vserver 1.9.3
[linux-2.6.git] / net / ipv4 / fib_rules.c
1 /*
2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
3  *              operating system.  INET is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              IPv4 Forwarding Information Base: policy rules.
7  *
8  * Version:     $Id: fib_rules.c,v 1.17 2001/10/31 21:55:54 davem Exp $
9  *
10  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
11  *
12  *              This program is free software; you can redistribute it and/or
13  *              modify it under the terms of the GNU General Public License
14  *              as published by the Free Software Foundation; either version
15  *              2 of the License, or (at your option) any later version.
16  *
17  * Fixes:
18  *              Rani Assaf      :       local_rule cannot be deleted
19  *              Marc Boucher    :       routing by fwmark
20  */
21
22 #include <linux/config.h>
23 #include <asm/uaccess.h>
24 #include <asm/system.h>
25 #include <asm/bitops.h>
26 #include <linux/types.h>
27 #include <linux/kernel.h>
28 #include <linux/sched.h>
29 #include <linux/mm.h>
30 #include <linux/string.h>
31 #include <linux/socket.h>
32 #include <linux/sockios.h>
33 #include <linux/errno.h>
34 #include <linux/in.h>
35 #include <linux/inet.h>
36 #include <linux/netdevice.h>
37 #include <linux/if_arp.h>
38 #include <linux/proc_fs.h>
39 #include <linux/skbuff.h>
40 #include <linux/netlink.h>
41 #include <linux/init.h>
42
43 #include <net/ip.h>
44 #include <net/protocol.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47 #include <net/sock.h>
48 #include <net/ip_fib.h>
49
50 #define FRprintk(a...)
51
52 struct fib_rule
53 {
54         struct fib_rule *r_next;
55         atomic_t        r_clntref;
56         u32             r_preference;
57         unsigned char   r_table;
58         unsigned char   r_action;
59         unsigned char   r_dst_len;
60         unsigned char   r_src_len;
61         u32             r_src;
62         u32             r_srcmask;
63         u32             r_dst;
64         u32             r_dstmask;
65         u32             r_srcmap;
66         u8              r_flags;
67         u8              r_tos;
68 #ifdef CONFIG_IP_ROUTE_FWMARK
69         u32             r_fwmark;
70 #endif
71         int             r_ifindex;
72 #ifdef CONFIG_NET_CLS_ROUTE
73         __u32           r_tclassid;
74 #endif
75         char            r_ifname[IFNAMSIZ];
76         int             r_dead;
77 };
78
79 static struct fib_rule default_rule = {
80         .r_clntref =    ATOMIC_INIT(2),
81         .r_preference = 0x7FFF,
82         .r_table =      RT_TABLE_DEFAULT,
83         .r_action =     RTN_UNICAST,
84 };
85
86 static struct fib_rule main_rule = {
87         .r_next =       &default_rule,
88         .r_clntref =    ATOMIC_INIT(2),
89         .r_preference = 0x7FFE,
90         .r_table =      RT_TABLE_MAIN,
91         .r_action =     RTN_UNICAST,
92 };
93
94 static struct fib_rule local_rule = {
95         .r_next =       &main_rule,
96         .r_clntref =    ATOMIC_INIT(2),
97         .r_table =      RT_TABLE_LOCAL,
98         .r_action =     RTN_UNICAST,
99 };
100
101 static struct fib_rule *fib_rules = &local_rule;
102 static rwlock_t fib_rules_lock = RW_LOCK_UNLOCKED;
103
104 int inet_rtm_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
105 {
106         struct rtattr **rta = arg;
107         struct rtmsg *rtm = NLMSG_DATA(nlh);
108         struct fib_rule *r, **rp;
109         int err = -ESRCH;
110
111         for (rp=&fib_rules; (r=*rp) != NULL; rp=&r->r_next) {
112                 if ((!rta[RTA_SRC-1] || memcmp(RTA_DATA(rta[RTA_SRC-1]), &r->r_src, 4) == 0) &&
113                     rtm->rtm_src_len == r->r_src_len &&
114                     rtm->rtm_dst_len == r->r_dst_len &&
115                     (!rta[RTA_DST-1] || memcmp(RTA_DATA(rta[RTA_DST-1]), &r->r_dst, 4) == 0) &&
116                     rtm->rtm_tos == r->r_tos &&
117 #ifdef CONFIG_IP_ROUTE_FWMARK
