[NETFILTER]: nfnetlink_log: fix sending of multipart messages
[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 #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 int __nfulnl_send(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                 /* timer "holds" one reference (we have one more) */
207                 if (del_timer(&inst->timer))
208                         instance_put(inst);
209                 if (inst->qlen)
210                         __nfulnl_send(inst);
211                 if (inst->skb) {
212                         kfree_skb(inst->skb);
213                         inst->skb = NULL;
214                 }
215         }
216         spin_unlock_bh(&inst->lock);
217
218         /* and finally put the refcount */
219         instance_put(inst);
220 }
221
222 static inline void
223 instance_destroy(struct nfulnl_instance *inst)
224 {
225         write_lock_bh(&instances_lock);
226         __instance_destroy(inst);
227         write_unlock_bh(&instances_lock);
228 }
229
230 static int
231 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
232                   unsigned int range)
233 {
234         int status = 0;
235
236         spin_lock_bh(&inst->lock);
237
238         switch (mode) {
239         case NFULNL_COPY_NONE:
240         case NFULNL_COPY_META:
241                 inst->copy_mode = mode;
242                 inst->copy_range = 0;
243                 break;
244
245         case NFULNL_COPY_PACKET:
246                 inst->copy_mode = mode;
247                 /* we're using struct nfattr which has 16bit nfa_len */
248                 if (range > 0xffff)
249                         inst->copy_range = 0xffff;
250                 else
251                         inst->copy_range = range;
252                 break;
253
254         default:
255                 status = -EINVAL;
256                 break;
257         }
258
259         spin_unlock_bh(&inst->lock);
260
261         return status;
262 }
263
264 static int
265 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
266 {
267         int status;
268
269         spin_lock_bh(&inst->lock);
270         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
271                 status = -ERANGE;
272         else if (nlbufsiz > 131072)
273                 status = -ERANGE;
274         else {
275                 inst->nlbufsiz = nlbufsiz;
276                 status = 0;
277         }
278         spin_unlock_bh(&inst->lock);
279
280         return status;
281 }
282
283 static int
284 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
285 {
286         spin_lock_bh(&inst->lock);
287         inst->flushtimeout = timeout;
288         spin_unlock_bh(&inst->lock);
289
290         return 0;
291 }
292
293 static int
294 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
295 {
296         spin_lock_bh(&inst->lock);
297         inst->qthreshold = qthresh;
298         spin_unlock_bh(&inst->lock);
299
300         return 0;
301 }
302
303 static int
304 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
305 {
306         spin_lock_bh(&inst->lock);
307         inst->flags = flags;
308         spin_unlock_bh(&inst->lock);
309
310         return 0;
311 }
312
313 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size,
314                                         unsigned int pkt_size)
315 {
316         struct sk_buff *skb;
317         unsigned int n;
318
319         UDEBUG("entered (%u, %u)\n", inst_size, pkt_size);
320
321         /* alloc skb which should be big enough for a whole multipart
322          * message.  WARNING: has to be <= 128k due to slab restrictions */
323
324         n = max(inst_size, pkt_size);
325         skb = alloc_skb(n, GFP_ATOMIC);
326         if (!skb) {
327                 PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
328                         inst_size);
329
330                 if (n > pkt_size) {
331                         /* try to allocate only as much as we need for current
332                          * packet */
333
334                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
335                         if (!skb)
336                                 PRINTR("nfnetlink_log: can't even alloc %u "
337                                        "bytes\n", pkt_size);
338                 }
339         }
340
341         return skb;
342 }
343
344 static int
345 __nfulnl_send(struct nfulnl_instance *inst)
346 {
347         int status = -1;
348
349         if (inst->qlen > 1)
350                 NLMSG_PUT(inst->skb, 0, 0,
351                           NLMSG_DONE,
352                           sizeof(struct nfgenmsg));
353
354         status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT);
355         if (status < 0) {
356                 UDEBUG("netlink_unicast() failed\n");
357                 /* FIXME: statistics */
358         }
359
360         inst->qlen = 0;
361         inst->skb = NULL;
362
363 nlmsg_failure:
364         return status;
365 }
366
367 static void nfulnl_timer(unsigned long data)
368 {
369         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
370
371         UDEBUG("timer function called, flushing buffer\n");
372
373         spin_lock_bh(&inst->lock);
