atomic: use <linux/atomic.h>
[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 <linux/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 char *prefix, unsigned int plen)
380 {
381         struct nfulnl_msg_packet_hdr pmsg;
382         struct nlmsghdr *nlh;
383         struct nfgenmsg *nfmsg;
384         sk_buff_data_t old_tail = inst->skb->tail;
385
386         nlh = NLMSG_PUT(inst->skb, 0, 0,
387                         NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET,
388                         sizeof(struct nfgenmsg));
389         nfmsg = NLMSG_DATA(nlh);
390         nfmsg->nfgen_family = pf;
391         nfmsg->version = NFNETLINK_V0;
392         nfmsg->res_id = htons(inst->group_num);
393
394         pmsg.hw_protocol        = skb->protocol;
395         pmsg.hook               = hooknum;
396
397         NLA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg);
398
399         if (prefix)
400                 NLA_PUT(inst->skb, NFULA_PREFIX, plen, prefix);
401
402         if (indev) {
403 #ifndef CONFIG_BRIDGE_NETFILTER
404                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
405                              htonl(indev->ifindex));
406 #else
407                 if (pf == PF_BRIDGE) {
408                         /* Case 1: outdev is physical input device, we need to
409                          * look for bridge group (when called from
410                          * netfilter_bridge) */
411                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
412                                      htonl(indev->ifindex));
413                         /* this is the bridge group "brX" */
414                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
415                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
416                                      htonl(br_port_get_rcu(indev)->br->dev->ifindex));
417                 } else {
418                         /* Case 2: indev is bridge group, we need to look for
419                          * physical device (when called from ipv4) */
420                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_INDEV,
421                                      htonl(indev->ifindex));
422                         if (skb->nf_bridge && skb->nf_bridge->physindev)
423                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSINDEV,
424                                              htonl(skb->nf_bridge->physindev->ifindex));
425                 }
426 #endif
427         }
428
429         if (outdev) {
430 #ifndef CONFIG_BRIDGE_NETFILTER
431                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
432                              htonl(outdev->ifindex));
433 #else
434                 if (pf == PF_BRIDGE) {
435                         /* Case 1: outdev is physical output device, we need to
436                          * look for bridge group (when called from
437                          * netfilter_bridge) */
438                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
439                                      htonl(outdev->ifindex));
440                         /* this is the bridge group "brX" */
441                         /* rcu_read_lock()ed by nf_hook_slow or nf_log_packet */
442                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
443                                      htonl(br_port_get_rcu(outdev)->br->dev->ifindex));
444                 } else {
445                         /* Case 2: indev is a bridge group, we need to look
446                          * for physical device (when called from ipv4) */
447                         NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_OUTDEV,
448                                      htonl(outdev->ifindex));
449                         if (skb->nf_bridge && skb->nf_bridge->physoutdev)
450                                 NLA_PUT_BE32(inst->skb, NFULA_IFINDEX_PHYSOUTDEV,
451                                              htonl(skb->nf_bridge->physoutdev->ifindex));
452                 }
453 #endif
454         }
455
456         if (skb->mark)
457                 NLA_PUT_BE32(inst->skb, NFULA_MARK, htonl(skb->mark));
458
459         if (indev && skb->dev &&
460             skb->mac_header != skb->network_header) {
461                 struct nfulnl_msg_packet_hw phw;
462                 int len = dev_parse_header(skb, phw.hw_addr);
463                 if (len > 0) {
464                         phw.hw_addrlen = htons(len);
465                         NLA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw);
466                 }
467         }
468
469         if (indev && skb_mac_header_was_set(skb)) {
470                 NLA_PUT_BE16(inst->skb, NFULA_HWTYPE, htons(skb->dev->type));
471                 NLA_PUT_BE16(inst->skb, NFULA_HWLEN,
472                              htons(skb->dev->hard_header_len));
