netfilter: add __rcu annotations
[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 <linux/slab.h>
32 #include <net/sock.h>
33 #include <net/netfilter/nf_log.h>
34 #include <net/netfilter/nfnetlink_log.h>
35
36 #include <asm/atomic.h>
37
38 #ifdef CONFIG_BRIDGE_NETFILTER
39 #include "../bridge/br_private.h"
40 #endif
41
42 #define NFULNL_NLBUFSIZ_DEFAULT NLMSG_GOODSIZE
43 #define NFULNL_TIMEOUT_DEFAULT  100     /* every second */
44 #define NFULNL_QTHRESH_DEFAULT  100     /* 100 packets */
45 #define NFULNL_COPY_RANGE_MAX   0xFFFF  /* max packet size is limited by 16-bit struct nfattr nfa_len field */
46
47 #define PRINTR(x, args...)      do { if (net_ratelimit()) \
48                                      printk(x, ## args); } while (0);
49
50 struct nfulnl_instance {
51         struct hlist_node hlist;        /* global list of instances */
52         spinlock_t lock;
53         atomic_t use;                   /* use count */
54
55         unsigned int qlen;              /* number of nlmsgs in skb */
56         struct sk_buff *skb;            /* pre-allocatd skb */
57         struct timer_list timer;
58         int peer_pid;                   /* PID of the peer process */
59
60         /* configurable parameters */
61         unsigned int flushtimeout;      /* timeout until queue flush */
62         unsigned int nlbufsiz;          /* netlink buffer allocation size */
63         unsigned int qthreshold;        /* threshold of the queue */
64         u_int32_t copy_range;
65         u_int32_t seq;                  /* instance-local sequential counter */
66         u_int16_t group_num;            /* number of this queue */
67         u_int16_t flags;
68         u_int8_t copy_mode;
69         struct rcu_head rcu;
70 };
71
72 static DEFINE_SPINLOCK(instances_lock);
73 static atomic_t global_seq;
74
75 #define INSTANCE_BUCKETS        16
76 static struct hlist_head instance_table[INSTANCE_BUCKETS];
77 static unsigned int hash_init;
78
79 static inline u_int8_t instance_hashfn(u_int16_t group_num)
80 {
81         return ((group_num & 0xff) % INSTANCE_BUCKETS);
82 }
83
84 static struct nfulnl_instance *
85 __instance_lookup(u_int16_t group_num)
86 {
87         struct hlist_head *head;
88         struct hlist_node *pos;
89         struct nfulnl_instance *inst;
90
91         head = &instance_table[instance_hashfn(group_num)];
92         hlist_for_each_entry_rcu(inst, pos, head, hlist) {
93                 if (inst->group_num == group_num)
94                         return inst;
95         }
96         return NULL;
97 }
98
99 static inline void
100 instance_get(struct nfulnl_instance *inst)
101 {
102         atomic_inc(&inst->use);
103 }
104
105 static struct nfulnl_instance *
106 instance_lookup_get(u_int16_t group_num)
107 {
108         struct nfulnl_instance *inst;
109
110         rcu_read_lock_bh();
111         inst = __instance_lookup(group_num);
112         if (inst && !atomic_inc_not_zero(&inst->use))
113                 inst = NULL;
114         rcu_read_unlock_bh();
115
116         return inst;
117 }
118
119 static void nfulnl_instance_free_rcu(struct rcu_head *head)
120 {
121         kfree(container_of(head, struct nfulnl_instance, rcu));
122         module_put(THIS_MODULE);
123 }
124
125 static void
126 instance_put(struct nfulnl_instance *inst)
127 {
128         if (inst && atomic_dec_and_test(&inst->use))
129                 call_rcu_bh(&inst->rcu, nfulnl_instance_free_rcu);
130 }
131
132 static void nfulnl_timer(unsigned long data);
133
134 static struct nfulnl_instance *
135 instance_create(u_int16_t group_num, int pid)
136 {
137         struct nfulnl_instance *inst;
138         int err;
139
140         spin_lock_bh(&instances_lock);
141         if (__instance_lookup(group_num)) {
142                 err = -EEXIST;
143                 goto out_unlock;
144         }
145
146         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
147         if (!inst) {
148                 err = -ENOMEM;
149                 goto out_unlock;
150         }
151
152         if (!try_module_get(THIS_MODULE)) {
153                 kfree(inst);
154                 err = -EAGAIN;
155                 goto out_unlock;
156         }
157
158         INIT_HLIST_NODE(&inst->hlist);
159         spin_lock_init(&inst->lock);
160         /* needs to be two, since we _put() after creation */
161         atomic_set(&inst->use, 2);
162
163         setup_timer(&inst->timer, nfulnl_timer, (unsigned long)inst);
164
165         inst->peer_pid = pid;
166         inst->group_num = group_num;
167
168         inst->qthreshold        = NFULNL_QTHRESH_DEFAULT;
169         inst->flushtimeout      = NFULNL_TIMEOUT_DEFAULT;
170         inst->nlbufsiz          = NFULNL_NLBUFSIZ_DEFAULT;
171         inst->copy_mode         = NFULNL_COPY_PACKET;
172         inst->copy_range        = NFULNL_COPY_RANGE_MAX;
173
174         hlist_add_head_rcu(&inst->hlist,
175                        &instance_table[instance_hashfn(group_num)]);
176
177         spin_unlock_bh(&instances_lock);
178
179         return inst;
180