374         if (inst->skb)
375                 __nfulnl_send(inst);
376         spin_unlock_bh(&inst->lock);
377         instance_put(inst);
378 }
379
380 /* This is an inline function, we don't really care about a long
381  * list of arguments */
382 static inline int
383 __build_packet_message(struct nfulnl_instance *inst,
384                         const struct sk_buff *skb,
385                         unsigned int data_len,
386                         unsigned int pf,
387                         unsigned int hooknum,
388                         const struct net_device *indev,
389                         const struct net_device *outdev,
390                         const struct nf_loginfo *li,
391                         const char *prefix, unsigned int plen)
392 {
393         struct nfulnl_msg_packet_hdr pmsg;
394         struct nlmsghdr *nlh;
395         struct nfgenmsg *nfmsg;
396         __be32 tmp_uint;
397         sk_buff_data_t old_tail = inst->skb->tail;
398
399         UDEBUG("entered\n");
400
401         nlh = NLMSG_PUT(inst->skb, 0, 0,
402                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
403                         sizeof(struct nfgenmsg));
404         nfmsg = NLMSG_DATA(nlh);
405         nfmsg->nfgen_family = pf;
406         nfmsg->version = NFNETLINK_V0;
407         nfmsg->res_id = htons(inst->group_num);
408
409         pmsg.hw_protocol        = skb->protocol;
410         pmsg.hook               = hooknum;
411
412         NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
413
414         if (prefix)
415                 NFA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
416
417         if (indev) {
418                 tmp_uint = htonl(indev->ifindex);
419 #ifndef CONFIG_BRIDGE_NETFILTER
420                 NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint),
421                         &tmp_uint);
422 #else
423                 if (pf == PF_BRIDGE) {
424                         /* Case 1: outdev is physical input device, we need to
425                          * look for bridge group (when called from
426                          * netfilter_bridge) */
427                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
428                                 sizeof(tmp_uint), &tmp_uint);
429                         /* this is the bridge group "brX" */
430                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
431                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
432                                 sizeof(tmp_uint), &tmp_uint);
433                 } else {
434                         /* Case 2: indev is bridge group, we need to look for
435                          * physical device (when called from ipv4) */
436                         NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV,
437                                 sizeof(tmp_uint), &tmp_uint);
438                         if (skb->nf_bridge && skb->nf_bridge->physindev) {
439                                 tmp_uint =
440                                     htonl(skb->nf_bridge->physindev->ifindex);
441                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSINDEV,
442                                         sizeof(tmp_uint), &tmp_uint);
443                         }
444                 }
445 #endif
446         }
447
448         if (outdev) {
449                 tmp_uint = htonl(outdev->ifindex);
450 #ifndef CONFIG_BRIDGE_NETFILTER
451                 NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint),
452                         &tmp_uint);
453 #else
454                 if (pf == PF_BRIDGE) {
455                         /* Case 1: outdev is physical output device, we need to
456                          * look for bridge group (when called from
457                          * netfilter_bridge) */
458                         NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
459                                 sizeof(tmp_uint), &tmp_uint);
460                         /* this is the bridge group "brX" */
461                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
462                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
463                                 sizeof(tmp_uint), &tmp_uint);
464                 } else {
465                         /* Case 2: indev is a bridge group, we need to look
466                          * for physical device (when called from ipv4) */
467                         NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV,
468                                 sizeof(tmp_uint), &tmp_uint);
469                         if (skb->nf_bridge && skb->nf_bridge->physoutdev) {
470                                 tmp_uint =
471                                     htonl(skb->nf_bridge->physoutdev->ifindex);
472                                 NFA_PUT(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
473                                         sizeof(tmp_uint), &tmp_uint);
474                         }
475                 }
476 #endif
477         }
478
479         if (skb->mark) {
480                 tmp_uint = htonl(skb->mark);
481                 NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint);
482         }
