Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* Cluster IP hashmark target |
2 | * (C) 2003-2004 by Harald Welte <laforge@netfilter.org> | |
3 | * based on ideas of Fabio Olive Leite <olive@unixforge.org> | |
4 | * | |
5 | * Development of this code funded by SuSE Linux AG, http://www.suse.com/ | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | * | |
11 | */ | |
12 | #include <linux/module.h> | |
1da177e4 LT |
13 | #include <linux/proc_fs.h> |
14 | #include <linux/jhash.h> | |
136e92bb | 15 | #include <linux/bitops.h> |
1da177e4 LT |
16 | #include <linux/skbuff.h> |
17 | #include <linux/ip.h> | |
18 | #include <linux/tcp.h> | |
19 | #include <linux/udp.h> | |
20 | #include <linux/icmp.h> | |
21 | #include <linux/if_arp.h> | |
22 | #include <linux/proc_fs.h> | |
23 | #include <linux/seq_file.h> | |
24 | ||
25 | #include <net/checksum.h> | |
26 | ||
27 | #include <linux/netfilter_arp.h> | |
28 | ||
29 | #include <linux/netfilter_ipv4/ip_tables.h> | |
30 | #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h> | |
9fb9cbb1 | 31 | #include <net/netfilter/nf_conntrack_compat.h> |
1da177e4 | 32 | |
136e92bb | 33 | #define CLUSTERIP_VERSION "0.8" |
1da177e4 LT |
34 | |
35 | #define DEBUG_CLUSTERIP | |
36 | ||
37 | #ifdef DEBUG_CLUSTERIP | |
38 | #define DEBUGP printk | |
39 | #else | |
40 | #define DEBUGP | |
41 | #endif | |
42 | ||
43 | MODULE_LICENSE("GPL"); | |
44 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); | |
45 | MODULE_DESCRIPTION("iptables target for CLUSTERIP"); | |
46 | ||
47 | struct clusterip_config { | |
48 | struct list_head list; /* list of all configs */ | |
49 | atomic_t refcount; /* reference count */ | |
44513624 KK |
50 | atomic_t entries; /* number of entries/rules |
51 | * referencing us */ | |
1da177e4 | 52 | |
6a19d614 | 53 | __be32 clusterip; /* the IP address */ |
1da177e4 LT |
54 | u_int8_t clustermac[ETH_ALEN]; /* the MAC address */ |
55 | struct net_device *dev; /* device */ | |
56 | u_int16_t num_total_nodes; /* total number of nodes */ | |
136e92bb | 57 | unsigned long local_nodes; /* node number array */ |
1da177e4 LT |
58 | |
59 | #ifdef CONFIG_PROC_FS | |
60 | struct proc_dir_entry *pde; /* proc dir entry */ | |
61 | #endif | |
62 | enum clusterip_hashmode hash_mode; /* which hashing mode */ | |
63 | u_int32_t hash_initval; /* hash initialization */ | |
64 | }; | |
65 | ||
66 | static LIST_HEAD(clusterip_configs); | |
67 | ||
136e92bb | 68 | /* clusterip_lock protects the clusterip_configs list */ |
e45b1be8 | 69 | static DEFINE_RWLOCK(clusterip_lock); |
1da177e4 LT |
70 | |
71 | #ifdef CONFIG_PROC_FS | |
72 | static struct file_operations clusterip_proc_fops; | |
73 | static struct proc_dir_entry *clusterip_procdir; | |
74 | #endif | |
75 | ||
76 | static inline void | |
44513624 KK |
77 | clusterip_config_get(struct clusterip_config *c) |
78 | { | |
1da177e4 LT |
79 | atomic_inc(&c->refcount); |
80 | } | |
81 | ||
82 | static inline void | |
44513624 KK |
83 | clusterip_config_put(struct clusterip_config *c) |
84 | { | |
85 | if (atomic_dec_and_test(&c->refcount)) | |
86 | kfree(c); | |
87 | } | |
88 | ||
89 | /* increase the count of entries(rules) using/referencing this config */ | |
90 | static inline void | |
91 | clusterip_config_entry_get(struct clusterip_config *c) | |
92 | { | |
93 | atomic_inc(&c->entries); | |
94 | } | |
95 | ||
96 | /* decrease the count of entries using/referencing this config. If last | |
