2 * Copyright (C) 2009 Luigi Rizzo, Marta Carbone, Universita` di Pisa
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * The main interface to build ipfw+dummynet as a linux module.
30 * (and possibly as a windows module as well, though that part
31 * is not complete yet).
33 * The control interface uses the sockopt mechanism
34 * on a socket(AF_INET, SOCK_RAW, IPPROTO_RAW).
36 * The data interface uses the netfilter interface, at the moment
37 * hooked to the PRE_ROUTING and POST_ROUTING hooks.
38 * Unfortunately the netfilter interface is a moving target,
39 * so we need a set of macros to adapt to the various cases.
41 * In the netfilter hook we just mark packet as 'QUEUE' and then
42 * let the queue handler to do the whole work (filtering and
43 * possibly emulation).
44 * As we receive packets, we wrap them with an mbuf descriptor
45 * so the existing ipfw+dummynet code runs unmodified.
48 #include <sys/cdefs.h>
49 #include <sys/mbuf.h> /* sizeof struct mbuf */
50 #include <sys/param.h> /* NGROUPS */
53 #include <linux/module.h>
54 #include <linux/kernel.h>
55 #include <linux/netfilter.h>
56 #include <linux/netfilter_ipv4.h> /* NF_IP_PRI_FILTER */
58 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,25)
59 #include <net/netfilter/nf_queue.h> /* nf_queue */
62 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
66 #endif /* !__linux__ */
68 #include <netinet/in.h> /* in_addr */
69 #include <netinet/ip_fw.h> /* ip_fw_ctl_t, ip_fw_chk_t */
70 #include <netinet/ip_dummynet.h> /* ip_dn_ctl_t, ip_dn_io_t */
71 #include <net/pfil.h> /* PFIL_IN, PFIL_OUT */
72 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
73 #warning --- inet_hashtables not present on 2.4
74 #include <linux/tcp.h>
75 #include <net/route.h>
77 static inline int inet_iif(const struct sk_buff *skb)
79 return ((struct rtable *)skb->dst)->rt_iif;
83 #include <net/inet_hashtables.h> /* inet_lookup */
85 #include <net/route.h> /* inet_iif */
88 * Here we allocate some global variables used in the firewall.
90 //ip_dn_ctl_t *ip_dn_ctl_ptr;
91 int (*ip_dn_ctl_ptr)(struct sockopt *);
93 ip_fw_ctl_t *ip_fw_ctl_ptr;
95 int (*ip_dn_io_ptr)(struct mbuf **m, int dir, struct ip_fw_args *fwa);
96 ip_fw_chk_t *ip_fw_chk_ptr;
98 void (*bridge_dn_p)(struct mbuf *, struct ifnet *);
101 * Glue code to implement the registration of children with the parent.
102 * Each child should call my_mod_register() when linking, so that
103 * module_init() and module_exit() can call init_children() and
104 * fini_children() to provide the necessary initialization.
106 #include <sys/module.h>
108 struct moduledata *mod;
113 static unsigned int mod_idx;
114 static struct mod_args mods[10]; /* hard limit to 10 modules */
117 * my_mod_register should be called automatically as the init
118 * functions in the submodules. Unfortunately this compiler/linker
119 * trick is not supported yet so we call it manually.
122 my_mod_register(struct moduledata *mod, const char *name, int order)
124 struct mod_args m = { mod, name, order };
126 printf("%s %s called\n", __FUNCTION__, name);
127 if (mod_idx < sizeof(mods) / sizeof(mods[0]))
137 /* Call the functions registered at init time. */
138 printf("%s mod_idx value %d\n", __FUNCTION__, mod_idx);
139 for (i = 0; i < mod_idx; i++) {
140 printf("+++ start module %d %s %s at %p order 0x%x\n",
141 i, mods[i].name, mods[i].mod->name,
142 mods[i].mod, mods[i].order);
143 mods[i].mod->evhand(NULL, MOD_LOAD, mods[i].mod->priv);
152 /* Call the functions registered at init time. */
153 for (i = mod_idx - 1; i >= 0; i--) {
154 printf("+++ end module %d %s %s at %p order 0x%x\n",
155 i, mods[i].name, mods[i].mod->name,
156 mods[i].mod, mods[i].order);
157 mods[i].mod->evhand(NULL, MOD_UNLOAD, mods[i].mod->priv);
160 /*--- end of module binding helper functions ---*/
164 * ipfw_ctl_h() is a wrapper for linux to FreeBSD sockopt call convention.