483
484         if (indev && skb->dev && skb->dev->hard_header_parse) {
485                 struct nfulnl_msg_packet_hw phw;
486                 int len = skb->dev->hard_header_parse((struct sk_buff *)skb,
487                                                     phw.hw_addr);
488                 phw.hw_addrlen = htons(len);
489                 NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
490         }
491
492         if (skb->tstamp.tv64) {
493                 struct nfulnl_msg_packet_timestamp ts;
494                 struct timeval tv = ktime_to_timeval(skb->tstamp);
495                 ts.sec = cpu_to_be64(tv.tv_sec);
496                 ts.usec = cpu_to_be64(tv.tv_usec);
497
498                 NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
499         }
500
501         /* UID */
502         if (skb->sk) {
503                 read_lock_bh(&skb->sk->sk_callback_lock);
504                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
505                         __be32 uid = htonl(skb->sk->sk_socket->file->f_uid);
506                         /* need to unlock here since NFA_PUT may goto */
507                         read_unlock_bh(&skb->sk->sk_callback_lock);
508                         NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid);
509                 } else
510                         read_unlock_bh(&skb->sk->sk_callback_lock);
511         }
512
513         /* local sequence number */
514         if (inst->flags & NFULNL_CFG_F_SEQ) {
515                 tmp_uint = htonl(inst->seq++);
516                 NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
517         }
518         /* global sequence number */
519         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
520                 tmp_uint = htonl(atomic_inc_return(&global_seq));
521                 NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
522         }
523
524         if (data_len) {
525                 struct nfattr *nfa;
526                 int size = NFA_LENGTH(data_len);
527
528                 if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) {
529                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
530                         goto nlmsg_failure;
531                 }
532
533                 nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size));
534                 nfa->nfa_type = NFULA_PAYLOAD;
535                 nfa->nfa_len = size;
536
537                 if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len))
538                         BUG();
539         }
540
541         nlh->nlmsg_len = inst->skb->tail - old_tail;
542         return 0;
543
544 nlmsg_failure:
545         UDEBUG("nlmsg_failure\n");
546 nfattr_failure:
547         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
548         return -1;
549 }
550
551 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
552
553 static struct nf_loginfo default_loginfo = {
554         .type =         NF_LOG_TYPE_ULOG,
555         .u = {
556                 .ulog = {
557                         .copy_len       = 0xffff,
558                         .group          = 0,
559                         .qthreshold     = 1,
560                 },
561         },
562 };
563
564 /* log handler for internal netfilter logging api */
565 static void
566 nfulnl_log_packet(unsigned int pf,
567                   unsigned int hooknum,
568                   const struct sk_buff *skb,
569                   const struct net_device *in,
570                   const struct net_device *out,
571                   const struct nf_loginfo *li_user,
572                   const char *prefix)
573 {
574         unsigned int size, data_len;
575         struct nfulnl_instance *inst;
576         const struct nf_loginfo *li;
577         unsigned int qthreshold;
578         unsigned int plen;
579
580         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
581                 li = li_user;
582         else
583                 li = &default_loginfo;
584
585         inst = instance_lookup_get(li->u.ulog.group);
586         if (!inst)
587                 return;
588
589         plen = 0;
590         if (prefix)
591                 plen = strlen(prefix) + 1;
592
593         /* all macros expand to constant values at compile time */
594         /* FIXME: do we want to make the size calculation conditional based on
595          * what is actually present?  way more branches and checks, but more
596          * memory efficient... */
597         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
598                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr))
599                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
600                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
601 #ifdef CONFIG_BRIDGE_NETFILTER
602                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
603                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
604 #endif
605                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
606                 + NFA_SPACE(sizeof(u_int32_t))  /* uid */
607                 + NFA_SPACE(plen)               /* prefix */
608                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw))
