Harald Welte | 0597f26 | 2005-08-09 19:58:39 -0700 | [diff] [blame] | 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 | */ |
| 15 | #include <linux/module.h> |
| 16 | #include <linux/skbuff.h> |
| 17 | #include <linux/init.h> |
| 18 | #include <linux/ip.h> |
| 19 | #include <linux/ipv6.h> |
| 20 | #include <linux/netdevice.h> |
| 21 | #include <linux/netfilter.h> |
| 22 | #include <linux/netlink.h> |
| 23 | #include <linux/netfilter/nfnetlink.h> |
| 24 | #include <linux/netfilter/nfnetlink_log.h> |
| 25 | #include <linux/spinlock.h> |
| 26 | #include <linux/sysctl.h> |
| 27 | #include <linux/proc_fs.h> |
| 28 | #include <linux/security.h> |
| 29 | #include <linux/list.h> |
| 30 | #include <linux/jhash.h> |
| 31 | #include <linux/random.h> |
| 32 | #include <net/sock.h> |
| 33 | |
| 34 | #include <asm/atomic.h> |
| 35 | |
| 36 | #define NFULNL_NLBUFSIZ_DEFAULT 4096 |
| 37 | #define NFULNL_TIMEOUT_DEFAULT 100 /* every second */ |
| 38 | #define NFULNL_QTHRESH_DEFAULT 100 /* 100 packets */ |
| 39 | |
| 40 | #define PRINTR(x, args...) do { if (net_ratelimit()) \ |
| 41 | printk(x, ## args); } while (0); |
| 42 | |
| 43 | #if 0 |
| 44 | #define UDEBUG(x, args ...) printk(KERN_DEBUG "%s(%d):%s(): " x, \ |
| 45 | __FILE__, __LINE__, __FUNCTION__, \ |
| 46 | ## args) |
| 47 | #else |
| 48 | #define UDEBUG(x, ...) |
| 49 | #endif |
| 50 | |
| 51 | struct nfulnl_instance { |
| 52 | struct hlist_node hlist; /* global list of instances */ |
| 53 | spinlock_t lock; |
| 54 | atomic_t use; /* use count */ |
| 55 | |
| 56 | unsigned int qlen; /* number of nlmsgs in skb */ |
| 57 | struct sk_buff *skb; /* pre-allocatd skb */ |
| 58 | struct nlmsghdr *lastnlh; /* netlink header of last msg in skb */ |
| 59 | struct timer_list timer; |
| 60 | int peer_pid; /* PID of the peer process */ |
| 61 | |
| 62 | /* configurable parameters */ |
| 63 | unsigned int flushtimeout; /* timeout until queue flush */ |
| 64 | unsigned int nlbufsiz; /* netlink buffer allocation size */ |
| 65 | unsigned int qthreshold; /* threshold of the queue */ |
| 66 | u_int32_t copy_range; |
| 67 | u_int16_t group_num; /* number of this queue */ |
| 68 | u_int8_t copy_mode; |
| 69 | }; |
| 70 | |
| 71 | static DEFINE_RWLOCK(instances_lock); |
| 72 | |
| 73 | #define INSTANCE_BUCKETS 16 |
| 74 | static struct hlist_head instance_table[INSTANCE_BUCKETS]; |
| 75 | static unsigned int hash_init; |
| 76 | |
| 77 | static inline u_int8_t instance_hashfn(u_int16_t group_num) |
| 78 | { |
| 79 | return ((group_num & 0xff) % INSTANCE_BUCKETS); |
| 80 | } |
| 81 | |
| 82 | static struct nfulnl_instance * |
| 83 | __instance_lookup(u_int16_t group_num) |
| 84 | { |
| 85 | struct hlist_head *head; |
| 86 | struct hlist_node *pos; |
| 87 | struct nfulnl_instance *inst; |
| 88 | |
| 89 | UDEBUG("entering (group_num=%u)\n", group_num); |
| 90 | |
| 91 | head = &instance_table[instance_hashfn(group_num)]; |
| 92 | hlist_for_each_entry(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 | read_lock_bh(&instances_lock); |
| 111 | inst = __instance_lookup(group_num); |
| 112 | if (inst) |
| 113 | instance_get(inst); |
| 114 | read_unlock_bh(&instances_lock); |
| 115 | |
| 116 | return inst; |
| 117 | } |
| 118 | |
| 119 | static void |
| 120 | instance_put(struct nfulnl_instance *inst) |
| 121 | { |
| 122 | if (inst && atomic_dec_and_test(&inst->use)) { |
| 123 | UDEBUG("kfree(inst=%p)\n", inst); |
| 124 | kfree(inst); |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | static void nfulnl_timer(unsigned long data); |
| 129 | |
| 130 | static struct nfulnl_instance * |
| 131 | instance_create(u_int16_t group_num, int pid) |
| 132 | { |
| 133 | struct nfulnl_instance *inst; |
| 134 | |
| 135 | UDEBUG("entering (group_num=%u, pid=%d)\n", group_num, |
| 136 | pid); |
| 137 | |
| 138 | write_lock_bh(&instances_lock); |
| 139 | if (__instance_lookup(group_num)) { |
| 140 | inst = NULL; |
| 141 | UDEBUG("aborting, instance already exists\n"); |