473                 NLA_PUT(inst->skb, NFULA_HWHEADER, skb->dev->hard_header_len,
474                         skb_mac_header(skb));
475         }
476
477         if (skb->tstamp.tv64) {
478                 struct nfulnl_msg_packet_timestamp ts;
479                 struct timeval tv = ktime_to_timeval(skb->tstamp);
480                 ts.sec = cpu_to_be64(tv.tv_sec);
481                 ts.usec = cpu_to_be64(tv.tv_usec);
482
483                 NLA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts);
484         }
485
486         /* UID */
487         if (skb->sk) {
488                 read_lock_bh(&skb->sk->sk_callback_lock);
489                 if (skb->sk->sk_socket && skb->sk->sk_socket->file) {
490                         struct file *file = skb->sk->sk_socket->file;
491                         __be32 uid = htonl(file->f_cred->fsuid);
492                         __be32 gid = htonl(file->f_cred->fsgid);
493                         /* need to unlock here since NLA_PUT may goto */
494                         read_unlock_bh(&skb->sk->sk_callback_lock);
495                         NLA_PUT_BE32(inst->skb, NFULA_UID, uid);
496                         NLA_PUT_BE32(inst->skb, NFULA_GID, gid);
497                 } else
498                         read_unlock_bh(&skb->sk->sk_callback_lock);
499         }
500
501         /* local sequence number */
502         if (inst->flags & NFULNL_CFG_F_SEQ)
503                 NLA_PUT_BE32(inst->skb, NFULA_SEQ, htonl(inst->seq++));
504
505         /* global sequence number */
506         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
507                 NLA_PUT_BE32(inst->skb, NFULA_SEQ_GLOBAL,
508                              htonl(atomic_inc_return(&global_seq)));
509
510         if (data_len) {
511                 struct nlattr *nla;
512                 int size = nla_attr_size(data_len);
513
514                 if (skb_tailroom(inst->skb) < nla_total_size(data_len)) {
515                         printk(KERN_WARNING "nfnetlink_log: no tailroom!\n");
516                         goto nlmsg_failure;
517                 }
518
519                 nla = (struct nlattr *)skb_put(inst->skb, nla_total_size(data_len));
520                 nla->nla_type = NFULA_PAYLOAD;
521                 nla->nla_len = size;
522
523                 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
524                         BUG();
525         }
526
527         nlh->nlmsg_len = inst->skb->tail - old_tail;
528         return 0;
529
530 nlmsg_failure:
531 nla_put_failure:
532         PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n");
533         return -1;
534 }
535
536 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
537
538 static struct nf_loginfo default_loginfo = {
539         .type =         NF_LOG_TYPE_ULOG,
540         .u = {
541                 .ulog = {
542                         .copy_len       = 0xffff,
543                         .group          = 0,
544                         .qthreshold     = 1,
545                 },
546         },
547 };
548
549 /* log handler for internal netfilter logging api */
550 void
551 nfulnl_log_packet(u_int8_t pf,
552                   unsigned int hooknum,
553                   const struct sk_buff *skb,
554                   const struct net_device *in,
555                   const struct net_device *out,
556                   const struct nf_loginfo *li_user,
557                   const char *prefix)
558 {
559         unsigned int size, data_len;
560         struct nfulnl_instance *inst;
561         const struct nf_loginfo *li;
562         unsigned int qthreshold;
563         unsigned int plen;
564
565         if (li_user && li_user->type == NF_LOG_TYPE_ULOG)
566                 li = li_user;
567         else
568                 li = &default_loginfo;
569
570         inst = instance_lookup_get(li->u.ulog.group);
571         if (!inst)
572                 return;
573
574         plen = 0;
575         if (prefix)
576                 plen = strlen(prefix) + 1;
577
578         /* FIXME: do we want to make the size calculation conditional based on
579          * what is actually present?  way more branches and checks, but more
580          * memory efficient... */
581         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
582                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hdr))
583                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
584                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
585 #ifdef CONFIG_BRIDGE_NETFILTER
586                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
587                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
588 #endif
589                 + nla_total_size(sizeof(u_int32_t))     /* mark */
590                 + nla_total_size(sizeof(u_int32_t))     /* uid */
591                 + nla_total_size(sizeof(u_int32_t))     /* gid */