609                 + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp));
610
611         UDEBUG("initial size=%u\n", size);
612
613         spin_lock_bh(&inst->lock);
614
615         if (inst->flags & NFULNL_CFG_F_SEQ)
616                 size += NFA_SPACE(sizeof(u_int32_t));
617         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
618                 size += NFA_SPACE(sizeof(u_int32_t));
619
620         qthreshold = inst->qthreshold;
621         /* per-rule qthreshold overrides per-instance */
622         if (qthreshold > li->u.ulog.qthreshold)
623                 qthreshold = li->u.ulog.qthreshold;
624
625         switch (inst->copy_mode) {
626         case NFULNL_COPY_META:
627         case NFULNL_COPY_NONE:
628                 data_len = 0;
629                 break;
630
631         case NFULNL_COPY_PACKET:
632                 if (inst->copy_range == 0
633                     || inst->copy_range > skb->len)
634                         data_len = skb->len;
635                 else
636                         data_len = inst->copy_range;
637
638                 size += NFA_SPACE(data_len);
639                 UDEBUG("copy_packet, therefore size now %u\n", size);
640                 break;
641
642         default:
643                 goto unlock_and_release;
644         }
645
646         if (inst->qlen >= qthreshold ||
647             (inst->skb && size >
648              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                 /* timer "holds" one reference (we have another one) */
654                 if (del_timer(&inst->timer))
655                         instance_put(inst);
656                 __nfulnl_send(inst);
657         }
658
659         if (!inst->skb) {
660                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
661                 if (!inst->skb)
662                         goto alloc_failure;
663         }
664
665         UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold);
666         inst->qlen++;
667
668         __build_packet_message(inst, skb, data_len, pf,
669                                 hooknum, in, out, li, prefix, plen);
670
671         /* timer_pending always called within inst->lock, so there
672          * is no chance of a race here */
673         if (!timer_pending(&inst->timer)) {
674                 instance_get(inst);
675                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
676                 add_timer(&inst->timer);
677         }
678
679 unlock_and_release:
680         spin_unlock_bh(&inst->lock);
681         instance_put(inst);
682         return;
683
684 alloc_failure:
685         UDEBUG("error allocating skb\n");
686         /* FIXME: statistics */
687         goto unlock_and_release;
688 }
689
690 static int
691 nfulnl_rcv_nl_event(struct notifier_block *this,
692                    unsigned long event, void *ptr)
693 {
694         struct netlink_notify *n = ptr;
695
696         if (event == NETLINK_URELEASE &&
697             n->protocol == NETLINK_NETFILTER && n->pid) {
698                 int i;
699
700                 /* destroy all instances for this pid */
701                 write_lock_bh(&instances_lock);
702                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
703                         struct hlist_node *tmp, *t2;
704                         struct nfulnl_instance *inst;
705                         struct hlist_head *head = &instance_table[i];
706
707                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
708                                 UDEBUG("node = %p\n", inst);
709                                 if (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 nfattr *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 int nfula_min[NFULA_MAX] = {
736         [NFULA_PACKET_HDR-1]    = sizeof(struct nfulnl_msg_packet_hdr),
737         [NFULA_MARK-1]          = sizeof(u_int32_t),
738         [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
739         [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
740         [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
741         [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
742         [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
743         [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
744         [NFULA_PAYLOAD-1]       = 0,
745         [NFULA_PREFIX-1]        = 0,
746         [NFULA_UID-1]           = sizeof(u_int32_t),
747         [NFULA_SEQ-1]           = sizeof(u_int32_t),
748         [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
749 };
750
751 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
752         [NFULA_CFG_CMD-1]       = sizeof(struct nfulnl_msg_config_cmd),
753         [NFULA_CFG_MODE-1]      = sizeof(struct nfulnl_msg_config_mode),
754         [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
755         [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
756         [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
757         [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
758 };
759
760 static int
761 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
762                    struct nlmsghdr *nlh, struct nfattr *nfula[])