| 142 | goto out_unlock; |
| 143 | } |
| 144 | |
| 145 | inst = kmalloc(sizeof(*inst), GFP_ATOMIC); |
| 146 | if (!inst) |
| 147 | goto out_unlock; |
| 148 | |
| 149 | memset(inst, 0, sizeof(*inst)); |
| 150 | INIT_HLIST_NODE(&inst->hlist); |
| 151 | inst->lock = SPIN_LOCK_UNLOCKED; |
| 152 | /* needs to be two, since we _put() after creation */ |
| 153 | atomic_set(&inst->use, 2); |
| 154 | |
| 155 | init_timer(&inst->timer); |
| 156 | inst->timer.function = nfulnl_timer; |
| 157 | inst->timer.data = (unsigned long)inst; |
| 158 | /* don't start timer yet. (re)start it with every packet */ |
| 159 | |
| 160 | inst->peer_pid = pid; |
| 161 | inst->group_num = group_num; |
| 162 | |
| 163 | inst->qthreshold = NFULNL_QTHRESH_DEFAULT; |
| 164 | inst->flushtimeout = NFULNL_TIMEOUT_DEFAULT; |
| 165 | inst->nlbufsiz = NFULNL_NLBUFSIZ_DEFAULT; |
| 166 | inst->copy_mode = NFULNL_COPY_PACKET; |
| 167 | inst->copy_range = 0xffff; |
| 168 | |
| 169 | if (!try_module_get(THIS_MODULE)) |
| 170 | goto out_free; |
| 171 | |
| 172 | hlist_add_head(&inst->hlist, |
| 173 | &instance_table[instance_hashfn(group_num)]); |
| 174 | |
| 175 | UDEBUG("newly added node: %p, next=%p\n", &inst->hlist, |
| 176 | inst->hlist.next); |
| 177 | |
| 178 | write_unlock_bh(&instances_lock); |
| 179 | |
| 180 | return inst; |
| 181 | |
| 182 | out_free: |
| 183 | instance_put(inst); |
| 184 | out_unlock: |
| 185 | write_unlock_bh(&instances_lock); |
| 186 | return NULL; |
| 187 | } |
| 188 | |
| 189 | static int __nfulnl_send(struct nfulnl_instance *inst); |
| 190 | |
| 191 | static void |
| 192 | _instance_destroy2(struct nfulnl_instance *inst, int lock) |
| 193 | { |
| 194 | /* first pull it out of the global list */ |
| 195 | if (lock) |
| 196 | write_lock_bh(&instances_lock); |
| 197 | |
| 198 | UDEBUG("removing instance %p (queuenum=%u) from hash\n", |
| 199 | inst, inst->group_num); |
| 200 | |
| 201 | hlist_del(&inst->hlist); |
| 202 | |
| 203 | if (lock) |
| 204 | write_unlock_bh(&instances_lock); |
| 205 | |
| 206 | /* then flush all pending packets from skb */ |
| 207 | |
| 208 | spin_lock_bh(&inst->lock); |
| 209 | if (inst->skb) { |
| 210 | if (inst->qlen) |
| 211 | __nfulnl_send(inst); |
| 212 | if (inst->skb) { |
| 213 | kfree_skb(inst->skb); |
| 214 | inst->skb = NULL; |
| 215 | } |
| 216 | } |
| 217 | spin_unlock_bh(&inst->lock); |
| 218 | |
| 219 | /* and finally put the refcount */ |
| 220 | instance_put(inst); |
| 221 | |
| 222 | module_put(THIS_MODULE); |
| 223 | } |
| 224 | |
| 225 | static inline void |
| 226 | __instance_destroy(struct nfulnl_instance *inst) |
| 227 | { |
| 228 | _instance_destroy2(inst, 0); |
| 229 | } |
| 230 | |
| 231 | static inline void |
| 232 | instance_destroy(struct nfulnl_instance *inst) |
| 233 | { |
| 234 | _instance_destroy2(inst, 1); |
| 235 | } |
| 236 | |
| 237 | static int |
| 238 | nfulnl_set_mode(struct nfulnl_instance *inst, u_int8_t mode, |
| 239 | unsigned int range) |
| 240 | { |
| 241 | int status = 0; |
| 242 | |
| 243 | spin_lock_bh(&inst->lock); |
| 244 | |
| 245 | switch (mode) { |
| 246 | case NFULNL_COPY_NONE: |
| 247 | case NFULNL_COPY_META: |
| 248 | inst->copy_mode = mode; |
| 249 | inst->copy_range = 0; |
| 250 | break; |
| 251 | |
| 252 | case NFULNL_COPY_PACKET: |
| 253 | inst->copy_mode = mode; |
| 254 | /* we're using struct nfattr which has 16bit nfa_len */ |
| 255 | if (range > 0xffff) |
| 256 | inst->copy_range = 0xffff; |
| 257 | else |
| 258 | inst->copy_range = range; |
| 259 | break; |
| 260 | |
| 261 | default: |
| 262 | status = -EINVAL; |
| 263 | break; |
| 264 | } |
| 265 | |
| 266 | spin_unlock_bh(&inst->lock); |
| 267 | |
| 268 | return status; |
| 269 | } |
| 270 | |
| 271 | static int |
| 272 | nfulnl_set_nlbufsiz(struct nfulnl_instance *inst, u_int32_t nlbufsiz) |
| 273 | { |
| 274 | int status; |
| 275 | |
| 276 | spin_lock_bh(&inst->lock); |
| 277 | if (nlbufsiz < NFULNL_NLBUFSIZ_DEFAULT) |