181 out_unlock:
182         spin_unlock_bh(&instances_lock);
183         return ERR_PTR(err);
184 }
185
186 static void __nfulnl_flush(struct nfulnl_instance *inst);
187
188 /* called with BH disabled */
189 static void
190 __instance_destroy(struct nfulnl_instance *inst)
191 {
192         /* first pull it out of the global list */
193         hlist_del_rcu(&inst->hlist);
194
195         /* then flush all pending packets from skb */
196
197         spin_lock(&inst->lock);
198
199         /* lockless readers wont be able to use us */
200         inst->copy_mode = NFULNL_COPY_DISABLED;
201
202         if (inst->skb)
203                 __nfulnl_flush(inst);
204         spin_unlock(&inst->lock);
205
206         /* and finally put the refcount */
207         instance_put(inst);
208 }
209
210 static inline void
211 instance_destroy(struct nfulnl_instance *inst)
212 {
213         spin_lock_bh(&instances_lock);
214         __instance_destroy(inst);
215         spin_unlock_bh(&instances_lock);
216 }
217
218 static int
219 nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode,
220                   unsigned int range)
221 {
222         int status = 0;
223
224         spin_lock_bh(&inst->lock);
225
226         switch (mode) {
227         case NFULNL_COPY_NONE:
228         case NFULNL_COPY_META:
229                 inst->copy_mode = mode;
230                 inst->copy_range = 0;
231                 break;
232
233         case NFULNL_COPY_PACKET:
234                 inst->copy_mode = mode;
235                 inst->copy_range = min_t(unsigned int,
236                                          range, NFULNL_COPY_RANGE_MAX);
237                 break;
238
239         default:
240                 status = -EINVAL;
241                 break;
242         }
243
244         spin_unlock_bh(&inst->lock);
245
246         return status;
247 }
248
249 static int
250 nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz)
251 {
252         int status;
253
254         spin_lock_bh(&inst->lock);
255         if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT)
256                 status = -ERANGE;
257         else if (nlbufsiz > 131072)
258                 status = -ERANGE;
259         else {
260                 inst->nlbufsiz = nlbufsiz;
261                 status = 0;
262         }
263         spin_unlock_bh(&inst->lock);
264
265         return status;
266 }
267
268 static int
269 nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout)
270 {
271         spin_lock_bh(&inst->lock);
272         inst->flushtimeout = timeout;
273         spin_unlock_bh(&inst->lock);
274
275         return 0;
276 }
277
278 static int
279 nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
280 {
281         spin_lock_bh(&inst->lock);
282         inst->qthreshold = qthresh;
283         spin_unlock_bh(&inst->lock);
284
285         return 0;
286 }
287
288 static int
289 nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
290 {
291         spin_lock_bh(&inst->lock);
292         inst->flags = flags;
293         spin_unlock_bh(&inst->lock);
294
295         return 0;
296 }
297
298 static struct sk_buff *
299 nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
300 {
301         struct sk_buff *skb;
302         unsigned int n;
303
304         /* alloc skb which should be big enough for a whole multipart
305          * message.  WARNING: has to be <= 128k due to slab restrictions */
306
307         n = max(inst_size, pkt_size);
308         skb = alloc_skb(n, GFP_ATOMIC);
309         if (!skb) {
310                 pr_notice("nfnetlink_log: can't alloc whole buffer (%u bytes)\n",
311                         inst_size);
312
313                 if (n > pkt_size) {
314                         /* try to allocate only as much as we need for current
315                          * packet */
316
317                         skb = alloc_skb(pkt_size, GFP_ATOMIC);
318                         if (!skb)
319                                 pr_err("nfnetlink_log: can't even alloc %u "
320                                        "bytes\n", pkt_size);
321                 }
322         }
323
324         return skb;
325 }
326
327 static int
328 __nfulnl_send(struct nfulnl_instance *inst)
329 {
330         int status = -1;
331
332         if (inst->qlen > 1)
333                 NLMSG_PUT(inst->skb, 0, 0,
334                           NLMSG_DONE,
335                           sizeof(struct nfgenmsg));
336
337         status = nfnetlink_unicast(inst->skb, &init_net, inst->peer_pid,
338                                    MSG_DONTWAIT);
339
340         inst->qlen = 0;
341         inst->skb = NULL;
342
343 nlmsg_failure:
344         return status;
345 }
346
347 static void
348 __nfulnl_flush(struct nfulnl_instance *inst)
349 {
350         /* timer holds a reference */
351         if (del_timer(&inst->timer))
352                 instance_put(inst);
353         if (inst->skb)
354                 __nfulnl_send(inst);
355 }
356
357 static void
358 nfulnl_timer(unsigned long data)
359 {
360         struct nfulnl_instance *inst = (struct nfulnl_instance *)data;