165 * then call the ipfw handler in order to manage requests.
166 * In turn this is called by the linux set/get handlers.
169 ipfw_ctl_h(struct sockopt *s, int cmd, int dir, int len, void __user *user)
174 memset(s, 0, sizeof(s));
177 s->sopt_valsize = len;
180 /* sopt_td is not used but it is referenced */
181 memset(&t, 0, sizeof(t));
184 // printf("%s called with cmd %d len %d\n", __FUNCTION__, cmd, len);
186 if (cmd < IP_DUMMYNET_CONFIGURE && ip_fw_ctl_ptr)
187 ret = ip_fw_ctl_ptr(s);
188 else if (cmd >= IP_DUMMYNET_CONFIGURE && ip_dn_ctl_ptr)
189 ret = ip_dn_ctl_ptr(s);
191 return -ret; /* errors are < 0 on linux */
197 netisr_dispatch(int __unused num, struct mbuf *m)
202 ip_output(struct mbuf *m, struct mbuf __unused *opt,
203 struct route __unused *ro, int __unused flags,
204 struct ip_moptions __unused *imo, struct inpcb __unused *inp)
206 netisr_dispatch(0, m);
210 #else /* this is the linux glue */
212 * setsockopt hook has no return value other than the error code.
215 do_ipfw_set_ctl(struct sock __unused *sk, int cmd,
216 void __user *user, unsigned int len)
218 struct sockopt s; /* pass arguments */
220 return ipfw_ctl_h(&s, cmd, SOPT_SET, len, user);
224 * getsockopt can can return a block of data in response.
227 do_ipfw_get_ctl(struct sock __unused *sk,
228 int cmd, void __user *user, int *len)
230 struct sockopt s; /* pass arguments */
231 int ret = ipfw_ctl_h(&s, cmd, SOPT_GET, *len, user);
233 *len = s.sopt_valsize; /* return lenght back to the caller */
238 * declare our [get|set]sockopt hooks
240 static struct nf_sockopt_ops ipfw_sockopts = {
242 .set_optmin = _IPFW_SOCKOPT_BASE,
243 .set_optmax = _IPFW_SOCKOPT_END,
244 .set = do_ipfw_set_ctl,
245 .get_optmin = _IPFW_SOCKOPT_BASE,
246 .get_optmax = _IPFW_SOCKOPT_END,
247 .get = do_ipfw_get_ctl,
248 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
249 .owner = THIS_MODULE,
254 * We need a number of macros to adapt to the various APIs in
255 * different linux versions. Among them:
257 * - the hook names change between macros (NF_IP*) and enum NF_INET_*
259 * - the second argument to the netfilter hook is
260 * struct sk_buff ** in kernels <= 2.6.22
261 * struct sk_buff * in kernels > 2.6.22
263 * - NF_STOP is not defined before 2.6 so we remap it to NF_ACCEPT
265 * - the packet descriptor passed to the queue handler is
266 * struct nf_info in kernels <= 2.6.24
267 * struct nf_queue_entry in kernels <= 2.6.24
269 * - the arguments to the queue handler also change;
273 * declare hook to grab packets from the netfilter interface.
274 * The NF_* names change in different versions of linux, in some
275 * cases they are #defines, in others they are enum, so we
278 #ifndef NF_IP_PRE_ROUTING
279 #define NF_IP_PRE_ROUTING NF_INET_PRE_ROUTING
281 #ifndef NF_IP_POST_ROUTING
282 #define NF_IP_POST_ROUTING NF_INET_POST_ROUTING
286 * ipfw hooks into the POST_ROUTING and the PRE_ROUTING chains.
287 * PlanetLab sets skb_tag to the slice id in the LOCAL_INPUT and
288 * POST_ROUTING chains, so if we want to use that information we
289 * need to hook the LOCAL_INPUT chain instead of the PRE_ROUTING.
290 * However at the moment the skb_tag info is not reliable so
291 * we stay with the standard hooks.
293 #if 0 // defined(IPFW_PLANETLAB)
294 #define IPFW_HOOK_IN NF_IP_LOCAL_IN
296 #define IPFW_HOOK_IN NF_IP_PRE_ROUTING
300 * The main netfilter hook.
