Commit | Line | Data |
---|---|---|
e905a9ed | 1 | /* Cluster IP hashmark target |
1da177e4 LT |
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> | |
1da177e4 | 22 | #include <linux/seq_file.h> |
1da177e4 | 23 | #include <linux/netfilter_arp.h> |
6709dbbb | 24 | #include <linux/netfilter/x_tables.h> |
1da177e4 LT |
25 | #include <linux/netfilter_ipv4/ip_tables.h> |
26 | #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h> | |
587aa641 | 27 | #include <net/netfilter/nf_conntrack.h> |
457c4cbc | 28 | #include <net/net_namespace.h> |
587aa641 | 29 | #include <net/checksum.h> |
1da177e4 | 30 | |
136e92bb | 31 | #define CLUSTERIP_VERSION "0.8" |
1da177e4 | 32 | |
1da177e4 LT |
33 | MODULE_LICENSE("GPL"); |
34 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); | |
2ae15b64 | 35 | MODULE_DESCRIPTION("Xtables: CLUSTERIP target"); |
1da177e4 LT |
36 | |
37 | struct clusterip_config { | |
38 | struct list_head list; /* list of all configs */ | |
39 | atomic_t refcount; /* reference count */ | |
44513624 KK |
40 | atomic_t entries; /* number of entries/rules |
41 | * referencing us */ | |
1da177e4 | 42 | |
6a19d614 | 43 | __be32 clusterip; /* the IP address */ |
1da177e4 LT |
44 | u_int8_t clustermac[ETH_ALEN]; /* the MAC address */ |
45 | struct net_device *dev; /* device */ | |
46 | u_int16_t num_total_nodes; /* total number of nodes */ | |
136e92bb | 47 | unsigned long local_nodes; /* node number array */ |
1da177e4 LT |
48 | |
49 | #ifdef CONFIG_PROC_FS | |
50 | struct proc_dir_entry *pde; /* proc dir entry */ | |
51 | #endif | |
52 | enum clusterip_hashmode hash_mode; /* which hashing mode */ | |
53 | u_int32_t hash_initval; /* hash initialization */ | |
54 | }; | |
55 | ||
56 | static LIST_HEAD(clusterip_configs); | |
57 | ||
136e92bb | 58 | /* clusterip_lock protects the clusterip_configs list */ |
e45b1be8 | 59 | static DEFINE_RWLOCK(clusterip_lock); |
1da177e4 LT |
60 | |
61 | #ifdef CONFIG_PROC_FS | |
9a32144e | 62 | static const struct file_operations clusterip_proc_fops; |
1da177e4 LT |
63 | static struct proc_dir_entry *clusterip_procdir; |
64 | #endif | |
65 | ||
66 | static inline void | |
44513624 KK |
67 | clusterip_config_get(struct clusterip_config *c) |
68 | { | |
1da177e4 LT |
69 | atomic_inc(&c->refcount); |
70 | } | |
71 | ||
72 | static inline void | |
44513624 KK |
73 | clusterip_config_put(struct clusterip_config *c) |
74 | { | |
75 | if (atomic_dec_and_test(&c->refcount)) | |
76 | kfree(c); | |
77 | } | |
78 | ||
44513624 KK |
79 | /* decrease the count of entries using/referencing this config. If last |
80 | * entry(rule) is removed, remove the config from lists, but don't free it | |
81 | * yet, since proc-files could still be holding references */ | |
82 | static inline void | |
83 | clusterip_config_entry_put(struct clusterip_config *c) | |
84 | { | |
4dee9597 | 85 | write_lock_bh(&clusterip_lock); |
44513624 | 86 | if (atomic_dec_and_test(&c->entries)) { |
1da177e4 | 87 | list_del(&c->list); |
e45b1be8 | 88 | write_unlock_bh(&clusterip_lock); |
44513624 | 89 | |
1da177e4 LT |
90 | dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0); |
91 | dev_put(c->dev); | |
44513624 KK |
92 | |
93 | /* In case anyone still accesses the file, the open/close | |
94 | * functions are also incrementing the refcount on their own, | |
95 | * so it's safe to remove the entry even if it's in use. */ | |
96 | #ifdef CONFIG_PROC_FS | |
97 | remove_proc_entry(c->pde->name, c->pde->parent); | |
98 | #endif | |
4dee9597 | 99 | return; |
1da177e4 | 100 | } |