| 278 | status = -ERANGE; |
| 279 | else if (nlbufsiz > 131072) |
| 280 | status = -ERANGE; |
| 281 | else { |
| 282 | inst->nlbufsiz = nlbufsiz; |
| 283 | status = 0; |
| 284 | } |
| 285 | spin_unlock_bh(&inst->lock); |
| 286 | |
| 287 | return status; |
| 288 | } |
| 289 | |
| 290 | static int |
| 291 | nfulnl_set_timeout(struct nfulnl_instance *inst, u_int32_t timeout) |
| 292 | { |
| 293 | spin_lock_bh(&inst->lock); |
| 294 | inst->flushtimeout = timeout; |
| 295 | spin_unlock_bh(&inst->lock); |
| 296 | |
| 297 | return 0; |
| 298 | } |
| 299 | |
| 300 | static int |
| 301 | nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh) |
| 302 | { |
| 303 | spin_lock_bh(&inst->lock); |
| 304 | inst->qthreshold = qthresh; |
| 305 | spin_unlock_bh(&inst->lock); |
| 306 | |
| 307 | return 0; |
| 308 | } |
| 309 | |
| 310 | static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size, |
| 311 | unsigned int pkt_size) |
| 312 | { |
| 313 | struct sk_buff *skb; |
| 314 | |
| 315 | UDEBUG("entered (%u, %u)\n", inst_size, pkt_size); |
| 316 | |
| 317 | /* alloc skb which should be big enough for a whole multipart |
| 318 | * message. WARNING: has to be <= 128k due to slab restrictions */ |
| 319 | |
| 320 | skb = alloc_skb(inst_size, GFP_ATOMIC); |
| 321 | if (!skb) { |
| 322 | PRINTR("nfnetlink_log: can't alloc whole buffer (%u bytes)\n", |
| 323 | inst_size); |
| 324 | |
| 325 | /* try to allocate only as much as we need for current |
| 326 | * packet */ |
| 327 | |
| 328 | skb = alloc_skb(pkt_size, GFP_ATOMIC); |
| 329 | if (!skb) |
| 330 | PRINTR("nfnetlink_log: can't even alloc %u bytes\n", |
| 331 | pkt_size); |
| 332 | } |
| 333 | |
| 334 | return skb; |
| 335 | } |
| 336 | |
| 337 | static int |
| 338 | __nfulnl_send(struct nfulnl_instance *inst) |
| 339 | { |
| 340 | int status; |
| 341 | |
| 342 | if (timer_pending(&inst->timer)) |
| 343 | del_timer(&inst->timer); |
| 344 | |
| 345 | if (inst->qlen > 1) |
| 346 | inst->lastnlh->nlmsg_type = NLMSG_DONE; |
| 347 | |
| 348 | status = nfnetlink_unicast(inst->skb, inst->peer_pid, MSG_DONTWAIT); |
| 349 | if (status < 0) { |
| 350 | UDEBUG("netlink_unicast() failed\n"); |
| 351 | /* FIXME: statistics */ |
| 352 | } |
| 353 | |
| 354 | inst->qlen = 0; |
| 355 | inst->skb = NULL; |
| 356 | inst->lastnlh = NULL; |
| 357 | |
| 358 | return status; |
| 359 | } |
| 360 | |
| 361 | static void nfulnl_timer(unsigned long data) |
| 362 | { |
| 363 | struct nfulnl_instance *inst = (struct nfulnl_instance *)data; |
| 364 | |
| 365 | UDEBUG("timer function called, flushing buffer\n"); |
| 366 | |
| 367 | spin_lock_bh(&inst->lock); |
| 368 | __nfulnl_send(inst); |
| 369 | instance_put(inst); |
| 370 | spin_unlock_bh(&inst->lock); |
| 371 | } |
| 372 | |
| 373 | static inline int |
| 374 | __build_packet_message(struct nfulnl_instance *inst, |
| 375 | const struct sk_buff *skb, |
| 376 | unsigned int data_len, |
| 377 | unsigned int pf, |
| 378 | unsigned int hooknum, |
| 379 | const struct net_device *indev, |
| 380 | const struct net_device *outdev, |
| 381 | const struct nf_loginfo *li, |
| 382 | const char *prefix) |
| 383 | { |
| 384 | unsigned char *old_tail; |
| 385 | struct nfulnl_msg_packet_hdr pmsg; |
| 386 | struct nlmsghdr *nlh; |
| 387 | struct nfgenmsg *nfmsg; |
| 388 | u_int32_t tmp_uint; |
| 389 | |
| 390 | UDEBUG("entered\n"); |
| 391 | |
| 392 | old_tail = inst->skb->tail; |
| 393 | nlh = NLMSG_PUT(inst->skb, 0, 0, |
| 394 | NFNL_SUBSYS_ULOG << 8 | NFULNL_MSG_PACKET, |
| 395 | sizeof(struct nfgenmsg)); |
| 396 | nfmsg = NLMSG_DATA(nlh); |
| 397 | nfmsg->nfgen_family = pf; |
| 398 | nfmsg->version = NFNETLINK_V0; |
| 399 | nfmsg->res_id = htons(inst->group_num); |
| 400 | |
| 401 | pmsg.hw_protocol = htons(skb->protocol); |
| 402 | pmsg.hook = hooknum; |
| 403 | |
| 404 | NFA_PUT(inst->skb, NFULA_PACKET_HDR, sizeof(pmsg), &pmsg); |
| 405 | |
| 406 | if (prefix) { |
| 407 | int slen = strlen(prefix); |