361
362         spin_lock_bh(&inst->lock);
363         if (inst->skb)
364                 __nfulnl_send(inst);
365         spin_unlock_bh(&inst->lock);
366         instance_put(inst);
367 }
368
369 /* This is an inline function, we don't really care about a long
370  * list of arguments */
371 static inline int
372 __build_packet_message(struct nfulnl_instance *inst,
373                         const struct sk_buff *skb,
374                         unsigned int data_len,
375                         u_int8_t pf,
376                         unsigned int hooknum,
377                         const struct net_device *indev,
378                         const struct net_device *outdev,
379                         const struct nf_loginfo *li,
380                         const char *prefix, unsigned int plen)
381 {
382         struct nfulnl_msg_packet_hdr pmsg;
383         struct nlmsghdr *nlh;
384         struct nfgenmsg *nfmsg;
385         __be32 tmp_uint;
386         sk_buff_data_t old_tail = inst->skb->tail;
387
388         nlh = NLMSG_PUT(inst->skb, 0, 0,
389                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
390                         sizeof(struct nfgenmsg));
391         nfmsg = NLMSG_DATA(nlh);
392         nfmsg->nfgen_family = pf;
393         nfmsg->version = NFNETLINK_V0;
394         nfmsg->res_id = htons(inst->group_num);
395
396         pmsg.hw_protocol        = skb->protocol;
397         pmsg.hook               = hooknum;
398
399         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
400
401         if (prefix)
402                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
403
404         if (indev) {
405 #ifndef CONFIG_BRIDGE_NETFILTER
406                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
407                              htonl(indev->ifindex));
408 #else
409                 if (pf == PF_BRIDGE) {
410                         /* Case 1: outdev is physical input device, we need to
411                          * look for bridge group (when called from
412                          * netfilter_bridge) */
413                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
414                                      htonl(indev->ifindex));
415                         /* this is the bridge group "brX" */
416                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
417                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
418                                      htonl(br_port_get_rcu(indev)->br->dev->ifindex));
419                 } else {
420                         /* Case 2: indev is bridge group, we need to look for
421                          * physical device (when called from ipv4) */
422                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
423                                      htonl(indev->ifindex));
424                         if (skb->nf_bridge && skb->nf_bridge->physindev)
425                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
426                                              htonl(skb->nf_bridge->physindev->ifindex));
427                 }
428 #endif
429         }
430
431         if (outdev) {
432                 tmp_uint = htonl(outdev->ifindex);
433 #ifndef CONFIG_BRIDGE_NETFILTER
434                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
435                              htonl(outdev->ifindex));
436 #else
437                 if (pf == PF_BRIDGE) {
438                         /* Case 1: outdev is physical output device, we need to
439                          * look for bridge group (when called from
440                          * netfilter_bridge) */
441                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
442                                      htonl(outdev->ifindex));
443                         /* this is the bridge group "brX" */
444                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
445                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
446                                      htonl(br_port_get_rcu(outdev)->br->dev->ifindex));
447                 } else {
448                         /* Case 2: indev is a bridge group, we need to look
449                          * for physical device (when called from ipv4) */
450                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
451                                      htonl(outdev->ifindex));
452                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
453                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
454                                              htonl(skb->nf_bridge->physoutdev->ifindex));
455                 }
456 #endif
457         }
458
459         if (skb->mark)
460                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
461
462         if (indev && skb->dev) {
463                 struct nfulnl_msg_packet_hw phw;
464                 int len = dev_parse_header(skb, phw.hw_addr);
465                 if (len > 0) {