97 | * entry(rule) is removed, remove the config from lists, but don't free it | |
98 | * yet, since proc-files could still be holding references */ | |
99 | static inline void | |
100 | clusterip_config_entry_put(struct clusterip_config *c) | |
101 | { | |
102 | if (atomic_dec_and_test(&c->entries)) { | |
e45b1be8 | 103 | write_lock_bh(&clusterip_lock); |
1da177e4 | 104 | list_del(&c->list); |
e45b1be8 | 105 | write_unlock_bh(&clusterip_lock); |
44513624 | 106 | |
1da177e4 LT |
107 | dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0); |
108 | dev_put(c->dev); | |
44513624 KK |
109 | |
110 | /* In case anyone still accesses the file, the open/close | |
111 | * functions are also incrementing the refcount on their own, | |
112 | * so it's safe to remove the entry even if it's in use. */ | |
113 | #ifdef CONFIG_PROC_FS | |
114 | remove_proc_entry(c->pde->name, c->pde->parent); | |
115 | #endif | |
1da177e4 LT |
116 | } |
117 | } | |
118 | ||
1da177e4 | 119 | static struct clusterip_config * |
6a19d614 | 120 | __clusterip_config_find(__be32 clusterip) |
1da177e4 LT |
121 | { |
122 | struct list_head *pos; | |
123 | ||
1da177e4 LT |
124 | list_for_each(pos, &clusterip_configs) { |
125 | struct clusterip_config *c = list_entry(pos, | |
126 | struct clusterip_config, list); | |
127 | if (c->clusterip == clusterip) { | |
128 | return c; | |
129 | } | |
130 | } | |
131 | ||
132 | return NULL; | |
133 | } | |
134 | ||
135 | static inline struct clusterip_config * | |
6a19d614 | 136 | clusterip_config_find_get(__be32 clusterip, int entry) |
1da177e4 LT |
137 | { |
138 | struct clusterip_config *c; | |
139 | ||
e45b1be8 | 140 | read_lock_bh(&clusterip_lock); |
1da177e4 LT |
141 | c = __clusterip_config_find(clusterip); |
142 | if (!c) { | |
e45b1be8 | 143 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
144 | return NULL; |
145 | } | |
146 | atomic_inc(&c->refcount); | |
44513624 KK |
147 | if (entry) |
148 | atomic_inc(&c->entries); | |
e45b1be8 | 149 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
150 | |
151 | return c; | |
152 | } | |
153 | ||
136e92bb KK |
154 | static void |
155 | clusterip_config_init_nodelist(struct clusterip_config *c, | |
156 | const struct ipt_clusterip_tgt_info *i) | |
157 | { | |
158 | int n; | |
159 | ||
160 | for (n = 0; n < i->num_local_nodes; n++) { | |
161 | set_bit(i->local_nodes[n] - 1, &c->local_nodes); | |
162 | } | |
163 | } | |
164 | ||
1da177e4 | 165 | static struct clusterip_config * |
6a19d614 | 166 | clusterip_config_init(struct ipt_clusterip_tgt_info *i, __be32 ip, |
1da177e4 LT |
167 | struct net_device *dev) |
168 | { | |
169 | struct clusterip_config *c; | |
1da177e4 | 170 | |
0da974f4 | 171 | c = kzalloc(sizeof(*c), GFP_ATOMIC); |
1da177e4 LT |
172 | if (!c) |
173 | return NULL; | |
174 | ||
1da177e4 LT |
175 | c->dev = dev; |
176 | c->clusterip = ip; | |
177 | memcpy(&c->clustermac, &i->clustermac, ETH_ALEN); | |
178 | c->num_total_nodes = i->num_total_nodes; | |
136e92bb | 179 | clusterip_config_init_nodelist(c, i); |
1da177e4 LT |
180 | c->hash_mode = i->hash_mode; |
181 | c->hash_initval = i->hash_initval; | |
182 | atomic_set(&c->refcount, 1); | |
44513624 | 183 | atomic_set(&c->entries, 1); |
1da177e4 LT |
184 | |
185 | #ifdef CONFIG_PROC_FS | |
76592584 PM |
186 | { |
187 | char buffer[16]; | |
188 | ||
189 | /* create proc dir entry */ | |
190 | sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip)); | |
191 | c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR, | |