301 * To make life simple, we queue everything and then do all the
302 * decision in the queue handler.
304 * XXX note that in 2.4 and up to 2.6.22 the skbuf is passed as sk_buff**
305 * so we have an #ifdef to set the proper argument type.
308 call_ipfw(unsigned int __unused hooknum,
309 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23) // in 2.6.22 we have **
310 struct sk_buff __unused **skb,
312 struct sk_buff __unused *skb,
314 const struct net_device __unused *in,
315 const struct net_device __unused *out,
316 int __unused (*okfn)(struct sk_buff *))
321 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
322 #define NF_STOP NF_ACCEPT
325 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,25)
328 * nf_queue_entry is a recent addition, in previous versions
329 * of the code the struct is called nf_info.
331 #define nf_queue_entry nf_info /* for simplicity */
333 /* also, 2.4 and perhaps something else have different arguments */
334 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0) /* unsure on the exact boundary */
335 /* on 2.4 we use nf_info */
336 #define QH_ARGS struct sk_buff *skb, struct nf_info *info, void *data
337 #else /* 2.6.1.. 2.6.24 */
338 #define QH_ARGS struct sk_buff *skb, struct nf_info *info, unsigned int qnum, void *data
341 #define DEFINE_SKB /* nothing, already an argument */
342 #define REINJECT(_inf, _verd) nf_reinject(skb, _inf, _verd)
344 #else /* 2.6.25 and above */
346 #define QH_ARGS struct nf_queue_entry *info, unsigned int queuenum
347 #define DEFINE_SKB struct sk_buff *skb = info->skb;
348 #define REINJECT(_inf, _verd) nf_reinject(_inf, _verd)
352 * used by dummynet when dropping packets
353 * XXX use dummynet_send()
356 reinject_drop(struct mbuf* m)
358 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,25) /* unsure on the exact boundary */
359 struct sk_buff *skb = (struct sk_buff *)m;
361 REINJECT(m->queue_entry, NF_DROP);
365 * The real call to the firewall. nf_queue_entry points to the skbuf,
366 * and eventually we need to return both through nf_reinject().
369 ipfw2_queue_handler(QH_ARGS)
371 DEFINE_SKB /* no semicolon here, goes in the macro */
372 int ret = 0; /* return value */
375 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
376 if (skb->nh.iph == NULL) {
377 printf("null dp, len %d reinject now\n", skb->len);
378 REINJECT(info, NF_ACCEPT);
382 m = malloc(sizeof(*m), 0, 0);
384 printf("malloc fail, len %d reinject now\n", skb->len);
385 REINJECT(info, NF_ACCEPT);
390 m->m_len = skb->len; /* len in this skbuf */
391 m->m_pkthdr.len = skb->len; /* total packet len */
392 m->m_pkthdr.rcvif = info->indev;
393 m->queue_entry = info;
394 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
395 m->m_data = skb->nh.iph;
397 m->m_data = skb_network_header(skb);
400 /* XXX add the interface */
401 if (info->hook == IPFW_HOOK_IN) {
402 ret = ipfw_check_in(NULL, &m, info->indev, PFIL_IN, NULL);
404 ret = ipfw_check_out(NULL, &m, info->outdev, PFIL_OUT, NULL);
407 if (m != NULL) { /* Accept. reinject and free the mbuf */
408 REINJECT(info, NF_ACCEPT);
410 } else if (ret == 0) {
411 /* dummynet has kept the packet, will reinject later. */
414 * Packet dropped by ipfw or dummynet, reinject as NF_DROP
415 * mbuf already released by ipfw itself
417 REINJECT(info, NF_DROP);
427 /* XXX should include prototypes for netisr_dispatch and ip_output */
429 * The reinjection routine after a packet comes out from dummynet.
430 * We must update the skb timestamp so ping reports the right time.
433 netisr_dispatch(int num, struct mbuf *m)
435 struct nf_queue_entry *info = m->queue_entry;
436 struct sk_buff *skb = m->m_skb; /* always used */
440 KASSERT((info != NULL), ("%s info null!\n", __FUNCTION__));
441 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) // XXX above 2.6.x ?