763 {
764         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
765         u_int16_t group_num = ntohs(nfmsg->res_id);
766         struct nfulnl_instance *inst;
767         int ret = 0;
768
769         UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
770
771         if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) {
772                 UDEBUG("bad attribute size\n");
773                 return -EINVAL;
774         }
775
776         inst = instance_lookup_get(group_num);
777         if (nfula[NFULA_CFG_CMD-1]) {
778                 u_int8_t pf = nfmsg->nfgen_family;
779                 struct nfulnl_msg_config_cmd *cmd;
780                 cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]);
781                 UDEBUG("found CFG_CMD for\n");
782
783                 switch (cmd->command) {
784                 case NFULNL_CFG_CMD_BIND:
785                         if (inst) {
786                                 ret = -EBUSY;
787                                 goto out_put;
788                         }
789
790                         inst = instance_create(group_num,
791                                                NETLINK_CB(skb).pid);
792                         if (!inst) {
793                                 ret = -EINVAL;
794                                 goto out;
795                         }
796                         break;
797                 case NFULNL_CFG_CMD_UNBIND:
798                         if (!inst) {
799                                 ret = -ENODEV;
800                                 goto out;
801                         }
802
803                         if (inst->peer_pid != NETLINK_CB(skb).pid) {
804                                 ret = -EPERM;
805                                 goto out_put;
806                         }
807
808                         instance_destroy(inst);
809                         goto out;
810                 case NFULNL_CFG_CMD_PF_BIND:
811                         UDEBUG("registering log handler for pf=%u\n", pf);
812                         ret = nf_log_register(pf, &nfulnl_logger);
813                         break;
814                 case NFULNL_CFG_CMD_PF_UNBIND:
815                         UDEBUG("unregistering log handler for pf=%u\n", pf);
816                         /* This is a bug and a feature.  We cannot unregister
817                          * other handlers, like nfnetlink_inst can */
818                         nf_log_unregister_pf(pf);
819                         break;
820                 default:
821                         ret = -EINVAL;
822                         break;
823                 }
824
825                 if (!inst)
826                         goto out;
827         } else {
828                 if (!inst) {
829                         UDEBUG("no config command, and no instance for "
830                                 "group=%u pid=%u =>ENOENT\n",
831                                 group_num, NETLINK_CB(skb).pid);
832                         ret = -ENOENT;
833                         goto out;
834                 }
835
836                 if (inst->peer_pid != NETLINK_CB(skb).pid) {
837                         UDEBUG("no config command, and wrong pid\n");
838                         ret = -EPERM;
839                         goto out_put;
840                 }
841         }
842
843         if (nfula[NFULA_CFG_MODE-1]) {
844                 struct nfulnl_msg_config_mode *params;
845                 params = NFA_DATA(nfula[NFULA_CFG_MODE-1]);
846
847                 nfulnl_set_mode(inst, params->copy_mode,
848                                 ntohl(params->copy_range));
849         }
850
851         if (nfula[NFULA_CFG_TIMEOUT-1]) {
852                 __be32 timeout =
853                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]);
854
855                 nfulnl_set_timeout(inst, ntohl(timeout));
856         }
857
858         if (nfula[NFULA_CFG_NLBUFSIZ-1]) {
859                 __be32 nlbufsiz =
860                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]);
861
862                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
863         }
864
865         if (nfula[NFULA_CFG_QTHRESH-1]) {
866                 __be32 qthresh =
867                         *(__be32 *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]);
868
869                 nfulnl_set_qthresh(inst, ntohl(qthresh));
870         }
871
872         if (nfula[NFULA_CFG_FLAGS-1]) {
873                 __be16 flags =
874                         *(__be16 *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
875                 nfulnl_set_flags(inst, ntohs(flags));
876         }
877
878 out_put:
879         instance_put(inst);
880 out:
881         return ret;
882 }
883
884 static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
885         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
886                                     .attr_count = NFULA_MAX, },
887         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
888                                     .attr_count = NFULA_CFG_MAX, },
889 };
890
891 static struct nfnetlink_subsystem nfulnl_subsys = {
892         .name           = "log",