192 | clusterip_procdir); | |
193 | if (!c->pde) { | |
194 | kfree(c); | |
195 | return NULL; | |
196 | } | |
1da177e4 LT |
197 | } |
198 | c->pde->proc_fops = &clusterip_proc_fops; | |
199 | c->pde->data = c; | |
200 | #endif | |
201 | ||
e45b1be8 | 202 | write_lock_bh(&clusterip_lock); |
1da177e4 | 203 | list_add(&c->list, &clusterip_configs); |
e45b1be8 | 204 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
205 | |
206 | return c; | |
207 | } | |
208 | ||
76592584 | 209 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
210 | static int |
211 | clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum) | |
212 | { | |
1da177e4 | 213 | |
136e92bb KK |
214 | if (nodenum == 0 || |
215 | nodenum > c->num_total_nodes) | |
1da177e4 | 216 | return 1; |
1da177e4 | 217 | |
136e92bb KK |
218 | /* check if we already have this number in our bitfield */ |
219 | if (test_and_set_bit(nodenum - 1, &c->local_nodes)) | |
220 | return 1; | |
1da177e4 | 221 | |
1da177e4 LT |
222 | return 0; |
223 | } | |
224 | ||
225 | static int | |
226 | clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum) | |
227 | { | |
136e92bb KK |
228 | if (nodenum == 0 || |
229 | nodenum > c->num_total_nodes) | |
1da177e4 | 230 | return 1; |
1da177e4 | 231 | |
136e92bb KK |
232 | if (test_and_clear_bit(nodenum - 1, &c->local_nodes)) |
233 | return 0; | |
1da177e4 | 234 | |
1da177e4 LT |
235 | return 1; |
236 | } | |
76592584 | 237 | #endif |
1da177e4 LT |
238 | |
239 | static inline u_int32_t | |
240 | clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config) | |
241 | { | |
242 | struct iphdr *iph = skb->nh.iph; | |
243 | unsigned long hashval; | |
244 | u_int16_t sport, dport; | |
957dc80a | 245 | u_int16_t *ports; |
1da177e4 LT |
246 | |
247 | switch (iph->protocol) { | |
248 | case IPPROTO_TCP: | |
1da177e4 | 249 | case IPPROTO_UDP: |
957dc80a PM |
250 | case IPPROTO_SCTP: |
251 | case IPPROTO_DCCP: | |
1da177e4 | 252 | case IPPROTO_ICMP: |
957dc80a PM |
253 | ports = (void *)iph+iph->ihl*4; |
254 | sport = ports[0]; | |
255 | dport = ports[1]; | |
1da177e4 LT |
256 | break; |
257 | default: | |
258 | if (net_ratelimit()) { | |
259 | printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n", | |
260 | iph->protocol); | |
261 | } | |
262 | sport = dport = 0; | |
263 | } | |
264 | ||
265 | switch (config->hash_mode) { | |
266 | case CLUSTERIP_HASHMODE_SIP: | |
267 | hashval = jhash_1word(ntohl(iph->saddr), | |
268 | config->hash_initval); | |
269 | break; | |
270 | case CLUSTERIP_HASHMODE_SIP_SPT: | |
271 | hashval = jhash_2words(ntohl(iph->saddr), sport, | |
272 | config->hash_initval); | |
273 | break; | |
274 | case CLUSTERIP_HASHMODE_SIP_SPT_DPT: | |
275 | hashval = jhash_3words(ntohl(iph->saddr), sport, dport, | |
276 | config->hash_initval); | |
277 | break; | |
278 | default: | |
279 | /* to make gcc happy */ | |
280 | hashval = 0; | |
281 | /* This cannot happen, unless the check function wasn't called | |
282 | * at rule load time */ | |
283 | printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode); | |
284 | BUG(); | |
285 | break; | |
286 | } | |
287 | ||
288 | /* node numbers are 1..n, not 0..n */ | |
289 | return ((hashval % config->num_total_nodes)+1); | |
290 | } | |
291 | ||
292 | static inline int | |
293 | clusterip_responsible(struct clusterip_config *config, u_int32_t hash) | |
294 | { | |
136e92bb | 295 | return test_bit(hash - 1, &config->local_nodes); |
1da177e4 LT |
296 | } |