118                     (!rta[RTA_PROTOINFO-1] || memcmp(RTA_DATA(rta[RTA_PROTOINFO-1]), &r->r_fwmark, 4) == 0) &&
119 #endif
120                     (!rtm->rtm_type || rtm->rtm_type == r->r_action) &&
121                     (!rta[RTA_PRIORITY-1] || memcmp(RTA_DATA(rta[RTA_PRIORITY-1]), &r->r_preference, 4) == 0) &&
122                     (!rta[RTA_IIF-1] || strcmp(RTA_DATA(rta[RTA_IIF-1]), r->r_ifname) == 0) &&
123                     (!rtm->rtm_table || (r && rtm->rtm_table == r->r_table))) {
124                         err = -EPERM;
125                         if (r == &local_rule)
126                                 break;
127
128                         write_lock_bh(&fib_rules_lock);
129                         *rp = r->r_next;
130                         r->r_dead = 1;
131                         write_unlock_bh(&fib_rules_lock);
132                         fib_rule_put(r);
133                         err = 0;
134                         break;
135                 }
136         }
137         return err;
138 }
139
140 /* Allocate new unique table id */
141
142 static struct fib_table *fib_empty_table(void)
143 {
144         int id;
145
146         for (id = 1; id <= RT_TABLE_MAX; id++)
147                 if (fib_tables[id] == NULL)
148                         return __fib_new_table(id);
149         return NULL;
150 }
151
152 void fib_rule_put(struct fib_rule *r)
153 {
154         if (atomic_dec_and_test(&r->r_clntref)) {
155                 if (r->r_dead)
156                         kfree(r);
157                 else
158                         printk("Freeing alive rule %p\n", r);
159         }
160 }
161
162 int inet_rtm_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
163 {
164         struct rtattr **rta = arg;
165         struct rtmsg *rtm = NLMSG_DATA(nlh);
166         struct fib_rule *r, *new_r, **rp;
167         unsigned char table_id;
168
169         if (rtm->rtm_src_len > 32 || rtm->rtm_dst_len > 32 ||
170             (rtm->rtm_tos & ~IPTOS_TOS_MASK))
171                 return -EINVAL;
172
173         if (rta[RTA_IIF-1] && RTA_PAYLOAD(rta[RTA_IIF-1]) > IFNAMSIZ)
174                 return -EINVAL;
175
176         table_id = rtm->rtm_table;
177         if (table_id == RT_TABLE_UNSPEC) {
178                 struct fib_table *table;
179                 if (rtm->rtm_type == RTN_UNICAST) {
180                         if ((table = fib_empty_table()) == NULL)
181                                 return -ENOBUFS;
182                         table_id = table->tb_id;
183                 }
184         }
185
186         new_r = kmalloc(sizeof(*new_r), GFP_KERNEL);
187         if (!new_r)
188                 return -ENOMEM;
189         memset(new_r, 0, sizeof(*new_r));
190         if (rta[RTA_SRC-1])
191                 memcpy(&new_r->r_src, RTA_DATA(rta[RTA_SRC-1]), 4);
192         if (rta[RTA_DST-1])
193                 memcpy(&new_r->r_dst, RTA_DATA(rta[RTA_DST-1]), 4);
194         if (rta[RTA_GATEWAY-1])
195                 memcpy(&new_r->r_srcmap, RTA_DATA(rta[RTA_GATEWAY-1]), 4);
196         new_r->r_src_len = rtm->rtm_src_len;
197         new_r->r_dst_len = rtm->rtm_dst_len;
198         new_r->r_srcmask = inet_make_mask(rtm->rtm_src_len);
199         new_r->r_dstmask = inet_make_mask(rtm->rtm_dst_len);
200         new_r->r_tos = rtm->rtm_tos;
201 #ifdef CONFIG_IP_ROUTE_FWMARK
202         if (rta[RTA_PROTOINFO-1])
203                 memcpy(&new_r->r_fwmark, RTA_DATA(rta[RTA_PROTOINFO-1]), 4);
204 #endif
205         new_r->r_action = rtm->rtm_type;
206         new_r->r_flags = rtm->rtm_flags;
207         if (rta[RTA_PRIORITY-1])
208                 memcpy(&new_r->r_preference, RTA_DATA(rta[RTA_PRIORITY-1]), 4);
209         new_r->r_table = table_id;
210         if (rta[RTA_IIF-1]) {
211                 struct net_device *dev;
212                 memcpy(new_r->r_ifname, RTA_DATA(rta[RTA_IIF-1]), IFNAMSIZ);
213                 new_r->r_ifname[IFNAMSIZ-1] = 0;
214                 new_r->r_ifindex = -1;
215                 dev = __dev_get_by_name(new_r->r_ifname);