4dee9597 | 101 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
102 | } |
103 | ||
1da177e4 | 104 | static struct clusterip_config * |
6a19d614 | 105 | __clusterip_config_find(__be32 clusterip) |
1da177e4 | 106 | { |
4c610979 | 107 | struct clusterip_config *c; |
1da177e4 | 108 | |
4c610979 | 109 | list_for_each_entry(c, &clusterip_configs, list) { |
7c4e36bc | 110 | if (c->clusterip == clusterip) |
1da177e4 | 111 | return c; |
1da177e4 LT |
112 | } |
113 | ||
114 | return NULL; | |
115 | } | |
116 | ||
117 | static inline struct clusterip_config * | |
6a19d614 | 118 | clusterip_config_find_get(__be32 clusterip, int entry) |
1da177e4 LT |
119 | { |
120 | struct clusterip_config *c; | |
121 | ||
e45b1be8 | 122 | read_lock_bh(&clusterip_lock); |
1da177e4 LT |
123 | c = __clusterip_config_find(clusterip); |
124 | if (!c) { | |
e45b1be8 | 125 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
126 | return NULL; |
127 | } | |
128 | atomic_inc(&c->refcount); | |
44513624 KK |
129 | if (entry) |
130 | atomic_inc(&c->entries); | |
e45b1be8 | 131 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
132 | |
133 | return c; | |
134 | } | |
135 | ||
136e92bb KK |
136 | static void |
137 | clusterip_config_init_nodelist(struct clusterip_config *c, | |
138 | const struct ipt_clusterip_tgt_info *i) | |
139 | { | |
140 | int n; | |
141 | ||
7c4e36bc | 142 | for (n = 0; n < i->num_local_nodes; n++) |
136e92bb | 143 | set_bit(i->local_nodes[n] - 1, &c->local_nodes); |
136e92bb KK |
144 | } |
145 | ||
1da177e4 | 146 | static struct clusterip_config * |
3cf93c96 | 147 | clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip, |
1da177e4 LT |
148 | struct net_device *dev) |
149 | { | |
150 | struct clusterip_config *c; | |
1da177e4 | 151 | |
0da974f4 | 152 | c = kzalloc(sizeof(*c), GFP_ATOMIC); |
1da177e4 LT |
153 | if (!c) |
154 | return NULL; | |
155 | ||
1da177e4 LT |
156 | c->dev = dev; |
157 | c->clusterip = ip; | |
158 | memcpy(&c->clustermac, &i->clustermac, ETH_ALEN); | |
159 | c->num_total_nodes = i->num_total_nodes; | |
136e92bb | 160 | clusterip_config_init_nodelist(c, i); |
1da177e4 LT |
161 | c->hash_mode = i->hash_mode; |
162 | c->hash_initval = i->hash_initval; | |
163 | atomic_set(&c->refcount, 1); | |
44513624 | 164 | atomic_set(&c->entries, 1); |
1da177e4 LT |
165 | |
166 | #ifdef CONFIG_PROC_FS | |
76592584 PM |
167 | { |
168 | char buffer[16]; | |
169 | ||
170 | /* create proc dir entry */ | |
cffee385 | 171 | sprintf(buffer, "%pI4", &ip); |
6e79d85d DL |
172 | c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR, |
173 | clusterip_procdir, | |
174 | &clusterip_proc_fops, c); | |
76592584 PM |
175 | if (!c->pde) { |
176 | kfree(c); | |
177 | return NULL; | |
178 | } | |
1da177e4 | 179 | } |
1da177e4 LT |
180 | #endif |
181 | ||
e45b1be8 | 182 | write_lock_bh(&clusterip_lock); |
1da177e4 | 183 | list_add(&c->list, &clusterip_configs); |
e45b1be8 | 184 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
185 | |
186 | return c; | |
187 | } | |
188 | ||
76592584 | 189 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
190 | static int |
191 | clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum) | |
192 | { | |
1da177e4 | 193 | |
136e92bb KK |
194 | if (nodenum == 0 || |
195 | nodenum > c->num_total_nodes) | |
1da177e4 | 196 | return 1; |
1da177e4 | 197 | |
136e92bb KK |
198 | /* check if we already have this number in our bitfield */ |
199 | if (test_and_set_bit(nodenum - 1, &c->local_nodes)) | |
200 | return 1; | |
1da177e4 | 201 | |
1da177e4 LT |
202 | return 0; |
203 | } | |
204 | ||
e1931b78 | 205 | static bool |