297 | ||
298 | /*********************************************************************** | |
299 | * IPTABLES TARGET | |
300 | ***********************************************************************/ | |
301 | ||
302 | static unsigned int | |
303 | target(struct sk_buff **pskb, | |
304 | const struct net_device *in, | |
305 | const struct net_device *out, | |
306 | unsigned int hooknum, | |
c4986734 | 307 | const struct xt_target *target, |
fe1cb108 | 308 | const void *targinfo) |
1da177e4 LT |
309 | { |
310 | const struct ipt_clusterip_tgt_info *cipinfo = targinfo; | |
311 | enum ip_conntrack_info ctinfo; | |
9fb9cbb1 | 312 | u_int32_t *mark, hash; |
1da177e4 LT |
313 | |
314 | /* don't need to clusterip_config_get() here, since refcount | |
315 | * is only decremented by destroy() - and ip_tables guarantees | |
316 | * that the ->target() function isn't called after ->destroy() */ | |
317 | ||
9fb9cbb1 YK |
318 | mark = nf_ct_get_mark((*pskb), &ctinfo); |
319 | if (mark == NULL) { | |
1da177e4 LT |
320 | printk(KERN_ERR "CLUSTERIP: no conntrack!\n"); |
321 | /* FIXME: need to drop invalid ones, since replies | |
322 | * to outgoing connections of other nodes will be | |
323 | * marked as INVALID */ | |
324 | return NF_DROP; | |
325 | } | |
326 | ||
327 | /* special case: ICMP error handling. conntrack distinguishes between | |
328 | * error messages (RELATED) and information requests (see below) */ | |
329 | if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP | |
330 | && (ctinfo == IP_CT_RELATED | |
5d927eb0 | 331 | || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY)) |
1da177e4 LT |
332 | return IPT_CONTINUE; |
333 | ||
334 | /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO, | |
335 | * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here | |
336 | * on, which all have an ID field [relevant for hashing]. */ | |
337 | ||
338 | hash = clusterip_hashfn(*pskb, cipinfo->config); | |
339 | ||
340 | switch (ctinfo) { | |
341 | case IP_CT_NEW: | |
9fb9cbb1 | 342 | *mark = hash; |
1da177e4 LT |
343 | break; |
344 | case IP_CT_RELATED: | |
345 | case IP_CT_RELATED+IP_CT_IS_REPLY: | |
346 | /* FIXME: we don't handle expectations at the | |
347 | * moment. they can arrive on a different node than | |
348 | * the master connection (e.g. FTP passive mode) */ | |
349 | case IP_CT_ESTABLISHED: | |
350 | case IP_CT_ESTABLISHED+IP_CT_IS_REPLY: | |
351 | break; | |
352 | default: | |
353 | break; | |
354 | } | |
355 | ||
356 | #ifdef DEBUG_CLUSTERP | |
357 | DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
358 | #endif | |
9fb9cbb1 | 359 | DEBUGP("hash=%u ct_hash=%u ", hash, *mark); |
1da177e4 LT |
360 | if (!clusterip_responsible(cipinfo->config, hash)) { |
361 | DEBUGP("not responsible\n"); | |
362 | return NF_DROP; | |
363 | } | |
364 | DEBUGP("responsible\n"); | |
365 | ||
366 | /* despite being received via linklayer multicast, this is | |
367 | * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */ | |
368 | (*pskb)->pkt_type = PACKET_HOST; | |
369 | ||
370 | return IPT_CONTINUE; | |
371 | } | |
372 | ||
373 | static int | |
374 | checkentry(const char *tablename, | |
2e4e6a17 | 375 | const void *e_void, |
c4986734 | 376 | const struct xt_target *target, |
1da177e4 | 377 | void *targinfo, |
1da177e4 LT |
378 | unsigned int hook_mask) |
379 | { | |
380 | struct ipt_clusterip_tgt_info *cipinfo = targinfo; | |
2e4e6a17 | 381 | const struct ipt_entry *e = e_void; |
1da177e4 LT |