592                 + nla_total_size(plen)                  /* prefix */
593                 + nla_total_size(sizeof(struct nfulnl_msg_packet_hw))
594                 + nla_total_size(sizeof(struct nfulnl_msg_packet_timestamp));
595
596         if (in && skb_mac_header_was_set(skb)) {
597                 size +=   nla_total_size(skb->dev->hard_header_len)
598                         + nla_total_size(sizeof(u_int16_t))     /* hwtype */
599                         + nla_total_size(sizeof(u_int16_t));    /* hwlen */
600         }
601
602         spin_lock_bh(&inst->lock);
603
604         if (inst->flags & NFULNL_CFG_F_SEQ)
605                 size += nla_total_size(sizeof(u_int32_t));
606         if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
607                 size += nla_total_size(sizeof(u_int32_t));
608
609         qthreshold = inst->qthreshold;
610         /* per-rule qthreshold overrides per-instance */
611         if (li->u.ulog.qthreshold)
612                 if (qthreshold > li->u.ulog.qthreshold)
613                         qthreshold = li->u.ulog.qthreshold;
614
615
616         switch (inst->copy_mode) {
617         case NFULNL_COPY_META:
618         case NFULNL_COPY_NONE:
619                 data_len = 0;
620                 break;
621
622         case NFULNL_COPY_PACKET:
623                 if (inst->copy_range == 0
624                     || inst->copy_range > skb->len)
625                         data_len = skb->len;
626                 else
627                         data_len = inst->copy_range;
628
629                 size += nla_total_size(data_len);
630                 break;
631
632         case NFULNL_COPY_DISABLED:
633         default:
634                 goto unlock_and_release;
635         }
636
637         if (inst->skb &&
638             size > skb_tailroom(inst->skb) - sizeof(struct nfgenmsg)) {
639                 /* either the queue len is too high or we don't have
640                  * enough room in the skb left. flush to userspace. */
641                 __nfulnl_flush(inst);
642         }
643
644         if (!inst->skb) {
645                 inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
646                 if (!inst->skb)
647                         goto alloc_failure;
648         }
649
650         inst->qlen++;
651
652         __build_packet_message(inst, skb, data_len, pf,
653                                 hooknum, in, out, prefix, plen);
654
655         if (inst->qlen >= qthreshold)
656                 __nfulnl_flush(inst);
657         /* timer_pending always called within inst->lock, so there
658          * is no chance of a race here */
659         else if (!timer_pending(&inst->timer)) {
660                 instance_get(inst);
661                 inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100);
662                 add_timer(&inst->timer);
663         }
664
665 unlock_and_release:
666         spin_unlock_bh(&inst->lock);
667         instance_put(inst);
668         return;
669
670 alloc_failure:
671         /* FIXME: statistics */
672         goto unlock_and_release;
673 }
674 EXPORT_SYMBOL_GPL(nfulnl_log_packet);
675
676 static int
677 nfulnl_rcv_nl_event(struct notifier_block *this,
678                    unsigned long event, void *ptr)
679 {
680         struct netlink_notify *n = ptr;
681
682         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
683                 int i;
684
685                 /* destroy all instances for this pid */
686                 spin_lock_bh(&instances_lock);
687                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
688                         struct hlist_node *tmp, *t2;
689                         struct nfulnl_instance *inst;
690                         struct hlist_head *head = &instance_table[i];
691
692                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
693                                 if ((net_eq(n->net, &init_net)) &&
694                                     (n->pid == inst->peer_pid))
695                                         __instance_destroy(inst);
696                         }
697                 }
698                 spin_unlock_bh(&instances_lock);
699         }
700         return NOTIFY_DONE;
701 }
702
703 static struct notifier_block nfulnl_rtnl_notifier = {
704         .notifier_call  = nfulnl_rcv_nl_event,
705 };
706
707 static int
708 nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
709                    const struct nlmsghdr *nlh,
710                    const struct nlattr * const nfqa[])
711 {
712         return -ENOTSUPP;
713 }
714
715 static struct nf_logger nfulnl_logger __read_mostly = {
716         .name   = "nfnetlink_log",
717         .logfn  = &nfulnl_log_packet,
718         .me     = THIS_MODULE,
719 };
720
721 static const struct nla_policy nfula_cfg_policy[NFULA_CFG_MAX+1] = {
722         [NFULA_CFG_CMD]         = { .len = sizeof(struct nfulnl_msg_config_cmd) },
