
467                          * for physical device (when called from ipv4) */
468                         NLA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
469                                 sizeof(tmp_uint), &tmp_uint);
470                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
471                                 tmp_uint =
472                                     htonl(skb->nf_bridge->physoutdev->ifindex);
473                                 NLA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
474                                         sizeof(tmp_uint), &tmp_uint);
475                         }
476                 }
477 #endif
478         }
479
480         if (skb->mark) {
481                 tmp_uint = htonl(skb->mark);
482                 NLA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
483         }
484
485         if (indev && skb->dev) {
486                 struct nfulnl_msg_packet_hw phw;
487                 int len = dev_parse_header(skb, phw.hw_addr);
488                 if (len > 0) {
489                         phw.hw_addrlen = htons(len);
490                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
491                 }
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                 NLA_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 NLA_PUT may goto */
509                         read_unlock_bh(&skb->sk->sk_callback_lock);
510                         NLA_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                 NLA_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                 NLA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
524         }
525
526         if (data_len) {
527                 struct nlattr *nla;
528                 int size = nla_attr_size(data_len);
529
530                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
531                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
532                         goto nlmsg_failure;
533                 }
534
535                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
536                 nla->nla_type = NFULA_PAYLOAD;
537                 nla->nla_len = size;
538
539                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
540                         BUG();
541         }
542
543         nlh->nlmsg_len = inst->skb->tail - old_tail;
544         return 0;
545
546 nlmsg_failure:
547         UDEBUG("nlmsg_failure\n");
548 nla_put_failure:
549         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
550         return -1;
551 }
552
553 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
554
555 static struct nf_loginfo default_loginfo = {
556         .type =         NF_LOG_TYPE_ULOG,
557         .u = {
558                 .ulog = {
559                         .copy_len       = 0xffff,
560                         .group          = 0,
561                         .qthreshold     = 1,
562                 },
563         },
564 };
565
566 /* log handler for internal netfilter logging api */
567 static void
568 nfulnl_log_packet(unsigned int pf,
569                   unsigned int hooknum,
570                   const struct sk_buff *skb,
571                   const struct net_device *in,
572                   const struct net_device *out,
573                   const struct nf_loginfo *li_user,
574                   const char *prefix)
575 {
576         unsigned int size, data_len;
577         struct nfulnl_instance *inst;
578         const struct nf_loginfo *li;
579         unsigned int qthreshold;
580         unsigned int plen;
581
582         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
583                 li = li_user;
584         else
585                 li = &default_loginfo;
586
587         inst = instance_lookup_get(li->u.ulog.group);
588         if (!inst)
589                 return;
590
591         plen = 0;
592         if (prefix)
593                 plen = strlen(prefix) + 1;
594
595         /* FIXME: do we want to make the size calculation conditional based on
596          * what is actually present?  way more branches and checks, but more
597          * memory efficient... */
598         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
599                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
600                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
601                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
602 #ifdef CONFIG_BRIDGE_NETFILTER
603                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
604                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
605 #endif
606                 + nla_total_size(sizeof(u_int32_t))     /* mark */
607                 + nla_total_size(sizeof(u_int32_t))     /* uid */
608                 + nla_total_size(plen)                  /* prefix */
609                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
610                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
611
612         UDEBUG("initial size=%u\n", size);
613