| 408 | if (slen > NFULNL_PREFIXLEN) |
| 409 | slen = NFULNL_PREFIXLEN; |
| 410 | NFA_PUT(inst->skb, NFULA_PREFIX, slen, prefix); |
| 411 | } |
| 412 | |
| 413 | if (indev) { |
| 414 | tmp_uint = htonl(indev->ifindex); |
| 415 | NFA_PUT(inst->skb, NFULA_IFINDEX_INDEV, sizeof(tmp_uint), |
| 416 | &tmp_uint); |
| 417 | } |
| 418 | |
| 419 | if (outdev) { |
| 420 | tmp_uint = htonl(outdev->ifindex); |
| 421 | NFA_PUT(inst->skb, NFULA_IFINDEX_OUTDEV, sizeof(tmp_uint), |
| 422 | &tmp_uint); |
| 423 | } |
| 424 | |
| 425 | if (skb->nfmark) { |
| 426 | tmp_uint = htonl(skb->nfmark); |
| 427 | NFA_PUT(inst->skb, NFULA_MARK, sizeof(tmp_uint), &tmp_uint); |
| 428 | } |
| 429 | |
| 430 | if (indev && skb->dev && skb->dev->hard_header_parse) { |
| 431 | struct nfulnl_msg_packet_hw phw; |
| 432 | |
| 433 | phw.hw_addrlen = |
| 434 | skb->dev->hard_header_parse((struct sk_buff *)skb, |
| 435 | phw.hw_addr); |
| 436 | phw.hw_addrlen = htons(phw.hw_addrlen); |
| 437 | NFA_PUT(inst->skb, NFULA_HWADDR, sizeof(phw), &phw); |
| 438 | } |
| 439 | |
| 440 | if (skb->stamp.tv_sec) { |
| 441 | struct nfulnl_msg_packet_timestamp ts; |
| 442 | |
| 443 | ts.sec = cpu_to_be64(skb->stamp.tv_sec); |
| 444 | ts.usec = cpu_to_be64(skb->stamp.tv_usec); |
| 445 | |
| 446 | NFA_PUT(inst->skb, NFULA_TIMESTAMP, sizeof(ts), &ts); |
| 447 | } |
| 448 | |
| 449 | /* UID */ |
| 450 | if (skb->sk) { |
| 451 | read_lock_bh(&skb->sk->sk_callback_lock); |
| 452 | if (skb->sk->sk_socket && skb->sk->sk_socket->file) { |
| 453 | u_int32_t uid = htonl(skb->sk->sk_socket->file->f_uid); |
| 454 | /* need to unlock here since NFA_PUT may goto */ |
| 455 | read_unlock_bh(&skb->sk->sk_callback_lock); |
| 456 | NFA_PUT(inst->skb, NFULA_UID, sizeof(uid), &uid); |
| 457 | } else |
| 458 | read_unlock_bh(&skb->sk->sk_callback_lock); |
| 459 | } |
| 460 | |
| 461 | if (data_len) { |
| 462 | struct nfattr *nfa; |
| 463 | int size = NFA_LENGTH(data_len); |
| 464 | |
| 465 | if (skb_tailroom(inst->skb) < (int)NFA_SPACE(data_len)) { |
| 466 | printk(KERN_WARNING "nfnetlink_log: no tailroom!\n"); |
| 467 | goto nlmsg_failure; |
| 468 | } |
| 469 | |
| 470 | nfa = (struct nfattr *)skb_put(inst->skb, NFA_ALIGN(size)); |
| 471 | nfa->nfa_type = NFULA_PAYLOAD; |
| 472 | nfa->nfa_len = size; |
| 473 | |
| 474 | if (skb_copy_bits(skb, 0, NFA_DATA(nfa), data_len)) |
| 475 | BUG(); |
| 476 | } |
| 477 | |
| 478 | nlh->nlmsg_len = inst->skb->tail - old_tail; |
| 479 | return 0; |
| 480 | |
| 481 | nlmsg_failure: |
| 482 | UDEBUG("nlmsg_failure\n"); |
| 483 | nfattr_failure: |
| 484 | PRINTR(KERN_ERR "nfnetlink_log: error creating log nlmsg\n"); |
| 485 | return -1; |
| 486 | } |
| 487 | |
| 488 | #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0) |
| 489 | |
| 490 | static struct nf_loginfo default_loginfo = { |
| 491 | .type = NF_LOG_TYPE_ULOG, |
| 492 | .u = { |
| 493 | .ulog = { |
| 494 | .copy_len = 0xffff, |
| 495 | .group = 0, |
| 496 | .qthreshold = 1, |
| 497 | }, |
| 498 | }, |
| 499 | }; |
| 500 | |
| 501 | /* log handler for internal netfilter logging api */ |
| 502 | static void |
| 503 | nfulnl_log_packet(unsigned int pf, |
| 504 | unsigned int hooknum, |
| 505 | const struct sk_buff *skb, |
| 506 | const struct net_device *in, |
| 507 | const struct net_device *out, |
| 508 | const struct nf_loginfo *li_user, |
| 509 | const char *prefix) |
| 510 | { |
| 511 | unsigned int size, data_len; |
| 512 | struct nfulnl_instance *inst; |
| 513 | const struct nf_loginfo *li; |
| 514 | unsigned int qthreshold; |
| 515 | unsigned int nlbufsiz; |
| 516 | |
| 517 | if (li_user && li_user->type == NF_LOG_TYPE_ULOG) |
| 518 | li = li_user; |
| 519 | else |
| 520 | li = &default_loginfo; |
| 521 | |
| 522 | inst = instance_lookup_get(li->u.ulog.group); |
| 523 | if (!inst) |
| 524 | inst = instance_lookup_get(0); |
| 525 | if (!inst) { |
| 526 | PRINTR("nfnetlink_log: trying to log packet, " |
| 527 | "but no instance for group %u\n", li->u.ulog.group); |