723         [NFULA_CFG_MODE]        = { .len = sizeof(struct nfulnl_msg_config_mode) },
724         [NFULA_CFG_TIMEOUT]     = { .type = NLA_U32 },
725         [NFULA_CFG_QTHRESH]     = { .type = NLA_U32 },
726         [NFULA_CFG_NLBUFSIZ]    = { .type = NLA_U32 },
727         [NFULA_CFG_FLAGS]       = { .type = NLA_U16 },
728 };
729
730 static int
731 nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
732                    const struct nlmsghdr *nlh,
733                    const struct nlattr * const nfula[])
734 {
735         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
736         u_int16_t group_num = ntohs(nfmsg->res_id);
737         struct nfulnl_instance *inst;
738         struct nfulnl_msg_config_cmd *cmd = NULL;
739         int ret = 0;
740
741         if (nfula[NFULA_CFG_CMD]) {
742                 u_int8_t pf = nfmsg->nfgen_family;
743                 cmd = nla_data(nfula[NFULA_CFG_CMD]);
744
745                 /* Commands without queue context */
746                 switch (cmd->command) {
747                 case NFULNL_CFG_CMD_PF_BIND:
748                         return nf_log_bind_pf(pf, &nfulnl_logger);
749                 case NFULNL_CFG_CMD_PF_UNBIND:
750                         nf_log_unbind_pf(pf);
751                         return 0;
752                 }
753         }
754
755         inst = instance_lookup_get(group_num);
756         if (inst && inst->peer_pid != NETLINK_CB(skb).pid) {
757                 ret = -EPERM;
758                 goto out_put;
759         }
760
761         if (cmd != NULL) {
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 (IS_ERR(inst)) {
772                                 ret = PTR_ERR(inst);
773                                 goto out;
774                         }
775                         break;
776                 case NFULNL_CFG_CMD_UNBIND:
777                         if (!inst) {
778                                 ret = -ENODEV;
779                                 goto out;
780                         }
781
782                         instance_destroy(inst);
783                         goto out_put;
784                 default:
785                         ret = -ENOTSUPP;
786                         break;
787                 }
788         }
789
790         if (nfula[NFULA_CFG_MODE]) {
791                 struct nfulnl_msg_config_mode *params;
792                 params = nla_data(nfula[NFULA_CFG_MODE]);
793
794                 if (!inst) {
795                         ret = -ENODEV;
796                         goto out;
797                 }
798                 nfulnl_set_mode(inst, params->copy_mode,
799                                 ntohl(params->copy_range));
800         }
801
802         if (nfula[NFULA_CFG_TIMEOUT]) {
803                 __be32 timeout = nla_get_be32(nfula[NFULA_CFG_TIMEOUT]);
804
805                 if (!inst) {
806                         ret = -ENODEV;
807                         goto out;
808                 }
809                 nfulnl_set_timeout(inst, ntohl(timeout));
810         }
811
812         if (nfula[NFULA_CFG_NLBUFSIZ]) {
813                 __be32 nlbufsiz = nla_get_be32(nfula[NFULA_CFG_NLBUFSIZ]);
814
815                 if (!inst) {
816                         ret = -ENODEV;
817                         goto out;
818                 }
819                 nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz));
820         }
821
822         if (nfula[NFULA_CFG_QTHRESH]) {
823                 __be32 qthresh = nla_get_be32(nfula[NFULA_CFG_QTHRESH]);
824
825                 if (!inst) {
826                         ret = -ENODEV;
827                         goto out;
828                 }
829                 nfulnl_set_qthresh(inst, ntohl(qthresh));
830         }
831
832         if (nfula[NFULA_CFG_FLAGS]) {
833                 __be16 flags = nla_get_be16(nfula[NFULA_CFG_FLAGS]);
834
835                 if (!inst) {
836                         ret = -ENODEV;
837                         goto out;
838                 }
839                 nfulnl_set_flags(inst, ntohs(flags));
840         }
841
842 out_put:
843         instance_put(inst);
844 out:
845         return ret;
846 }
847
848 static const struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = {
849         [NFULNL_MSG_PACKET]     = { .call = nfulnl_recv_unsupp,
850                                     .attr_count = NFULA_MAX, },
851         [NFULNL_MSG_CONFIG]     = { .call = nfulnl_recv_config,
852                                     .attr_count = NFULA_CFG_MAX,
853                                     .policy = nfula_cfg_policy },
854 };
855
856 static const struct nfnetlink_subsystem nfulnl_subsys = {
857         .name           = "log",
858         .subsys_id      = NFNL_SUBSYS_ULOG,
859         .cb_count       = NFULNL_MSG_MAX,
860         .cb             = nfulnl_cb,
861 };
862
863 #ifdef CONFIG_PROC_FS
864 struct iter_state {
865         unsigned int bucket;