1da177e4 LT |
206 | clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum) |
207 | { | |
136e92bb KK |
208 | if (nodenum == 0 || |
209 | nodenum > c->num_total_nodes) | |
e1931b78 | 210 | return true; |
e905a9ed | 211 | |
136e92bb | 212 | if (test_and_clear_bit(nodenum - 1, &c->local_nodes)) |
e1931b78 | 213 | return false; |
1da177e4 | 214 | |
e1931b78 | 215 | return true; |
1da177e4 | 216 | } |
76592584 | 217 | #endif |
1da177e4 LT |
218 | |
219 | static inline u_int32_t | |
a47362a2 JE |
220 | clusterip_hashfn(const struct sk_buff *skb, |
221 | const struct clusterip_config *config) | |
1da177e4 | 222 | { |
a47362a2 | 223 | const struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
224 | unsigned long hashval; |
225 | u_int16_t sport, dport; | |
a47362a2 | 226 | const u_int16_t *ports; |
1da177e4 LT |
227 | |
228 | switch (iph->protocol) { | |
229 | case IPPROTO_TCP: | |
1da177e4 | 230 | case IPPROTO_UDP: |
a8d0f952 | 231 | case IPPROTO_UDPLITE: |
957dc80a PM |
232 | case IPPROTO_SCTP: |
233 | case IPPROTO_DCCP: | |
1da177e4 | 234 | case IPPROTO_ICMP: |
a47362a2 | 235 | ports = (const void *)iph+iph->ihl*4; |
957dc80a PM |
236 | sport = ports[0]; |
237 | dport = ports[1]; | |
1da177e4 LT |
238 | break; |
239 | default: | |
7c4e36bc | 240 | if (net_ratelimit()) |
1da177e4 LT |
241 | printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n", |
242 | iph->protocol); | |
1da177e4 LT |
243 | sport = dport = 0; |
244 | } | |
245 | ||
246 | switch (config->hash_mode) { | |
247 | case CLUSTERIP_HASHMODE_SIP: | |
248 | hashval = jhash_1word(ntohl(iph->saddr), | |
249 | config->hash_initval); | |
250 | break; | |
251 | case CLUSTERIP_HASHMODE_SIP_SPT: | |
e905a9ed | 252 | hashval = jhash_2words(ntohl(iph->saddr), sport, |
1da177e4 LT |
253 | config->hash_initval); |
254 | break; | |
255 | case CLUSTERIP_HASHMODE_SIP_SPT_DPT: | |
256 | hashval = jhash_3words(ntohl(iph->saddr), sport, dport, | |
257 | config->hash_initval); | |
258 | break; | |
259 | default: | |
260 | /* to make gcc happy */ | |
261 | hashval = 0; | |
262 | /* This cannot happen, unless the check function wasn't called | |
263 | * at rule load time */ | |
264 | printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode); | |
265 | BUG(); | |
266 | break; | |
267 | } | |
268 | ||
269 | /* node numbers are 1..n, not 0..n */ | |
34498825 | 270 | return (((u64)hashval * config->num_total_nodes) >> 32) + 1; |
1da177e4 LT |
271 | } |
272 | ||
273 | static inline int | |
a47362a2 | 274 | clusterip_responsible(const struct clusterip_config *config, u_int32_t hash) |
1da177e4 | 275 | { |
136e92bb | 276 | return test_bit(hash - 1, &config->local_nodes); |
1da177e4 LT |
277 | } |
278 | ||
e905a9ed YH |
279 | /*********************************************************************** |
280 | * IPTABLES TARGET | |
1da177e4 LT |
281 | ***********************************************************************/ |
282 | ||
283 | static unsigned int | |
7eb35586 | 284 | clusterip_tg(struct sk_buff *skb, const struct xt_target_param *par) |
1da177e4 | 285 | { |
7eb35586 | 286 | const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
587aa641 | 287 | struct nf_conn *ct; |
1da177e4 | 288 | enum ip_conntrack_info ctinfo; |
587aa641 | 289 | u_int32_t hash; |
1da177e4 LT |
290 | |
291 | /* don't need to clusterip_config_get() here, since refcount | |
292 | * is only decremented by destroy() - and ip_tables guarantees | |
293 | * that the ->target() function isn't called after ->destroy() */ | |
294 | ||
3db05fea | 295 | ct = nf_ct_get(skb, &ctinfo); |