216                 if (dev)
217                         new_r->r_ifindex = dev->ifindex;
218         }
219 #ifdef CONFIG_NET_CLS_ROUTE
220         if (rta[RTA_FLOW-1])
221                 memcpy(&new_r->r_tclassid, RTA_DATA(rta[RTA_FLOW-1]), 4);
222 #endif
223
224         rp = &fib_rules;
225         if (!new_r->r_preference) {
226                 r = fib_rules;
227                 if (r && (r = r->r_next) != NULL) {
228                         rp = &fib_rules->r_next;
229                         if (r->r_preference)
230                                 new_r->r_preference = r->r_preference - 1;
231                 }
232         }
233
234         while ( (r = *rp) != NULL ) {
235                 if (r->r_preference > new_r->r_preference)
236                         break;
237                 rp = &r->r_next;
238         }
239
240         new_r->r_next = r;
241         atomic_inc(&new_r->r_clntref);
242         write_lock_bh(&fib_rules_lock);
243         *rp = new_r;
244         write_unlock_bh(&fib_rules_lock);
245         return 0;
246 }
247
248 u32 fib_rules_map_destination(u32 daddr, struct fib_result *res)
249 {
250         u32 mask = inet_make_mask(res->prefixlen);
251         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
252 }
253
254 #ifdef CONFIG_NET_CLS_ROUTE
255 u32 fib_rules_tclass(struct fib_result *res)
256 {
257         if (res->r)
258                 return res->r->r_tclassid;
259         return 0;
260 }
261 #endif
262
263
264 static void fib_rules_detach(struct net_device *dev)
265 {
266         struct fib_rule *r;
267
268         for (r=fib_rules; r; r=r->r_next) {
269                 if (r->r_ifindex == dev->ifindex) {
270                         write_lock_bh(&fib_rules_lock);
271                         r->r_ifindex = -1;
272                         write_unlock_bh(&fib_rules_lock);
273                 }
274         }
275 }
276
277 static void fib_rules_attach(struct net_device *dev)
278 {
279         struct fib_rule *r;
280
281         for (r=fib_rules; r; r=r->r_next) {
282                 if (r->r_ifindex == -1 && strcmp(dev->name, r->r_ifname) == 0) {
283                         write_lock_bh(&fib_rules_lock);
284                         r->r_ifindex = dev->ifindex;
285                         write_unlock_bh(&fib_rules_lock);
286                 }
287         }
288 }
289
290 int fib_lookup(const struct flowi *flp, struct fib_result *res)
291 {
292         int err;
293         struct fib_rule *r, *policy;
294         struct fib_table *tb;
295
296         u32 daddr = flp->fl4_dst;
297         u32 saddr = flp->fl4_src;
298
299 FRprintk("Lookup: %u.%u.%u.%u <- %u.%u.%u.%u ",
300         NIPQUAD(flp->fl4_dst), NIPQUAD(flp->fl4_src));
301         read_lock(&fib_rules_lock);
302         for (r = fib_rules; r; r=r->r_next) {
303                 if (((saddr^r->r_src) & r->r_srcmask) ||
304                     ((daddr^r->r_dst) & r->r_dstmask) ||
305                     (r->r_tos && r->r_tos != flp->fl4_tos) ||
306 #ifdef CONFIG_IP_ROUTE_FWMARK
307                     (r->r_fwmark && r->r_fwmark != flp->fl4_fwmark) ||
308 #endif
309                     (r->r_ifindex && r->r_ifindex != flp->iif))
310                         continue;
311
312 FRprintk("tb %d r %d ", r->r_table, r->r_action);
313                 switch (r->r_action) {
314                 case RTN_UNICAST:
315                         policy = r;
316                         break;
317                 case RTN_UNREACHABLE:
318                         read_unlock(&fib_rules_lock);
319                         return -ENETUNREACH;
320                 default:
321                 case RTN_BLACKHOLE:
322                         read_unlock(&fib_rules_lock);
323                         return -EINVAL;
324                 case RTN_PROHIBIT:
325                         read_unlock(&fib_rules_lock);
326                         return -EACCES;
327                 }
328
329                 if ((tb = fib_get_table(r->r_table)) == NULL)