466                         phw.hw_addrlen = htons(len);
467                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
468                 }
469         }
470
471         if (indev && skb_mac_header_was_set(skb)) {
472                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
473                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
474                              htons(skb->dev->hard_header_len));
475                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
476                         skb_mac_header(skb));
477         }
478
479         if (skb->tstamp.tv64) {
480                 struct nfulnl_msg_packet_timestamp ts;
481                 struct timeval tv = ktime_to_timeval(skb->tstamp);
482                 ts.sec = cpu_to_be64(tv.tv_sec);
483                 ts.usec = cpu_to_be64(tv.tv_usec);
484
485                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
486         }
487
488         /* UID */
489         if (skb->sk) {
490                 read_lock_bh(&skb->sk->sk_callback_lock);
491                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
492                         struct file *file = skb->sk->sk_socket->file;
493                         __be32 uid = htonl(file->f_cred->fsuid);
494                         __be32 gid = htonl(file->f_cred->fsgid);
495                         /* need to unlock here since NLA_PUT may goto */
496                         read_unlock_bh(&skb->sk->sk_callback_lock);
497                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
498                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
499                 } else
500                         read_unlock_bh(&skb->sk->sk_callback_lock);
501         }
502
503         /* local sequence number */
504         if (inst->flags & NFULNL_CFG_F_SEQ)
505                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
506
507         /* global sequence number */
508         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
509                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
510                              htonl(atomic_inc_return(&global_seq)));
511
512         if (data_len) {
513                 struct nlattr *nla;
514                 int size = nla_attr_size(data_len);
515
516                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
517                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
518                         goto nlmsg_failure;
519                 }
520
521                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
522                 nla->nla_type = NFULA_PAYLOAD;
523                 nla->nla_len = size;
524
525                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
526                         BUG();
527         }
528
529         nlh->nlmsg_len = inst->skb->tail - old_tail;
530         return 0;
531
532 nlmsg_failure:
533 nla_put_failure:
534         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
535         return -1;
536 }
537
538 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
539
540 static struct nf_loginfo default_loginfo = {
541         .type =         NF_LOG_TYPE_ULOG,
542         .u = {
543                 .ulog = {
544                         .copy_len       = 0xffff,
545                         .group          = 0,
546                         .qthreshold     = 1,
547                 },
548         },
549 };
550
551 /* log handler for internal netfilter logging api */
552 void
553 nfulnl_log_packet(u_int8_t pf,
554                   unsigned int hooknum,
555                   const struct sk_buff *skb,
556                   const struct net_device *in,
557                   const struct net_device *out,
558                   const struct nf_loginfo *li_user,
559                   const char *prefix)
560 {
561         unsigned int size, data_len;
562         struct nfulnl_instance *inst;
563         const struct nf_loginfo *li;
564         unsigned int qthreshold;
565         unsigned int plen;
566
567         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
568                 li = li_user;
569         else
570                 li = &default_loginfo;
571
572         inst = instance_lookup_get(li->u.ulog.group);
573         if (!inst)
574                 return;
575
576         plen = 0;
577         if (prefix)
578                 plen = strlen(prefix) + 1;
579
580         /* FIXME: do we want to make the size calculation conditional based on
581          * what is actually present?  way more branches and checks, but more
582          * memory efficient... */
583         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
584                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
585                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
586                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
587 #ifdef CONFIG_BRIDGE_NETFILTER
588                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
589                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