866 };
867
868 static struct hlist_node *get_first(struct iter_state *st)
869 {
870         if (!st)
871                 return NULL;
872
873         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
874                 if (!hlist_empty(&instance_table[st->bucket]))
875                         return rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
876         }
877         return NULL;
878 }
879
880 static struct hlist_node *get_next(struct iter_state *st, struct hlist_node *h)
881 {
882         h = rcu_dereference_bh(hlist_next_rcu(h));
883         while (!h) {
884                 if (++st->bucket >= INSTANCE_BUCKETS)
885                         return NULL;
886
887                 h = rcu_dereference_bh(hlist_first_rcu(&instance_table[st->bucket]));
888         }
889         return h;
890 }
891
892 static struct hlist_node *get_idx(struct iter_state *st, loff_t pos)
893 {
894         struct hlist_node *head;
895         head = get_first(st);
896
897         if (head)
898                 while (pos && (head = get_next(st, head)))
899                         pos--;
900         return pos ? NULL : head;
901 }
902
903 static void *seq_start(struct seq_file *seq, loff_t *pos)
904         __acquires(rcu_bh)
905 {
906         rcu_read_lock_bh();
907         return get_idx(seq->private, *pos);
908 }
909
910 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
911 {
912         (*pos)++;
913         return get_next(s->private, v);
914 }
915
916 static void seq_stop(struct seq_file *s, void *v)
917         __releases(rcu_bh)
918 {
919         rcu_read_unlock_bh();
920 }
921
922 static int seq_show(struct seq_file *s, void *v)
923 {
924         const struct nfulnl_instance *inst = v;
925
926         return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n",
927                           inst->group_num,
928                           inst->peer_pid, inst->qlen,
929                           inst->copy_mode, inst->copy_range,
930                           inst->flushtimeout, atomic_read(&inst->use));
931 }
932
933 static const struct seq_operations nful_seq_ops = {
934         .start  = seq_start,
935         .next   = seq_next,
936         .stop   = seq_stop,
937         .show   = seq_show,
938 };
939
940 static int nful_open(struct inode *inode, struct file *file)
941 {
942         return seq_open_private(file, &nful_seq_ops,
943                         sizeof(struct iter_state));
944 }
945
946 static const struct file_operations nful_file_ops = {
947         .owner   = THIS_MODULE,
948         .open    = nful_open,
949         .read    = seq_read,
950         .llseek  = seq_lseek,
951         .release = seq_release_private,
952 };
953
954 #endif /* PROC_FS */
955
956 static int __init nfnetlink_log_init(void)
957 {
958         int i, status = -ENOMEM;
959
960         for (i = 0; i < INSTANCE_BUCKETS; i++)
961                 INIT_HLIST_HEAD(&instance_table[i]);
962
963         /* it's not really all that important to have a random value, so
964          * we can do this from the init function, even if there hasn't
965          * been that much entropy yet */
966         get_random_bytes(&hash_init, sizeof(hash_init));
967
968         netlink_register_notifier(&nfulnl_rtnl_notifier);
969         status = nfnetlink_subsys_register(&nfulnl_subsys);
970         if (status < 0) {
971                 printk(KERN_ERR "log: failed to create netlink socket\n");
972                 goto cleanup_netlink_notifier;
973         }
974
975         status = nf_log_register(NFPROTO_UNSPEC, &nfulnl_logger);
976         if (status < 0) {
977                 printk(KERN_ERR "log: failed to register logger\n");
978                 goto cleanup_subsys;
979         }
980
981 #ifdef CONFIG_PROC_FS
982         if (!proc_create("nfnetlink_log", 0440,
983                          proc_net_netfilter, &nful_file_ops))
984                 goto cleanup_logger;
985 #endif
986         return status;
987
988 #ifdef CONFIG_PROC_FS
989 cleanup_logger:
990         nf_log_unregister(&nfulnl_logger);
991 #endif
992 cleanup_subsys:
993         nfnetlink_subsys_unregister(&nfulnl_subsys);
994 cleanup_netlink_notifier:
995         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
996         return status;
997 }
998
999 static void __exit nfnetlink_log_fini(void)
1000 {
1001         nf_log_unregister(&nfulnl_logger);
1002 #ifdef CONFIG_PROC_FS
1003         remove_proc_entry("nfnetlink_log", proc_net_netfilter);
1004 #endif
1005         nfnetlink_subsys_unregister(&nfulnl_subsys);
1006         netlink_unregister_notifier(&nfulnl_rtnl_notifier);
1007 }
1008
1009 MODULE_DESCRIPTION("netfilter userspace logging");
1010 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1011 MODULE_LICENSE("GPL");
1012 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG);
1013
1014 module_init(nfnetlink_log_init);
1015 module_exit(nfnetlink_log_fini);