614         spin_lock_bh(&inst->lock);
615
616         if (inst->flags & NFULNL_CFG_F_SEQ)
617                 size += nla_total_size(sizeof(u_int32_t));
618         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
619                 size += nla_total_size(sizeof(u_int32_t));
620
621         qthreshold = inst->qthreshold;
622         /* per-rule qthreshold overrides per-instance */
623         if (qthreshold > li->u.ulog.qthreshold)
624                 qthreshold = li->u.ulog.qthreshold;
625
626         switch (inst->copy_mode) {
627         case NFULNL_COPY_META:
628         case NFULNL_COPY_NONE:
629                 data_len = 0;
630                 break;
631
632         case NFULNL_COPY_PACKET:
633                 if (inst->copy_range == 0
634                     || inst->copy_range > skb->len)
635                         data_len = skb->len;
636                 else
637                         data_len = inst->copy_range;
638
639                 size += nla_total_size(data_len);
640                 UDEBUG("copy_packet, therefore size now %u\n", size);
641                 break;
642
643         default:
644                 goto unlock_and_release;
645         }
646
647         if (inst->skb &&
648             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
649                 /* either the queue len is too high or we don't have
650                  * enough room in the skb left. flush to userspace. */
651                 UDEBUG("flushing old skb\n");
652
653                 __nfulnl_flush(inst);
654         }
655
656         if (!inst->skb) {
657                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
658                 if (!inst->skb)
659                         goto alloc_failure;
660         }
661
662         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
663         inst->qlen++;
664
665         __build_packet_message(inst, skb, data_len, pf,
666                                 hooknum, in, out, li, prefix, plen);
667
668         if (inst->qlen >= qthreshold)
669                 __nfulnl_flush(inst);
670         /* timer_pending always called within inst->lock, so there
671          * is no chance of a race here */
672         else if (!timer_pending(&inst->timer)) {
673                 instance_get(inst);
674                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
675                 add_timer(&inst->timer);
676         }
677
678 unlock_and_release:
679         spin_unlock_bh(&inst->lock);
680         instance_put(inst);
681         return;
682
683 alloc_failure:
684         UDEBUG("error allocating skb\n");
685         /* FIXME: statistics */
686         goto unlock_and_release;
687 }
688
689 static int
690 nfulnl_rcv_nl_event(struct notifier_block *this,
691                    unsigned long event, void *ptr)
692 {
693         struct netlink_notify *n = ptr;
694
695         if (event == NETLINK_URELEASE &&
696             n->protocol == NETLINK_NETFILTER && n->pid) {
697                 int i;
698
699                 /* destroy all instances for this pid */
700                 write_lock_bh(&instances_lock);
701                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
702                         struct hlist_node *tmp, *t2;
703                         struct nfulnl_instance *inst;
704                         struct hlist_head *head = &instance_table[i];
705
706                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
707                                 UDEBUG("node = %p\n", inst);
708                                 if ((n->net == &init_net) &&
709                                     (n->pid == inst->peer_pid))
710                                         __instance_destroy(inst);
711                         }
712                 }
713                 write_unlock_bh(&instances_lock);
714         }
715         return NOTIFY_DONE;
716 }
717
718 static struct notifier_block nfulnl_rtnl_notifier = {
719         .notifier_call  = nfulnl_rcv_nl_event,
720 };
721
722 static int
723 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
724                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
725 {
726         return -ENOTSUPP;
727 }
728
729 static struct nf_logger nfulnl_logger = {
730         .name   = "nfnetlink_log",
731         .logfn  = &nfulnl_log_packet,
732         .me     = THIS_MODULE,
733 };
734
735 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
736         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
737         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
738         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
739         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
740         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
741         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
742 };
743
744 static int
745 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
746                    struct nlmsghdr *nlh, struct nlattr *nfula[])
747 {
748         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