590 #endif
591                 + nla_total_size(sizeof(u_int32_t))     /* mark */
592                 + nla_total_size(sizeof(u_int32_t))     /* uid */
593                 + nla_total_size(sizeof(u_int32_t))     /* gid */
594                 + nla_total_size(plen)                  /* prefix */
595                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
596                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
597
598         if (in && skb_mac_header_was_set(skb)) {
599                 size +=   nla_total_size(skb->dev->hard_header_len)
600                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
601                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
602         }
603
604         spin_lock_bh(&inst->lock);
605
606         if (inst->flags & NFULNL_CFG_F_SEQ)
607                 size += nla_total_size(sizeof(u_int32_t));
608         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
609                 size += nla_total_size(sizeof(u_int32_t));
610
611         qthreshold = inst->qthreshold;
612         /* per-rule qthreshold overrides per-instance */
613         if (li->u.ulog.qthreshold)
614                 if (qthreshold > li->u.ulog.qthreshold)
615                         qthreshold = li->u.ulog.qthreshold;
616
617
618         switch (inst->copy_mode) {
619         case NFULNL_COPY_META:
620         case NFULNL_COPY_NONE:
621                 data_len = 0;
622                 break;
623
624         case NFULNL_COPY_PACKET:
625                 if (inst->copy_range == 0
626                     || inst->copy_range > skb->len)
627                         data_len = skb->len;
628                 else
629                         data_len = inst->copy_range;
630
631                 size += nla_total_size(data_len);
632                 break;
633
634         case NFULNL_COPY_DISABLED:
635         default:
636                 goto unlock_and_release;
637         }
638
639         if (inst->skb &&
640             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
641                 /* either the queue len is too high or we don't have
642                  * enough room in the skb left. flush to userspace. */
643                 __nfulnl_flush(inst);
644         }
645
646         if (!inst->skb) {
647                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
648                 if (!inst->skb)
649                         goto alloc_failure;
650         }
651
652         inst->qlen++;
653
654         __build_packet_message(inst, skb, data_len, pf,
655                                 hooknum, in, out, li, prefix, plen);
656
657         if (inst->qlen >= qthreshold)
658                 __nfulnl_flush(inst);
659         /* timer_pending always called within inst->lock, so there
660          * is no chance of a race here */
661         else if (!timer_pending(&inst->timer)) {
662                 instance_get(inst);
663                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
664                 add_timer(&inst->timer);
665         }
666
667 unlock_and_release:
668         spin_unlock_bh(&inst->lock);
669         instance_put(inst);
670         return;
671
672 alloc_failure:
673         /* FIXME: statistics */
674         goto unlock_and_release;
675 }
676 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
677
678 static int
679 nfulnl_rcv_nl_event(struct notifier_block *this,
680                    unsigned long event, void *ptr)
681 {
682         struct netlink_notify *n = ptr;
683
684         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
685                 int i;
686
687                 /* destroy all instances for this pid */
688                 spin_lock_bh(&instances_lock);
689                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
690                         struct hlist_node *tmp, *t2;
691                         struct nfulnl_instance *inst;
692                         struct hlist_head *head = &instance_table[i];
693
694                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
695                                 if ((net_eq(n->net, &init_net)) &&
696                                     (n->pid == inst->peer_pid))
697                                         __instance_destroy(inst);
698                         }
699                 }
700                 spin_unlock_bh(&instances_lock);
701         }
702         return NOTIFY_DONE;
703 }
704
705 static struct notifier_block nfulnl_rtnl_notifier = {
706         .notifier_call  = nfulnl_rcv_nl_event,
707 };
708
709 static int
710 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
711                    const struct nlmsghdr *nlh,
712                    const struct nlattr * const nfqa[])
713 {
714         return -ENOTSUPP;
715 }
716
717 static struct nf_logger nfulnl_logger __read_mostly = {
718         .name   = "nfnetlink_log",
719         .logfn  = &nfulnl_log_packet,
720         .me     = THIS_MODULE,
721 };
722
723 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