442 __net_timestamp(skb); /* update timestamp */
445 /* XXX to obey one-pass, possibly call the queue handler here */
446 REINJECT(info, ((num == -1)?NF_DROP:NF_STOP)); /* accept but no more firewall */
450 ip_output(struct mbuf *m, struct mbuf __unused *opt,
451 struct route __unused *ro, int __unused flags,
452 struct ip_moptions __unused *imo, struct inpcb __unused *inp)
454 netisr_dispatch(0, m);
459 * socket lookup function for linux.
460 * This code is used to associate uid, gid, jail/xid to packets,
461 * and store the info in a cache *ugp where they can be accessed quickly.
462 * The function returns 1 if the info is found, -1 otherwise.
464 * We do this only on selected protocols: TCP, ...
466 * The chain is the following
467 * sk_buff* sock* socket* file*
468 * skb -> sk ->sk_socket->file ->f_owner ->pid
469 * skb -> sk ->sk_socket->file ->f_uid (direct)
470 * skb -> sk ->sk_socket->file ->f_cred->fsuid (2.6.29+)
473 * linux/skbuff.h struct skbuff
474 * net/sock.h struct sock
475 * linux/net.h struct socket
476 * linux/fs.h struct file
478 * With vserver we may have sk->sk_xid and sk->sk_nid that
479 * which we store in fw_groups[1] (matches O_JAIL) and fw_groups[2]
482 * Note- for locally generated, outgoing packets we should not need
483 * need a lookup because the sk_buff already points to the socket where
486 extern struct inet_hashinfo tcp_hashinfo;
488 linux_lookup(const int proto, const __be32 saddr, const __be16 sport,
489 const __be32 daddr, const __be16 dport,
490 struct sk_buff *skb, int dir, struct bsd_ucred *u)
492 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,0)
496 int ret = -1; /* default return value */
497 int st = -1; /* state */
500 if (proto != IPPROTO_TCP) /* XXX extend for UDP */
503 if ((dir ? (void *)skb_dst(skb) : (void *)skb->dev) == NULL) {
504 panic(" -- this should not happen\n");
512 * Try a lookup. On a match, sk has a refcount that we must
513 * release on exit (we know it because skb->sk = NULL).
515 * inet_lookup above 2.6.24 has an additional 'net' parameter
516 * so we use a macro to conditionally supply it.
517 * swap dst and src depending on the direction.
519 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,24)
522 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,26)
523 /* there is no dev_net() on 2.6.25 */
524 #define _OPT_NET_ARG (skb->dev->nd_net),
525 #else /* 2.6.26 and above */
526 #define _OPT_NET_ARG dev_net(skb->dev),
529 sk = (dir) ? /* dir != 0 on output */
530 inet_lookup(_OPT_NET_ARG &tcp_hashinfo,
531 daddr, dport, saddr, sport, // match outgoing
533 inet_lookup(_OPT_NET_ARG &tcp_hashinfo,
534 saddr, sport, daddr, dport, // match incoming
538 if (sk == NULL) /* no match, nothing to be done */
541 ret = 1; /* retrying won't make things better */
543 #ifdef CONFIG_VSERVER
550 * Exclude tcp states where sk points to a inet_timewait_sock which
551 * has no sk_socket field (surely TCP_TIME_WAIT, perhaps more).
552 * To be safe, use a whitelist and not a blacklist.
553 * Before dereferencing sk_socket grab a lock on sk_callback_lock.
555 * Once again we need conditional code because the UID and GID
556 * location changes between kernels.
558 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,28)
559 /* use the current's real uid/gid */
560 #define _CURR_UID f_uid
561 #define _CURR_GID f_gid
562 #else /* 2.6.29 and above */
563 /* use the current's file access real uid/gid */
564 #define _CURR_UID f_cred->fsuid
565 #define _CURR_GID f_cred->fsgid
568 #define GOOD_STATES ( \
569 (1<<TCP_LISTEN) | (1<<TCP_SYN_RECV) | (1<<TCP_SYN_SENT) | \
570 (1<<TCP_ESTABLISHED) | (1<<TCP_FIN_WAIT1) | (1<<TCP_FIN_WAIT2) )
571 // surely exclude TCP_CLOSE, TCP_TIME_WAIT, TCP_LAST_ACK
572 // uncertain TCP_CLOSE_WAIT and TCP_CLOSING
574 if ((1<<st) & GOOD_STATES) {
575 read_lock_bh(&sk->sk_callback_lock);
576 if (sk->sk_socket && sk->sk_socket->file) {
577 u->uid = sk->sk_socket->file->_CURR_UID;
578 u->gid = sk->sk_socket->file->_CURR_GID;
580 read_unlock_bh(&sk->sk_callback_lock);
584 if (!skb->sk) /* return the reference that came from the lookup */
591 #endif /* LINUX > 2.4 */
595 * Now prepare to hook the various functions.