893         .subsys_id      = NFNL_SUBSYS_ULOG,
894         .cb_count       = NFULNL_MSG_MAX,
895         .cb             = nfulnl_cb,
896 };
897
898 #ifdef CONFIG_PROC_FS
899 struct iter_state {
900         unsigned int bucket;
901 };
902
903 static struct hlist_node *get_first(struct iter_state *st)
904 {
905         if (!st)
906                 return NULL;
907
908         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
909                 if (!hlist_empty(&instance_table[st->bucket]))
910                         return instance_table[st->bucket].first;
911         }
912         return NULL;
913 }
914
915 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
916 {
917         h = h->next;
918         while (!h) {
919                 if (++st->bucket >= INSTANCE_BUCKETS)
920                         return NULL;
921
922                 h = instance_table[st->bucket].first;
923         }
924         return h;
925 }
926
927 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
928 {
929         struct hlist_node *head;
930         head = get_first(st);
931
932         if (head)
933                 while (pos && (head = get_next(st, head)))
934                         pos--;
935         return pos ? NULL : head;
936 }
937
938 static void *seq_start(struct seq_file *seq, loff_t *pos)
939 {
940         read_lock_bh(&instances_lock);
941         return get_idx(seq->private, *pos);
942 }
943
944 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
945 {
946         (*pos)++;
947         return get_next(s->private, v);
948 }
949
950 static void seq_stop(struct seq_file *s, void *v)
951 {
952         read_unlock_bh(&instances_lock);
953 }
954
955 static int seq_show(struct seq_file *s, void *v)
956 {
957         const struct nfulnl_instance *inst = v;
958
959         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
960                           inst->group_num,
961                           inst->peer_pid, inst->qlen,
962                           inst->copy_mode, inst->copy_range,
963                           inst->flushtimeout, atomic_read(&inst->use));
964 }
965
966 static const struct seq_operations nful_seq_ops = {
967         .start  = seq_start,
968         .next   = seq_next,
969         .stop   = seq_stop,
970         .show   = seq_show,
971 };
972
973 static int nful_open(struct inode *inode, struct file *file)
974 {
975         struct seq_file *seq;
976         struct iter_state *is;
977         int ret;
978
979         is = kzalloc(sizeof(*is), GFP_KERNEL);
980         if (!is)
981                 return -ENOMEM;
982         ret = seq_open(file, &nful_seq_ops);
983         if (ret < 0)
984                 goto out_free;
985         seq = file->private_data;
986         seq->private = is;
987         return ret;
988 out_free:
989         kfree(is);
990         return ret;
991 }
992
993 static const struct file_operations nful_file_ops = {
994         .owner   = THIS_MODULE,
995         .open    = nful_open,
996         .read    = seq_read,
997         .llseek  = seq_lseek,
998         .release = seq_release_private,
999 };
1000
1001 #endif /* PROC_FS */
1002
1003 static int __init nfnetlink_log_init(void)
1004 {
1005         int i, status = -ENOMEM;
1006 #ifdef CONFIG_PROC_FS
1007         struct proc_dir_entry *proc_nful;
1008 #endif
1009
1010         for (i = 0; i < INSTANCE_BUCKETS; i++)
1011                 INIT_HLIST_HEAD(&instance_table[i]);
1012
1013         /* it's not really all that important to have a random value, so
1014          * we can do this from the init function, even if there hasn't
1015          * been that much entropy yet */
1016         get_random_bytes(&hash_init, sizeof(hash_init));
1017
1018         netlink_register_notifier(&nfulnl_rtnl_notifier);
1019         status = nfnetlink_subsys_register(&nfulnl_subsys);
1020         if (status < 0) {
1021                 printk(KERN_ERR "log: failed to create netlink socket\n");
1022                 goto cleanup_netlink_notifier;
1023         }
1024
1025 #ifdef CONFIG_PROC_FS
1026         proc_nful = create_proc_entry("nfnetlink_log", 0440,
1027                                       proc_net_netfilter);
1028         if (!proc_nful)
1029                 goto cleanup_subsys;
1030         proc_nful->proc_fops = &nful_file_ops;
1031 #endif
1032         return status;
1033
1034 #ifdef CONFIG_PROC_FS
1035 cleanup_subsys:
1036         nfnetlink_subsys_unregister(&nfulnl_subsys);
1037 #endif
1038 cleanup_netlink_notifier:
1039         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1040         return status;
1041 }
1042
1043 static void __exit nfnetlink_log_fini(void)
1044 {
1045         nf_log_unregister(&nfulnl_logger);
1046 #ifdef CONFIG_PROC_FS
1047         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1048 #endif
1049         nfnetlink_subsys_unregister(&nfulnl_subsys);
1050         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1051 }
1052
1053 MODULE_DESCRIPTION("netfilter userspace logging");
1054 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1055 MODULE_LICENSE("GPL");
1056 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1057
1058 module_init(nfnetlink_log_init);
1059 module_exit(nfnetlink_log_fini);