749         u_int16_t group_num = ntohs(nfmsg->res_id);
750         struct nfulnl_instance *inst;
751         int ret = 0;
752
753         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
754
755         inst = instance_lookup_get(group_num);
756         if (nfula[NFULA_CFG_CMD]) {
757                 u_int8_t pf = nfmsg->nfgen_family;
758                 struct nfulnl_msg_config_cmd *cmd;
759                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
760                 UDEBUG("found CFG_CMD for\n");
761
762                 switch (cmd->command) {
763                 case NFULNL_CFG_CMD_BIND:
764                         if (inst) {
765                                 ret = -EBUSY;
766                                 goto out_put;
767                         }
768
769                         inst = instance_create(group_num,
770                                                NETLINK_CB(skb).pid);
771                         if (!inst) {
772                                 ret = -EINVAL;
773                                 goto out;
774                         }
775                         break;
776                 case NFULNL_CFG_CMD_UNBIND:
777                         if (!inst) {
778                                 ret = -ENODEV;
779                                 goto out;
780                         }
781
782                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
783                                 ret = -EPERM;
784                                 goto out_put;
785                         }
786
787                         instance_destroy(inst);
788                         goto out;
789                 case NFULNL_CFG_CMD_PF_BIND:
790                         UDEBUG("registering log handler for pf=%u\n", pf);
791                         ret = nf_log_register(pf, &nfulnl_logger);
792                         break;
793                 case NFULNL_CFG_CMD_PF_UNBIND:
794                         UDEBUG("unregistering log handler for pf=%u\n", pf);
795                         /* This is a bug and a feature.  We cannot unregister
796                          * other handlers, like nfnetlink_inst can */
797                         nf_log_unregister_pf(pf);
798                         break;
799                 default:
800                         ret = -EINVAL;
801                         break;
802                 }
803
804                 if (!inst)
805                         goto out;
806         } else {
807                 if (!inst) {
808                         UDEBUG("no config command, and no instance for "
809                                 "group=%u pid=%u =>ENOENT\n",
810                                 group_num, NETLINK_CB(skb).pid);
811                         ret = -ENOENT;
812                         goto out;
813                 }
814
815                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
816                         UDEBUG("no config command, and wrong pid\n");
817                         ret = -EPERM;
818                         goto out_put;
819                 }
820         }
821
822         if (nfula[NFULA_CFG_MODE]) {
823                 struct nfulnl_msg_config_mode *params;
824                 params = nla_data(nfula[NFULA_CFG_MODE]);
825
826                 nfulnl_set_mode(inst, params->copy_mode,
827                                 ntohl(params->copy_range));
828         }
829
830         if (nfula[NFULA_CFG_TIMEOUT]) {
831                 __be32 timeout =
832                         *(__be32 *)nla_data(nfula[NFULA_CFG_TIMEOUT]);
833
834                 nfulnl_set_timeout(inst, ntohl(timeout));
835         }
836
837         if (nfula[NFULA_CFG_NLBUFSIZ]) {
838                 __be32 nlbufsiz =
839                         *(__be32 *)nla_data(nfula[NFULA_CFG_NLBUFSIZ]);
840
841                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
842         }
843
844         if (nfula[NFULA_CFG_QTHRESH]) {
845                 __be32 qthresh =
846                         *(__be32 *)nla_data(nfula[NFULA_CFG_QTHRESH]);
847
848                 nfulnl_set_qthresh(inst, ntohl(qthresh));
849         }
850
851         if (nfula[NFULA_CFG_FLAGS]) {
852                 __be16 flags =
853                         *(__be16 *)nla_data(nfula[NFULA_CFG_FLAGS]);
854                 nfulnl_set_flags(inst, ntohs(flags));
855         }
856
857 out_put:
858         instance_put(inst);
859 out:
860         return ret;
861 }
862
863 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
864         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
865                                     .attr_count = NFULA_MAX, },
866         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
867                                     .attr_count = NFULA_CFG_MAX,
868                                     .policy = nfula_cfg_policy },
869 };
870
871 static const struct nfnetlink_subsystem nfulnl_subsys = {
872         .name           = "log",
873         .subsys_id      = NFNL_SUBSYS_ULOG,
874         .cb_count       = NFULNL_MSG_MAX,
875         .cb             = nfulnl_cb,