382 | |
383 | struct clusterip_config *config; | |
384 | ||
1da177e4 LT |
385 | if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP && |
386 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT && | |
387 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) { | |
388 | printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n", | |
389 | cipinfo->hash_mode); | |
390 | return 0; | |
391 | ||
392 | } | |
6a19d614 | 393 | if (e->ip.dmsk.s_addr != htonl(0xffffffff) |
1da177e4 LT |
394 | || e->ip.dst.s_addr == 0) { |
395 | printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n"); | |
396 | return 0; | |
397 | } | |
398 | ||
399 | /* FIXME: further sanity checks */ | |
400 | ||
44513624 KK |
401 | config = clusterip_config_find_get(e->ip.dst.s_addr, 1); |
402 | if (config) { | |
403 | if (cipinfo->config != NULL) { | |
404 | /* Case A: This is an entry that gets reloaded, since | |
405 | * it still has a cipinfo->config pointer. Simply | |
406 | * increase the entry refcount and return */ | |
407 | if (cipinfo->config != config) { | |
408 | printk(KERN_ERR "CLUSTERIP: Reloaded entry " | |
409 | "has invalid config pointer!\n"); | |
410 | return 0; | |
411 | } | |
412 | clusterip_config_entry_get(cipinfo->config); | |
413 | } else { | |
414 | /* Case B: This is a new rule referring to an existing | |
415 | * clusterip config. */ | |
416 | cipinfo->config = config; | |
417 | clusterip_config_entry_get(cipinfo->config); | |
418 | } | |
419 | } else { | |
420 | /* Case C: This is a completely new clusterip config */ | |
1da177e4 LT |
421 | if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) { |
422 | printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr)); | |
423 | return 0; | |
424 | } else { | |
425 | struct net_device *dev; | |
426 | ||
427 | if (e->ip.iniface[0] == '\0') { | |
428 | printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n"); | |
429 | return 0; | |
430 | } | |
431 | ||
432 | dev = dev_get_by_name(e->ip.iniface); | |
433 | if (!dev) { | |
434 | printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface); | |
435 | return 0; | |
436 | } | |
437 | ||
438 | config = clusterip_config_init(cipinfo, | |
439 | e->ip.dst.s_addr, dev); | |
440 | if (!config) { | |
441 | printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n"); | |
442 | dev_put(dev); | |
443 | return 0; | |
444 | } | |
445 | dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0); | |
446 | } | |
44513624 | 447 | cipinfo->config = config; |
1da177e4 LT |
448 | } |
449 | ||
11078c37 YK |
450 | if (nf_ct_l3proto_try_module_get(target->family) < 0) { |
451 | printk(KERN_WARNING "can't load conntrack support for " | |
452 | "proto=%d\n", target->family); | |
453 | return 0; | |
454 | } | |
455 | ||
1da177e4 LT |
456 | return 1; |
457 | } | |
458 | ||
459 | /* drop reference count of cluster config when rule is deleted */ | |
efa74165 | 460 | static void destroy(const struct xt_target *target, void *targinfo) |
1da177e4 | 461 | { |
c4986734 | 462 | struct ipt_clusterip_tgt_info *cipinfo = targinfo; |
1da177e4 | 463 | |
44513624 KK |
464 | /* if no more entries are referencing the config, remove it |
465 | * from the list and destroy the proc entry */ | |
466 | clusterip_config_entry_put(cipinfo->config); | |
467 | ||
1da177e4 | 468 | clusterip_config_put(cipinfo->config); |
11078c37 YK |
469 | |
470 | nf_ct_l3proto_module_put(target->family); | |
1da177e4 LT |
471 | } |
472 | ||
1d5cd909 PM |