| 528 | return; |
| 529 | } |
| 530 | |
| 531 | /* all macros expand to constant values at compile time */ |
| 532 | /* FIXME: do we want to make the size calculation conditional based on |
| 533 | * what is actually present? way more branches and checks, but more |
| 534 | * memory efficient... */ |
| 535 | size = NLMSG_SPACE(sizeof(struct nfgenmsg)) |
| 536 | + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hdr)) |
| 537 | + NFA_SPACE(sizeof(u_int32_t)) /* ifindex */ |
| 538 | + NFA_SPACE(sizeof(u_int32_t)) /* ifindex */ |
| 539 | + NFA_SPACE(sizeof(u_int32_t)) /* mark */ |
| 540 | + NFA_SPACE(sizeof(u_int32_t)) /* uid */ |
| 541 | + NFA_SPACE(NFULNL_PREFIXLEN) /* prefix */ |
| 542 | + NFA_SPACE(sizeof(struct nfulnl_msg_packet_hw)) |
| 543 | + NFA_SPACE(sizeof(struct nfulnl_msg_packet_timestamp)); |
| 544 | |
| 545 | UDEBUG("initial size=%u\n", size); |
| 546 | |
| 547 | spin_lock_bh(&inst->lock); |
| 548 | |
| 549 | qthreshold = inst->qthreshold; |
| 550 | /* per-rule qthreshold overrides per-instance */ |
| 551 | if (qthreshold > li->u.ulog.qthreshold) |
| 552 | qthreshold = li->u.ulog.qthreshold; |
| 553 | |
| 554 | switch (inst->copy_mode) { |
| 555 | case NFULNL_COPY_META: |
| 556 | case NFULNL_COPY_NONE: |
| 557 | data_len = 0; |
| 558 | break; |
| 559 | |
| 560 | case NFULNL_COPY_PACKET: |
| 561 | if (inst->copy_range == 0 |
| 562 | || inst->copy_range > skb->len) |
| 563 | data_len = skb->len; |
| 564 | else |
| 565 | data_len = inst->copy_range; |
| 566 | |
| 567 | size += NFA_SPACE(data_len); |
| 568 | UDEBUG("copy_packet, therefore size now %u\n", size); |
| 569 | break; |
| 570 | |
| 571 | default: |
| 572 | spin_unlock_bh(&inst->lock); |
| 573 | instance_put(inst); |
| 574 | return; |
| 575 | } |
| 576 | |
| 577 | if (size > inst->nlbufsiz) |
| 578 | nlbufsiz = size; |
| 579 | else |
| 580 | nlbufsiz = inst->nlbufsiz; |
| 581 | |
| 582 | if (!inst->skb) { |
| 583 | if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) { |
| 584 | UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n", |
| 585 | inst->nlbufsiz, size); |
| 586 | goto alloc_failure; |
| 587 | } |
| 588 | } else if (inst->qlen >= qthreshold || |
| 589 | size > skb_tailroom(inst->skb)) { |
| 590 | /* either the queue len is too high or we don't have |
| 591 | * enough room in the skb left. flush to userspace. */ |
| 592 | UDEBUG("flushing old skb\n"); |
| 593 | |
| 594 | __nfulnl_send(inst); |
| 595 | |
| 596 | if (!(inst->skb = nfulnl_alloc_skb(nlbufsiz, size))) { |
| 597 | UDEBUG("error in nfulnl_alloc_skb(%u, %u)\n", |
| 598 | inst->nlbufsiz, size); |
| 599 | goto alloc_failure; |
| 600 | } |
| 601 | } |
| 602 | |
| 603 | UDEBUG("qlen %d, qthreshold %d\n", inst->qlen, qthreshold); |
| 604 | inst->qlen++; |
| 605 | |
| 606 | __build_packet_message(inst, skb, data_len, pf, |
| 607 | hooknum, in, out, li, prefix); |
| 608 | |
| 609 | /* timer_pending always called within inst->lock, so there |
| 610 | * is no chance of a race here */ |
| 611 | if (!timer_pending(&inst->timer)) { |
| 612 | instance_get(inst); |
| 613 | inst->timer.expires = jiffies + (inst->flushtimeout*HZ/100); |
| 614 | add_timer(&inst->timer); |
| 615 | } |
| 616 | spin_unlock_bh(&inst->lock); |
| 617 | |
| 618 | return; |
| 619 | |
| 620 | alloc_failure: |
| 621 | spin_unlock_bh(&inst->lock); |
| 622 | instance_put(inst); |
| 623 | UDEBUG("error allocating skb\n"); |
| 624 | /* FIXME: statistics */ |
| 625 | } |
| 626 | |
| 627 | static int |
| 628 | nfulnl_rcv_nl_event(struct notifier_block *this, |
| 629 | unsigned long event, void *ptr) |
| 630 | { |
| 631 | struct netlink_notify *n = ptr; |
| 632 | |
| 633 | if (event == NETLINK_URELEASE && |
| 634 | n->protocol == NETLINK_NETFILTER && n->pid) { |
| 635 | int i; |
| 636 | |
| 637 | /* destroy all instances for this pid */ |
| 638 | write_lock_bh(&instances_lock); |
| 639 | for (i = 0; i < INSTANCE_BUCKETS; i++) { |
| 640 | struct hlist_node *tmp, *t2; |