330                         continue;
331                 err = tb->tb_lookup(tb, flp, res);
332                 if (err == 0) {
333                         res->r = policy;
334                         if (policy)
335                                 atomic_inc(&policy->r_clntref);
336                         read_unlock(&fib_rules_lock);
337                         return 0;
338                 }
339                 if (err < 0 && err != -EAGAIN) {
340                         read_unlock(&fib_rules_lock);
341                         return err;
342                 }
343         }
344 FRprintk("FAILURE\n");
345         read_unlock(&fib_rules_lock);
346         return -ENETUNREACH;
347 }
348
349 void fib_select_default(const struct flowi *flp, struct fib_result *res)
350 {
351         if (res->r && res->r->r_action == RTN_UNICAST &&
352             FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) {
353                 struct fib_table *tb;
354                 if ((tb = fib_get_table(res->r->r_table)) != NULL)
355                         tb->tb_select_default(tb, flp, res);
356         }
357 }
358
359 static int fib_rules_event(struct notifier_block *this, unsigned long event, void *ptr)
360 {
361         struct net_device *dev = ptr;
362
363         if (event == NETDEV_UNREGISTER)
364                 fib_rules_detach(dev);
365         else if (event == NETDEV_REGISTER)
366                 fib_rules_attach(dev);
367         return NOTIFY_DONE;
368 }
369
370
371 struct notifier_block fib_rules_notifier = {
372         .notifier_call =fib_rules_event,
373 };
374
375 static __inline__ int inet_fill_rule(struct sk_buff *skb,
376                                      struct fib_rule *r,
377                                      struct netlink_callback *cb)
378 {
379         struct rtmsg *rtm;
380         struct nlmsghdr  *nlh;
381         unsigned char    *b = skb->tail;
382
383         nlh = NLMSG_PUT(skb, NETLINK_CREDS(cb->skb)->pid, cb->nlh->nlmsg_seq, RTM_NEWRULE, sizeof(*rtm));
384         rtm = NLMSG_DATA(nlh);
385         rtm->rtm_family = AF_INET;
386         rtm->rtm_dst_len = r->r_dst_len;
387         rtm->rtm_src_len = r->r_src_len;
388         rtm->rtm_tos = r->r_tos;
389 #ifdef CONFIG_IP_ROUTE_FWMARK
390         if (r->r_fwmark)
391                 RTA_PUT(skb, RTA_PROTOINFO, 4, &r->r_fwmark);
392 #endif
393         rtm->rtm_table = r->r_table;
394         rtm->rtm_protocol = 0;
395         rtm->rtm_scope = 0;
396         rtm->rtm_type = r->r_action;
397         rtm->rtm_flags = r->r_flags;
398
399         if (r->r_dst_len)
400                 RTA_PUT(skb, RTA_DST, 4, &r->r_dst);
401         if (r->r_src_len)
402                 RTA_PUT(skb, RTA_SRC, 4, &r->r_src);
403         if (r->r_ifname[0])
404                 RTA_PUT(skb, RTA_IIF, IFNAMSIZ, &r->r_ifname);
405         if (r->r_preference)
406                 RTA_PUT(skb, RTA_PRIORITY, 4, &r->r_preference);
407         if (r->r_srcmap)
408                 RTA_PUT(skb, RTA_GATEWAY, 4, &r->r_srcmap);
409 #ifdef CONFIG_NET_CLS_ROUTE
410         if (r->r_tclassid)
411                 RTA_PUT(skb, RTA_FLOW, 4, &r->r_tclassid);
412 #endif
413         nlh->nlmsg_len = skb->tail - b;
414         return skb->len;
415
416 nlmsg_failure:
417 rtattr_failure:
418         skb_trim(skb, b - skb->data);
419         return -1;
420 }
421
422 int inet_dump_rules(struct sk_buff *skb, struct netlink_callback *cb)
423 {
424         int idx;
425         int s_idx = cb->args[0];
426         struct fib_rule *r;
427
428         read_lock(&fib_rules_lock);
429         for (r=fib_rules, idx=0; r; r = r->r_next, idx++) {
430                 if (idx < s_idx)
431                         continue;
432                 if (inet_fill_rule(skb, r, cb) < 0)
433                         break;
434         }
435         read_unlock(&fib_rules_lock);
436         cb->args[0] = idx;
437
438         return skb->len;
439 }
440
441 void __init fib_rules_init(void)
442 {
443         register_netdevice_notifier(&fib_rules_notifier);
444 }