473 | static struct ipt_target clusterip_tgt = { |
474 | .name = "CLUSTERIP", | |
475 | .target = target, | |
476 | .targetsize = sizeof(struct ipt_clusterip_tgt_info), | |
477 | .checkentry = checkentry, | |
478 | .destroy = destroy, | |
479 | .me = THIS_MODULE | |
1da177e4 LT |
480 | }; |
481 | ||
482 | ||
483 | /*********************************************************************** | |
484 | * ARP MANGLING CODE | |
485 | ***********************************************************************/ | |
486 | ||
487 | /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */ | |
488 | struct arp_payload { | |
489 | u_int8_t src_hw[ETH_ALEN]; | |
6a19d614 | 490 | __be32 src_ip; |
1da177e4 | 491 | u_int8_t dst_hw[ETH_ALEN]; |
6a19d614 | 492 | __be32 dst_ip; |
1da177e4 LT |
493 | } __attribute__ ((packed)); |
494 | ||
495 | #ifdef CLUSTERIP_DEBUG | |
496 | static void arp_print(struct arp_payload *payload) | |
497 | { | |
498 | #define HBUFFERLEN 30 | |
499 | char hbuffer[HBUFFERLEN]; | |
500 | int j,k; | |
501 | const char hexbuf[]= "0123456789abcdef"; | |
502 | ||
503 | for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) { | |
504 | hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15]; | |
505 | hbuffer[k++]=hexbuf[payload->src_hw[j]&15]; | |
506 | hbuffer[k++]=':'; | |
507 | } | |
508 | hbuffer[--k]='\0'; | |
509 | ||
510 | printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n", | |
511 | NIPQUAD(payload->src_ip), hbuffer, | |
512 | NIPQUAD(payload->dst_ip)); | |
513 | } | |
514 | #endif | |
515 | ||
516 | static unsigned int | |
517 | arp_mangle(unsigned int hook, | |
518 | struct sk_buff **pskb, | |
519 | const struct net_device *in, | |
520 | const struct net_device *out, | |
521 | int (*okfn)(struct sk_buff *)) | |
522 | { | |
523 | struct arphdr *arp = (*pskb)->nh.arph; | |
524 | struct arp_payload *payload; | |
525 | struct clusterip_config *c; | |
526 | ||
527 | /* we don't care about non-ethernet and non-ipv4 ARP */ | |
528 | if (arp->ar_hrd != htons(ARPHRD_ETHER) | |
529 | || arp->ar_pro != htons(ETH_P_IP) | |
530 | || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN) | |
531 | return NF_ACCEPT; | |
532 | ||
4095ebf1 HW |
533 | /* we only want to mangle arp requests and replies */ |
534 | if (arp->ar_op != htons(ARPOP_REPLY) | |
535 | && arp->ar_op != htons(ARPOP_REQUEST)) | |
1da177e4 LT |
536 | return NF_ACCEPT; |
537 | ||
538 | payload = (void *)(arp+1); | |
539 | ||
540 | /* if there is no clusterip configuration for the arp reply's | |
541 | * source ip, we don't want to mangle it */ | |
44513624 | 542 | c = clusterip_config_find_get(payload->src_ip, 0); |
1da177e4 LT |
543 | if (!c) |
544 | return NF_ACCEPT; | |
545 | ||
546 | /* normally the linux kernel always replies to arp queries of | |
547 | * addresses on different interfacs. However, in the CLUSTERIP case | |
548 | * this wouldn't work, since we didn't subscribe the mcast group on | |
549 | * other interfaces */ | |
550 | if (c->dev != out) { | |
551 | DEBUGP("CLUSTERIP: not mangling arp reply on different " | |
552 | "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name); | |
553 | clusterip_config_put(c); | |
554 | return NF_ACCEPT; | |
555 | } | |
556 | ||
557 | /* mangle reply hardware address */ | |
558 | memcpy(payload->src_hw, c->clustermac, arp->ar_hln); | |
559 | ||
560 | #ifdef CLUSTERIP_DEBUG | |
561 | DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: "); | |
562 | arp_print(payload); | |