| 641 | struct nfulnl_instance *inst; |
| 642 | struct hlist_head *head = &instance_table[i]; |
| 643 | |
| 644 | hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) { |
| 645 | UDEBUG("node = %p\n", inst); |
| 646 | if (n->pid == inst->peer_pid) |
| 647 | __instance_destroy(inst); |
| 648 | } |
| 649 | } |
| 650 | write_unlock_bh(&instances_lock); |
| 651 | } |
| 652 | return NOTIFY_DONE; |
| 653 | } |
| 654 | |
| 655 | static struct notifier_block nfulnl_rtnl_notifier = { |
| 656 | .notifier_call = nfulnl_rcv_nl_event, |
| 657 | }; |
| 658 | |
| 659 | static int |
| 660 | nfulnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb, |
| 661 | struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp) |
| 662 | { |
| 663 | return -ENOTSUPP; |
| 664 | } |
| 665 | |
| 666 | static struct nf_logger nfulnl_logger = { |
| 667 | .name = "nfnetlink_log", |
| 668 | .logfn = &nfulnl_log_packet, |
| 669 | .me = THIS_MODULE, |
| 670 | }; |
| 671 | |
| 672 | static const int nfula_min[NFULA_MAX] = { |
| 673 | [NFULA_PACKET_HDR-1] = sizeof(struct nfulnl_msg_packet_hdr), |
| 674 | [NFULA_MARK-1] = sizeof(u_int32_t), |
| 675 | [NFULA_TIMESTAMP-1] = sizeof(struct nfulnl_msg_packet_timestamp), |
| 676 | [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t), |
| 677 | [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t), |
| 678 | [NFULA_HWADDR-1] = sizeof(struct nfulnl_msg_packet_hw), |
| 679 | [NFULA_PAYLOAD-1] = 0, |
| 680 | [NFULA_PREFIX-1] = 0, |
| 681 | [NFULA_UID-1] = sizeof(u_int32_t), |
| 682 | }; |
| 683 | |
| 684 | static const int nfula_cfg_min[NFULA_CFG_MAX] = { |
| 685 | [NFULA_CFG_CMD-1] = sizeof(struct nfulnl_msg_config_cmd), |
| 686 | [NFULA_CFG_MODE-1] = sizeof(struct nfulnl_msg_config_mode), |
| 687 | [NFULA_CFG_TIMEOUT-1] = sizeof(u_int32_t), |
| 688 | [NFULA_CFG_QTHRESH-1] = sizeof(u_int32_t), |
| 689 | [NFULA_CFG_NLBUFSIZ-1] = sizeof(u_int32_t), |
| 690 | }; |
| 691 | |
| 692 | static int |
| 693 | nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb, |
| 694 | struct nlmsghdr *nlh, struct nfattr *nfula[], int *errp) |
| 695 | { |
| 696 | struct nfgenmsg *nfmsg = NLMSG_DATA(nlh); |
| 697 | u_int16_t group_num = ntohs(nfmsg->res_id); |
| 698 | struct nfulnl_instance *inst; |
| 699 | int ret = 0; |
| 700 | |
| 701 | UDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type)); |
| 702 | |
| 703 | if (nfattr_bad_size(nfula, NFULA_CFG_MAX, nfula_cfg_min)) { |
| 704 | UDEBUG("bad attribute size\n"); |
| 705 | return -EINVAL; |
| 706 | } |
| 707 | |
| 708 | inst = instance_lookup_get(group_num); |
| 709 | if (nfula[NFULA_CFG_CMD-1]) { |
| 710 | u_int8_t pf = nfmsg->nfgen_family; |
| 711 | struct nfulnl_msg_config_cmd *cmd; |
| 712 | cmd = NFA_DATA(nfula[NFULA_CFG_CMD-1]); |
| 713 | UDEBUG("found CFG_CMD for\n"); |
| 714 | |
| 715 | switch (cmd->command) { |
| 716 | case NFULNL_CFG_CMD_BIND: |
| 717 | if (inst) { |
| 718 | ret = -EBUSY; |
| 719 | goto out_put; |
| 720 | } |
| 721 | |
| 722 | inst = instance_create(group_num, |
| 723 | NETLINK_CB(skb).pid); |
| 724 | if (!inst) { |
| 725 | ret = -EINVAL; |
| 726 | goto out_put; |
| 727 | } |
| 728 | break; |
| 729 | case NFULNL_CFG_CMD_UNBIND: |
| 730 | if (!inst) { |
| 731 | ret = -ENODEV; |
| 732 | goto out_put; |
| 733 | } |
| 734 | |
| 735 | if (inst->peer_pid != NETLINK_CB(skb).pid) { |
| 736 | ret = -EPERM; |
| 737 | goto out_put; |
| 738 | } |
| 739 | |
| 740 | instance_destroy(inst); |
| 741 | break; |
| 742 | case NFULNL_CFG_CMD_PF_BIND: |
| 743 | UDEBUG("registering log handler for pf=%u\n", pf); |
| 744 | ret = nf_log_register(pf, &nfulnl_logger); |
| 745 | break; |
| 746 | case NFULNL_CFG_CMD_PF_UNBIND: |
| 747 | UDEBUG("unregistering log handler for pf=%u\n", pf); |
| 748 | /* This is a bug and a feature. We cannot unregister |
| 749 | * other handlers, like nfnetlink_inst can */ |
| 750 | nf_log_unregister_pf(pf); |
| 751 | break; |
| 752 | default: |
| 753 | ret = -EINVAL; |