596 * Linux 2.4 has a different API so we need some adaptation
597 * for register and unregister hooks
599 * the unregister function changed arguments between 2.6.22 and 2.6.24
601 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
603 nf_register_hooks(struct nf_hook_ops *ops, int n)
606 for (i = 0; i < n; i++) {
607 ret = nf_register_hook(ops + i);
615 nf_unregister_hooks(struct nf_hook_ops *ops, int n)
618 for (i = 0; i < n; i++) {
619 nf_unregister_hook(ops + i);
622 #define REG_QH_ARG(fn) fn, NULL /* argument for nf_[un]register_queue_handler */
623 #define UNREG_QH_ARG(fn) //fn /* argument for nf_[un]register_queue_handler */
624 #define SET_MOD_OWNER
626 #else /* linux >= 2.6.0 */
628 struct nf_queue_handler ipfw2_queue_handler_desc = {
629 .outfn = ipfw2_queue_handler,
630 .name = "ipfw2 dummynet queue",
632 #define REG_QH_ARG(fn) &(fn ## _desc)
634 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
635 #define UNREG_QH_ARG(fn) //fn /* argument for nf_[un]register_queue_handler */
637 #define UNREG_QH_ARG(fn) , &(fn ## _desc)
638 #endif /* 2.6.0 < LINUX > 2.6.24 */
640 #define SET_MOD_OWNER .owner = THIS_MODULE,
642 #endif /* !LINUX < 2.6.0 */
644 static struct nf_hook_ops ipfw_ops[] __read_mostly = {
648 .hooknum = IPFW_HOOK_IN,
649 .priority = NF_IP_PRI_FILTER,
655 .hooknum = NF_IP_POST_ROUTING,
656 .priority = NF_IP_PRI_FILTER,
660 #endif /* !__linux__ */
662 /* descriptors for the children */
663 extern moduledata_t *moddesc_ipfw;
664 extern moduledata_t *moddesc_dummynet;
666 extern void rn_init(void);
668 * Module glue - init and exit function.
671 ipfw_module_init(void)
675 printf("%s in-hook %d svn id %s\n", __FUNCTION__, IPFW_HOOK_IN, "$Id$");
679 my_mod_register(moddesc_ipfw, "ipfw", 1);
680 my_mod_register(moddesc_dummynet, "dummynet", 2);
686 #else /* linux hook */
687 /* sockopt register, in order to talk with user space */
688 ret = nf_register_sockopt(&ipfw_sockopts);
690 printf("error %d in nf_register_sockopt\n", ret);
694 /* queue handler registration, in order to get network
695 * packet under a private queue */
696 ret = nf_register_queue_handler(PF_INET, REG_QH_ARG(ipfw2_queue_handler) );
697 if (ret < 0) /* queue busy */
698 goto unregister_sockopt;
700 ret = nf_register_hooks(ipfw_ops, ARRAY_SIZE(ipfw_ops));
702 goto unregister_sockopt;
704 printf("%s loaded\n", __FUNCTION__);
708 /* handle errors on load */
710 nf_unregister_queue_handler(PF_INET UNREG_QH_ARG(ipfw2_queue_handler) );
711 nf_unregister_sockopt(&ipfw_sockopts);
715 printf("%s error\n", __FUNCTION__);
721 /* module shutdown */
723 ipfw_module_exit(void)
726 #else /* linux hook */
727 nf_unregister_hooks(ipfw_ops, ARRAY_SIZE(ipfw_ops));
728 /* maybe drain the queue before unregistering ? */
729 nf_unregister_queue_handler(PF_INET UNREG_QH_ARG(ipfw2_queue_handler) );
730 nf_unregister_sockopt(&ipfw_sockopts);
735 printf("%s unloaded\n", __FUNCTION__);
739 module_init(ipfw_module_init)
740 module_exit(ipfw_module_exit)
741 MODULE_LICENSE("Dual BSD/GPL"); /* the code here is all BSD. */