724         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
725         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
726         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
727         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
728         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
729         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
730 };
731
732 static int
733 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
734                    const struct nlmsghdr *nlh,
735                    const struct nlattr * const nfula[])
736 {
737         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
738         u_int16_t group_num = ntohs(nfmsg->res_id);
739         struct nfulnl_instance *inst;
740         struct nfulnl_msg_config_cmd *cmd = NULL;
741         int ret = 0;
742
743         if (nfula[NFULA_CFG_CMD]) {
744                 u_int8_t pf = nfmsg->nfgen_family;
745                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
746
747                 /* Commands without queue context */
748                 switch (cmd->command) {
749                 case NFULNL_CFG_CMD_PF_BIND:
750                         return nf_log_bind_pf(pf, &nfulnl_logger);
751                 case NFULNL_CFG_CMD_PF_UNBIND:
752                         nf_log_unbind_pf(pf);
753                         return 0;
754                 }
755         }
756
757         inst = instance_lookup_get(group_num);
758         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
759                 ret = -EPERM;
760                 goto out_put;
761         }
762
763         if (cmd != NULL) {
764                 switch (cmd->command) {
765                 case NFULNL_CFG_CMD_BIND:
766                         if (inst) {
767                                 ret = -EBUSY;
768                                 goto out_put;
769                         }
770
771                         inst = instance_create(group_num,
772                                                NETLINK_CB(skb).pid);
773                         if (IS_ERR(inst)) {
774                                 ret = PTR_ERR(inst);
775                                 goto out;
776                         }
777                         break;
778                 case NFULNL_CFG_CMD_UNBIND:
779                         if (!inst) {
780                                 ret = -ENODEV;
781                                 goto out;
782                         }
783
784                         instance_destroy(inst);
785                         goto out_put;
786                 default:
787                         ret = -ENOTSUPP;
788                         break;
789                 }
790         }
791
792         if (nfula[NFULA_CFG_MODE]) {
793                 struct nfulnl_msg_config_mode *params;
794                 params = nla_data(nfula[NFULA_CFG_MODE]);
795
796                 if (!inst) {
797                         ret = -ENODEV;
798                         goto out;
799                 }
800                 nfulnl_set_mode(inst, params->copy_mode,
801                                 ntohl(params->copy_range));
802         }
803
804         if (nfula[NFULA_CFG_TIMEOUT]) {
805                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
806
807                 if (!inst) {
808                         ret = -ENODEV;
809                         goto out;
810                 }
811                 nfulnl_set_timeout(inst, ntohl(timeout));
812         }
813
814         if (nfula[NFULA_CFG_NLBUFSIZ]) {
815                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
816
817                 if (!inst) {
818                         ret = -ENODEV;
819                         goto out;
820                 }
821                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
822         }
823
824         if (nfula[NFULA_CFG_QTHRESH]) {
825                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
826
827                 if (!inst) {
828                         ret = -ENODEV;
829                         goto out;
830                 }
831                 nfulnl_set_qthresh(inst, ntohl(qthresh));
832         }
833
834         if (nfula[NFULA_CFG_FLAGS]) {
835                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
836
837                 if (!inst) {
838                         ret = -ENODEV;
839                         goto out;
840                 }
841                 nfulnl_set_flags(inst, ntohs(flags));
842         }
843
844 out_put:
845         instance_put(inst);
846 out:
847         return ret;
848 }
849
850 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
851         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
852                                     .attr_count = NFULA_MAX, },
853         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
854                                     .attr_count = NFULA_CFG_MAX,
855                                     .policy = nfula_cfg_policy },
856 };
857
858 static const struct nfnetlink_subsystem nfulnl_subsys = {