| 754 | break; |
| 755 | } |
| 756 | } else { |
| 757 | if (!inst) { |
| 758 | UDEBUG("no config command, and no instance for " |
| 759 | "group=%u pid=%u =>ENOENT\n", |
| 760 | group_num, NETLINK_CB(skb).pid); |
| 761 | ret = -ENOENT; |
| 762 | goto out_put; |
| 763 | } |
| 764 | |
| 765 | if (inst->peer_pid != NETLINK_CB(skb).pid) { |
| 766 | UDEBUG("no config command, and wrong pid\n"); |
| 767 | ret = -EPERM; |
| 768 | goto out_put; |
| 769 | } |
| 770 | } |
| 771 | |
| 772 | if (nfula[NFULA_CFG_MODE-1]) { |
| 773 | struct nfulnl_msg_config_mode *params; |
| 774 | params = NFA_DATA(nfula[NFULA_CFG_MODE-1]); |
| 775 | |
| 776 | nfulnl_set_mode(inst, params->copy_mode, |
| 777 | ntohs(params->copy_range)); |
| 778 | } |
| 779 | |
| 780 | if (nfula[NFULA_CFG_TIMEOUT-1]) { |
| 781 | u_int32_t timeout = |
| 782 | *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_TIMEOUT-1]); |
| 783 | |
| 784 | nfulnl_set_timeout(inst, ntohl(timeout)); |
| 785 | } |
| 786 | |
| 787 | if (nfula[NFULA_CFG_NLBUFSIZ-1]) { |
| 788 | u_int32_t nlbufsiz = |
| 789 | *(u_int32_t *)NFA_DATA(nfula[NFULA_CFG_NLBUFSIZ-1]); |
| 790 | |
| 791 | nfulnl_set_nlbufsiz(inst, ntohl(nlbufsiz)); |
| 792 | } |
| 793 | |
| 794 | if (nfula[NFULA_CFG_QTHRESH-1]) { |
| 795 | u_int32_t qthresh = |
| 796 | *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_QTHRESH-1]); |
| 797 | |
| 798 | nfulnl_set_qthresh(inst, ntohl(qthresh)); |
| 799 | } |
| 800 | |
| 801 | out_put: |
| 802 | instance_put(inst); |
| 803 | return ret; |
| 804 | } |
| 805 | |
| 806 | static struct nfnl_callback nfulnl_cb[NFULNL_MSG_MAX] = { |
| 807 | [NFULNL_MSG_PACKET] = { .call = nfulnl_recv_unsupp, |
Harald Welte | 927ccbc | 2005-08-09 20:03:40 -0700 | [diff] [blame] | 808 | .attr_count = NFULA_MAX, |
| 809 | .cap_required = CAP_NET_ADMIN, }, |
Harald Welte | 0597f26 | 2005-08-09 19:58:39 -0700 | [diff] [blame] | 810 | [NFULNL_MSG_CONFIG] = { .call = nfulnl_recv_config, |
Harald Welte | 927ccbc | 2005-08-09 20:03:40 -0700 | [diff] [blame] | 811 | .attr_count = NFULA_CFG_MAX, |
Harald Welte | 0597f26 | 2005-08-09 19:58:39 -0700 | [diff] [blame] | 812 | .cap_required = CAP_NET_ADMIN }, |
| 813 | }; |
| 814 | |
| 815 | static struct nfnetlink_subsystem nfulnl_subsys = { |
| 816 | .name = "log", |
| 817 | .subsys_id = NFNL_SUBSYS_ULOG, |
| 818 | .cb_count = NFULNL_MSG_MAX, |
Harald Welte | 0597f26 | 2005-08-09 19:58:39 -0700 | [diff] [blame] | 819 | .cb = nfulnl_cb, |
| 820 | }; |
| 821 | |
| 822 | #ifdef CONFIG_PROC_FS |
| 823 | struct iter_state { |
| 824 | unsigned int bucket; |
| 825 | }; |
| 826 | |
| 827 | static struct hlist_node *get_first(struct seq_file *seq) |
| 828 | { |
| 829 | struct iter_state *st = seq->private; |
| 830 | |
| 831 | if (!st) |
| 832 | return NULL; |
| 833 | |
| 834 | for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) { |
| 835 | if (!hlist_empty(&instance_table[st->bucket])) |
| 836 | return instance_table[st->bucket].first; |
| 837 | } |
| 838 | return NULL; |
| 839 | } |
| 840 | |
| 841 | static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h) |
| 842 | { |
| 843 | struct iter_state *st = seq->private; |
| 844 | |
| 845 | h = h->next; |
| 846 | while (!h) { |
| 847 | if (++st->bucket >= INSTANCE_BUCKETS) |
| 848 | return NULL; |
| 849 | |
| 850 | h = instance_table[st->bucket].first; |
| 851 | } |
| 852 | return h; |
| 853 | } |
| 854 | |
| 855 | static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos) |
| 856 | { |
| 857 | struct hlist_node *head; |
| 858 | head = get_first(seq); |
| 859 | |
| 860 | if (head) |
| 861 | while (pos && (head = get_next(seq, head))) |
| 862 | pos--; |
| 863 | return pos ? NULL : head; |
| 864 | } |
| 865 | |
| 866 | static void *seq_start(struct seq_file *seq, loff_t *pos) |
| 867 | { |
| 868 | read_lock_bh(&instances_lock); |
| 869 | return get_idx(seq, *pos); |
| 870 | } |
| 871 | |
| 872 | static void *seq_next(struct seq_file *s, void *v, loff_t *pos) |