563 | #endif | |
564 | ||
565 | clusterip_config_put(c); | |
566 | ||
567 | return NF_ACCEPT; | |
568 | } | |
569 | ||
570 | static struct nf_hook_ops cip_arp_ops = { | |
571 | .hook = arp_mangle, | |
572 | .pf = NF_ARP, | |
573 | .hooknum = NF_ARP_OUT, | |
574 | .priority = -1 | |
575 | }; | |
576 | ||
577 | /*********************************************************************** | |
578 | * PROC DIR HANDLING | |
579 | ***********************************************************************/ | |
580 | ||
581 | #ifdef CONFIG_PROC_FS | |
582 | ||
136e92bb KK |
583 | struct clusterip_seq_position { |
584 | unsigned int pos; /* position */ | |
585 | unsigned int weight; /* number of bits set == size */ | |
586 | unsigned int bit; /* current bit */ | |
587 | unsigned long val; /* current value */ | |
588 | }; | |
589 | ||
1da177e4 LT |
590 | static void *clusterip_seq_start(struct seq_file *s, loff_t *pos) |
591 | { | |
592 | struct proc_dir_entry *pde = s->private; | |
593 | struct clusterip_config *c = pde->data; | |
136e92bb KK |
594 | unsigned int weight; |
595 | u_int32_t local_nodes; | |
596 | struct clusterip_seq_position *idx; | |
597 | ||
598 | /* FIXME: possible race */ | |
599 | local_nodes = c->local_nodes; | |
600 | weight = hweight32(local_nodes); | |
601 | if (*pos >= weight) | |
1da177e4 LT |
602 | return NULL; |
603 | ||
136e92bb KK |
604 | idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL); |
605 | if (!idx) | |
1da177e4 LT |
606 | return ERR_PTR(-ENOMEM); |
607 | ||
136e92bb KK |
608 | idx->pos = *pos; |
609 | idx->weight = weight; | |
610 | idx->bit = ffs(local_nodes); | |
611 | idx->val = local_nodes; | |
612 | clear_bit(idx->bit - 1, &idx->val); | |
613 | ||
614 | return idx; | |
1da177e4 LT |
615 | } |
616 | ||
617 | static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos) | |
618 | { | |
136e92bb | 619 | struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v; |
1da177e4 | 620 | |
136e92bb KK |
621 | *pos = ++idx->pos; |
622 | if (*pos >= idx->weight) { | |
1da177e4 LT |
623 | kfree(v); |
624 | return NULL; | |
625 | } | |
136e92bb KK |
626 | idx->bit = ffs(idx->val); |
627 | clear_bit(idx->bit - 1, &idx->val); | |
628 | return idx; | |
1da177e4 LT |
629 | } |
630 | ||
631 | static void clusterip_seq_stop(struct seq_file *s, void *v) | |
632 | { | |
633 | kfree(v); | |
1da177e4 LT |
634 | } |
635 | ||
636 | static int clusterip_seq_show(struct seq_file *s, void *v) | |
637 | { | |
136e92bb | 638 | struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v; |
1da177e4 | 639 | |
136e92bb | 640 | if (idx->pos != 0) |
1da177e4 | 641 | seq_putc(s, ','); |
1da177e4 | 642 | |
136e92bb KK |
643 | seq_printf(s, "%u", idx->bit); |
644 | ||
645 | if (idx->pos == idx->weight - 1) | |
1da177e4 LT |
646 | seq_putc(s, '\n'); |
647 | ||
648 | return 0; | |
649 | } | |
650 | ||
651 | static struct seq_operations clusterip_seq_ops = { | |
652 | .start = clusterip_seq_start, | |
653 | .next = clusterip_seq_next, | |
654 | .stop = clusterip_seq_stop, | |
655 | .show = clusterip_seq_show, | |
656 | }; | |
657 | ||
658 | static int clusterip_proc_open(struct inode *inode, struct file *file) | |
659 | { | |
660 | int ret = seq_open(file, &clusterip_seq_ops); | |
661 | ||
662 | if (!ret) { | |
663 | struct seq_file *sf = file->private_data; | |
664 | struct proc_dir_entry *pde = PDE(inode); | |
665 | struct clusterip_config *c = pde->data; | |
666 | ||
667 | sf->private = pde; | |
668 | ||