876 };
877
878 #ifdef CONFIG_PROC_FS
879 struct iter_state {
880         unsigned int bucket;
881 };
882
883 static struct hlist_node *get_first(struct iter_state *st)
884 {
885         if (!st)
886                 return NULL;
887
888         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
889                 if (!hlist_empty(&instance_table[st->bucket]))
890                         return instance_table[st->bucket].first;
891         }
892         return NULL;
893 }
894
895 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
896 {
897         h = h->next;
898         while (!h) {
899                 if (++st->bucket >= INSTANCE_BUCKETS)
900                         return NULL;
901
902                 h = instance_table[st->bucket].first;
903         }
904         return h;
905 }
906
907 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
908 {
909         struct hlist_node *head;
910         head = get_first(st);
911
912         if (head)
913                 while (pos && (head = get_next(st, head)))
914                         pos--;
915         return pos ? NULL : head;
916 }
917
918 static void *seq_start(struct seq_file *seq, loff_t *pos)
919 {
920         read_lock_bh(&instances_lock);
921         return get_idx(seq->private, *pos);
922 }
923
924 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
925 {
926         (*pos)++;
927         return get_next(s->private, v);
928 }
929
930 static void seq_stop(struct seq_file *s, void *v)
931 {
932         read_unlock_bh(&instances_lock);
933 }
934
935 static int seq_show(struct seq_file *s, void *v)
936 {
937         const struct nfulnl_instance *inst = v;
938
939         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
940                           inst->group_num,
941                           inst->peer_pid, inst->qlen,
942                           inst->copy_mode, inst->copy_range,
943                           inst->flushtimeout, atomic_read(&inst->use));
944 }
945
946 static const struct seq_operations nful_seq_ops = {
947         .start  = seq_start,
948         .next   = seq_next,
949         .stop   = seq_stop,
950         .show   = seq_show,
951 };
952
953 static int nful_open(struct inode *inode, struct file *file)
954 {
955         struct seq_file *seq;
956         struct iter_state *is;
957         int ret;
958
959         is = kzalloc(sizeof(*is), GFP_KERNEL);
960         if (!is)
961                 return -ENOMEM;
962         ret = seq_open(file, &nful_seq_ops);
963         if (ret < 0)
964                 goto out_free;
965         seq = file->private_data;
966         seq->private = is;
967         return ret;
968 out_free:
969         kfree(is);
970         return ret;
971 }
972
973 static const struct file_operations nful_file_ops = {
974         .owner   = THIS_MODULE,
975         .open    = nful_open,
976         .read    = seq_read,
977         .llseek  = seq_lseek,
978         .release = seq_release_private,
979 };
980
981 #endif /* PROC_FS */
982
983 static int __init nfnetlink_log_init(void)
984 {
985         int i, status = -ENOMEM;
986 #ifdef CONFIG_PROC_FS
987         struct proc_dir_entry *proc_nful;
988 #endif
989
990         for (i = 0; i < INSTANCE_BUCKETS; i++)
991                 INIT_HLIST_HEAD(&instance_table[i]);
992
993         /* it's not really all that important to have a random value, so
994          * we can do this from the init function, even if there hasn't
995          * been that much entropy yet */
996         get_random_bytes(&hash_init, sizeof(hash_init));
997
998         netlink_register_notifier(&nfulnl_rtnl_notifier);
999         status = nfnetlink_subsys_register(&nfulnl_subsys);
1000         if (status < 0) {
1001                 printk(KERN_ERR "log: failed to create netlink socket\n");
1002                 goto cleanup_netlink_notifier;
1003         }
1004
1005 #ifdef CONFIG_PROC_FS
1006         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1007                                       proc_net_netfilter);
1008         if (!proc_nful)
1009                 goto cleanup_subsys;
1010         proc_nful->proc_fops = &nful_file_ops;
1011 #endif
1012         return status;
1013
1014 #ifdef CONFIG_PROC_FS
1015 cleanup_subsys:
1016         nfnetlink_subsys_unregister(&nfulnl_subsys);
1017 #endif
1018 cleanup_netlink_notifier:
1019         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1020         return status;
1021 }
1022
1023 static void __exit nfnetlink_log_fini(void)
1024 {
1025         nf_log_unregister(&nfulnl_logger);
1026 #ifdef CONFIG_PROC_FS
1027         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1028 #endif
1029         nfnetlink_subsys_unregister(&nfulnl_subsys);
1030         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1031 }
1032
1033 MODULE_DESCRIPTION("netfilter userspace logging");
1034 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1035 MODULE_LICENSE("GPL");
1036 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1037
1038 module_init(nfnetlink_log_init);
1039 module_exit(nfnetlink_log_fini);