| 873 | { |
| 874 | (*pos)++; |
| 875 | return get_next(s, v); |
| 876 | } |
| 877 | |
| 878 | static void seq_stop(struct seq_file *s, void *v) |
| 879 | { |
| 880 | read_unlock_bh(&instances_lock); |
| 881 | } |
| 882 | |
| 883 | static int seq_show(struct seq_file *s, void *v) |
| 884 | { |
| 885 | const struct nfulnl_instance *inst = v; |
| 886 | |
| 887 | return seq_printf(s, "%5d %6d %5d %1d %5d %6d %2d\n", |
| 888 | inst->group_num, |
| 889 | inst->peer_pid, inst->qlen, |
| 890 | inst->copy_mode, inst->copy_range, |
| 891 | inst->flushtimeout, atomic_read(&inst->use)); |
| 892 | } |
| 893 | |
| 894 | static struct seq_operations nful_seq_ops = { |
| 895 | .start = seq_start, |
| 896 | .next = seq_next, |
| 897 | .stop = seq_stop, |
| 898 | .show = seq_show, |
| 899 | }; |
| 900 | |
| 901 | static int nful_open(struct inode *inode, struct file *file) |
| 902 | { |
| 903 | struct seq_file *seq; |
| 904 | struct iter_state *is; |
| 905 | int ret; |
| 906 | |
| 907 | is = kmalloc(sizeof(*is), GFP_KERNEL); |
| 908 | if (!is) |
| 909 | return -ENOMEM; |
| 910 | memset(is, 0, sizeof(*is)); |
| 911 | ret = seq_open(file, &nful_seq_ops); |
| 912 | if (ret < 0) |
| 913 | goto out_free; |
| 914 | seq = file->private_data; |
| 915 | seq->private = is; |
| 916 | return ret; |
| 917 | out_free: |
| 918 | kfree(is); |
| 919 | return ret; |
| 920 | } |
| 921 | |
| 922 | static struct file_operations nful_file_ops = { |
| 923 | .owner = THIS_MODULE, |
| 924 | .open = nful_open, |
| 925 | .read = seq_read, |
| 926 | .llseek = seq_lseek, |
| 927 | .release = seq_release_private, |
| 928 | }; |
| 929 | |
| 930 | #endif /* PROC_FS */ |
| 931 | |
| 932 | static int |
| 933 | init_or_cleanup(int init) |
| 934 | { |
| 935 | int i, status = -ENOMEM; |
| 936 | #ifdef CONFIG_PROC_FS |
| 937 | struct proc_dir_entry *proc_nful; |
| 938 | #endif |
| 939 | |
| 940 | if (!init) |
| 941 | goto cleanup; |
| 942 | |
| 943 | for (i = 0; i < INSTANCE_BUCKETS; i++) |
| 944 | INIT_HLIST_HEAD(&instance_table[i]); |
| 945 | |
| 946 | /* it's not really all that important to have a random value, so |
| 947 | * we can do this from the init function, even if there hasn't |
| 948 | * been that much entropy yet */ |
| 949 | get_random_bytes(&hash_init, sizeof(hash_init)); |
| 950 | |
| 951 | netlink_register_notifier(&nfulnl_rtnl_notifier); |
| 952 | status = nfnetlink_subsys_register(&nfulnl_subsys); |
| 953 | if (status < 0) { |
| 954 | printk(KERN_ERR "log: failed to create netlink socket\n"); |
| 955 | goto cleanup_netlink_notifier; |
| 956 | } |
| 957 | |
| 958 | #ifdef CONFIG_PROC_FS |
| 959 | proc_nful = create_proc_entry("nfnetlink_log", 0440, |
| 960 | proc_net_netfilter); |
| 961 | if (!proc_nful) |
| 962 | goto cleanup_subsys; |
| 963 | proc_nful->proc_fops = &nful_file_ops; |
| 964 | #endif |
| 965 | |
| 966 | return status; |
| 967 | |
| 968 | cleanup: |
| 969 | nf_log_unregister_logger(&nfulnl_logger); |
| 970 | #ifdef CONFIG_PROC_FS |
| 971 | remove_proc_entry("nfnetlink_log", proc_net_netfilter); |
| 972 | cleanup_subsys: |
| 973 | #endif |
| 974 | nfnetlink_subsys_unregister(&nfulnl_subsys); |
| 975 | cleanup_netlink_notifier: |
| 976 | netlink_unregister_notifier(&nfulnl_rtnl_notifier); |
| 977 | return status; |
| 978 | } |
| 979 | |
| 980 | static int __init init(void) |
| 981 | { |
| 982 | |
| 983 | return init_or_cleanup(1); |
| 984 | } |
| 985 | |
| 986 | static void __exit fini(void) |
| 987 | { |
| 988 | init_or_cleanup(0); |
| 989 | } |
| 990 | |
| 991 | MODULE_DESCRIPTION("netfilter userspace logging"); |
| 992 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); |
| 993 | MODULE_LICENSE("GPL"); |
Harald Welte | f682fae | 2005-08-09 20:20:34 -0700 | [diff] [blame^] | 994 | MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ULOG); |
Harald Welte | 0597f26 | 2005-08-09 19:58:39 -0700 | [diff] [blame] | 995 | |
| 996 | module_init(init); |
| 997 | module_exit(fini); |