669 | clusterip_config_get(c); | |
670 | } | |
671 | ||
672 | return ret; | |
673 | } | |
674 | ||
675 | static int clusterip_proc_release(struct inode *inode, struct file *file) | |
676 | { | |
677 | struct proc_dir_entry *pde = PDE(inode); | |
678 | struct clusterip_config *c = pde->data; | |
679 | int ret; | |
680 | ||
681 | ret = seq_release(inode, file); | |
682 | ||
683 | if (!ret) | |
684 | clusterip_config_put(c); | |
685 | ||
686 | return ret; | |
687 | } | |
688 | ||
689 | static ssize_t clusterip_proc_write(struct file *file, const char __user *input, | |
690 | size_t size, loff_t *ofs) | |
691 | { | |
692 | #define PROC_WRITELEN 10 | |
693 | char buffer[PROC_WRITELEN+1]; | |
6df81ab2 | 694 | struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode); |
1da177e4 LT |
695 | struct clusterip_config *c = pde->data; |
696 | unsigned long nodenum; | |
697 | ||
698 | if (copy_from_user(buffer, input, PROC_WRITELEN)) | |
699 | return -EFAULT; | |
700 | ||
701 | if (*buffer == '+') { | |
702 | nodenum = simple_strtoul(buffer+1, NULL, 10); | |
703 | if (clusterip_add_node(c, nodenum)) | |
704 | return -ENOMEM; | |
705 | } else if (*buffer == '-') { | |
706 | nodenum = simple_strtoul(buffer+1, NULL,10); | |
707 | if (clusterip_del_node(c, nodenum)) | |
708 | return -ENOENT; | |
709 | } else | |
710 | return -EIO; | |
711 | ||
712 | return size; | |
713 | } | |
714 | ||
715 | static struct file_operations clusterip_proc_fops = { | |
716 | .owner = THIS_MODULE, | |
717 | .open = clusterip_proc_open, | |
718 | .read = seq_read, | |
719 | .write = clusterip_proc_write, | |
720 | .llseek = seq_lseek, | |
721 | .release = clusterip_proc_release, | |
722 | }; | |
723 | ||
724 | #endif /* CONFIG_PROC_FS */ | |
725 | ||
32292a7f | 726 | static int __init ipt_clusterip_init(void) |
1da177e4 LT |
727 | { |
728 | int ret; | |
729 | ||
32292a7f PM |
730 | ret = ipt_register_target(&clusterip_tgt); |
731 | if (ret < 0) | |
732 | return ret; | |
1da177e4 | 733 | |
32292a7f PM |
734 | ret = nf_register_hook(&cip_arp_ops); |
735 | if (ret < 0) | |
1da177e4 | 736 | goto cleanup_target; |
1da177e4 LT |
737 | |
738 | #ifdef CONFIG_PROC_FS | |
739 | clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net); | |
740 | if (!clusterip_procdir) { | |
741 | printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n"); | |
742 | ret = -ENOMEM; | |
743 | goto cleanup_hook; | |
744 | } | |
745 | #endif /* CONFIG_PROC_FS */ | |
746 | ||
747 | printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n", | |
748 | CLUSTERIP_VERSION); | |
1da177e4 LT |
749 | return 0; |
750 | ||
76592584 | 751 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
752 | cleanup_hook: |
753 | nf_unregister_hook(&cip_arp_ops); | |
76592584 | 754 | #endif /* CONFIG_PROC_FS */ |
1da177e4 LT |
755 | cleanup_target: |
756 | ipt_unregister_target(&clusterip_tgt); | |
32292a7f | 757 | return ret; |
1da177e4 LT |
758 | } |
759 | ||
65b4b4e8 | 760 | static void __exit ipt_clusterip_fini(void) |
1da177e4 | 761 | { |
32292a7f PM |
762 | printk(KERN_NOTICE "ClusterIP Version %s unloading\n", |
763 | CLUSTERIP_VERSION); | |
764 | #ifdef CONFIG_PROC_FS | |
765 | remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent); | |
766 | #endif | |
767 | nf_unregister_hook(&cip_arp_ops); | |
768 | ipt_unregister_target(&clusterip_tgt); | |
1da177e4 LT |
769 | } |
770 | ||
65b4b4e8 AM |
771 | module_init(ipt_clusterip_init); |
772 | module_exit(ipt_clusterip_fini); |