859         .name           = "log",
860         .subsys_id      = NFNL_SUBSYS_ULOG,
861         .cb_count       = NFULNL_MSG_MAX,
862         .cb             = nfulnl_cb,
863 };
864
865 #ifdef CONFIG_PROC_FS
866 struct iter_state {
867         unsigned int bucket;
868 };
869
870 static struct hlist_node *get_first(struct iter_state *st)
871 {
872         if (!st)
873                 return NULL;
874
875         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
876                 if (!hlist_empty(&instance_table[st->bucket]))
877                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
878         }
879         return NULL;
880 }
881
882 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
883 {
884         h = rcu_dereference_bh(hlist_next_rcu(h));
885         while (!h) {
886                 if (++st->bucket >= INSTANCE_BUCKETS)
887                         return NULL;
888
889                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
890         }
891         return h;
892 }
893
894 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
895 {
896         struct hlist_node *head;
897         head = get_first(st);
898
899         if (head)
900                 while (pos && (head = get_next(st, head)))
901                         pos--;
902         return pos ? NULL : head;
903 }
904
905 static void *seq_start(struct seq_file *seq, loff_t *pos)
906         __acquires(rcu_bh)
907 {
908         rcu_read_lock_bh();
909         return get_idx(seq->private, *pos);
910 }
911
912 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
913 {
914         (*pos)++;
915         return get_next(s->private, v);
916 }
917
918 static void seq_stop(struct seq_file *s, void *v)
919         __releases(rcu_bh)
920 {
921         rcu_read_unlock_bh();
922 }
923
924 static int seq_show(struct seq_file *s, void *v)
925 {
926         const struct nfulnl_instance *inst = v;
927
928         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
929                           inst->group_num,
930                           inst->peer_pid, inst->qlen,
931                           inst->copy_mode, inst->copy_range,
932                           inst->flushtimeout, atomic_read(&inst->use));
933 }
934
935 static const struct seq_operations nful_seq_ops = {
936         .start  = seq_start,
937         .next   = seq_next,
938         .stop   = seq_stop,
939         .show   = seq_show,
940 };
941
942 static int nful_open(struct inode *inode, struct file *file)
943 {
944         return seq_open_private(file, &nful_seq_ops,
945                         sizeof(struct iter_state));
946 }
947
948 static const struct file_operations nful_file_ops = {
949         .owner   = THIS_MODULE,
950         .open    = nful_open,
951         .read    = seq_read,
952         .llseek  = seq_lseek,
953         .release = seq_release_private,
954 };
955
956 #endif /* PROC_FS */
957
958 static int __init nfnetlink_log_init(void)
959 {
960         int i, status = -ENOMEM;
961
962         for (i = 0; i < INSTANCE_BUCKETS; i++)
963                 INIT_HLIST_HEAD(&instance_table[i]);
964
965         /* it's not really all that important to have a random value, so
966          * we can do this from the init function, even if there hasn't
967          * been that much entropy yet */
968         get_random_bytes(&hash_init, sizeof(hash_init));
969
970         netlink_register_notifier(&nfulnl_rtnl_notifier);
971         status = nfnetlink_subsys_register(&nfulnl_subsys);
972         if (status < 0) {
973                 printk(KERN_ERR "log: failed to create netlink socket\n");
974                 goto cleanup_netlink_notifier;
975         }
976
977         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
978         if (status < 0) {
979                 printk(KERN_ERR "log: failed to register logger\n");
980                 goto cleanup_subsys;
981         }
982
983 #ifdef CONFIG_PROC_FS
984         if (!proc_create("nfnetlink_log", 0440,
985                          proc_net_netfilter, &nful_file_ops))
986                 goto cleanup_logger;
987 #endif
988         return status;
989
990 #ifdef CONFIG_PROC_FS
991 cleanup_logger:
992         nf_log_unregister(&nfulnl_logger);
993 #endif
994 cleanup_subsys:
995         nfnetlink_subsys_unregister(&nfulnl_subsys);
996 cleanup_netlink_notifier:
997         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
998         return status;
999 }
1000
1001 static void __exit nfnetlink_log_fini(void)
1002 {
1003         nf_log_unregister(&nfulnl_logger);
1004 #ifdef CONFIG_PROC_FS
1005         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1006 #endif
1007         nfnetlink_subsys_unregister(&nfulnl_subsys);
1008         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1009 }
1010
1011 MODULE_DESCRIPTION("netfilter userspace logging");
1012 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1013 MODULE_LICENSE("GPL");
1014 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1015
1016 module_init(nfnetlink_log_init);
1017 module_exit(nfnetlink_log_fini);