587aa641 | 296 | if (ct == NULL) { |
1da177e4 LT |
297 | printk(KERN_ERR "CLUSTERIP: no conntrack!\n"); |
298 | /* FIXME: need to drop invalid ones, since replies | |
e905a9ed | 299 | * to outgoing connections of other nodes will be |
1da177e4 LT |
300 | * marked as INVALID */ |
301 | return NF_DROP; | |
302 | } | |
303 | ||
304 | /* special case: ICMP error handling. conntrack distinguishes between | |
305 | * error messages (RELATED) and information requests (see below) */ | |
3db05fea | 306 | if (ip_hdr(skb)->protocol == IPPROTO_ICMP |
e905a9ed | 307 | && (ctinfo == IP_CT_RELATED |
5d927eb0 | 308 | || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY)) |
6709dbbb | 309 | return XT_CONTINUE; |
1da177e4 | 310 | |
e905a9ed | 311 | /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO, |
1da177e4 LT |
312 | * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here |
313 | * on, which all have an ID field [relevant for hashing]. */ | |
314 | ||
3db05fea | 315 | hash = clusterip_hashfn(skb, cipinfo->config); |
1da177e4 LT |
316 | |
317 | switch (ctinfo) { | |
318 | case IP_CT_NEW: | |
587aa641 | 319 | ct->mark = hash; |
1da177e4 LT |
320 | break; |
321 | case IP_CT_RELATED: | |
322 | case IP_CT_RELATED+IP_CT_IS_REPLY: | |
323 | /* FIXME: we don't handle expectations at the | |
324 | * moment. they can arrive on a different node than | |
325 | * the master connection (e.g. FTP passive mode) */ | |
326 | case IP_CT_ESTABLISHED: | |
327 | case IP_CT_ESTABLISHED+IP_CT_IS_REPLY: | |
328 | break; | |
329 | default: | |
330 | break; | |
331 | } | |
332 | ||
0d53778e | 333 | #ifdef DEBUG |
3c9fba65 | 334 | nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
1da177e4 | 335 | #endif |
0d53778e | 336 | pr_debug("hash=%u ct_hash=%u ", hash, ct->mark); |
1da177e4 | 337 | if (!clusterip_responsible(cipinfo->config, hash)) { |
0d53778e | 338 | pr_debug("not responsible\n"); |
1da177e4 LT |
339 | return NF_DROP; |
340 | } | |
0d53778e | 341 | pr_debug("responsible\n"); |
1da177e4 LT |
342 | |
343 | /* despite being received via linklayer multicast, this is | |
344 | * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */ | |
3db05fea | 345 | skb->pkt_type = PACKET_HOST; |
1da177e4 | 346 | |
6709dbbb | 347 | return XT_CONTINUE; |
1da177e4 LT |
348 | } |
349 | ||
af5d6dc2 | 350 | static bool clusterip_tg_check(const struct xt_tgchk_param *par) |
1da177e4 | 351 | { |
af5d6dc2 JE |
352 | struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
353 | const struct ipt_entry *e = par->entryinfo; | |
1da177e4 LT |
354 | |
355 | struct clusterip_config *config; | |
356 | ||
1da177e4 LT |
357 | if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP && |
358 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT && | |
359 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) { | |
360 | printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n", | |
361 | cipinfo->hash_mode); | |
e1931b78 | 362 | return false; |
1da177e4 LT |
363 | |
364 | } | |
6a19d614 | 365 | if (e->ip.dmsk.s_addr != htonl(0xffffffff) |
1da177e4 LT |
366 | || e->ip.dst.s_addr == 0) { |
367 | printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n"); | |
e1931b78 | 368 | return false; |
1da177e4 LT |
369 | } |
370 | ||
371 | /* FIXME: further sanity checks */ | |
372 | ||
44513624 | 373 | config = clusterip_config_find_get(e->ip.dst.s_addr, 1); |
d3c3f424 | 374 | if (!config) { |
1da177e4 | 375 | if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) { |
cffee385 | 376 | printk(KERN_WARNING "CLUSTERIP: no config found for %pI4, need 'new'\n", &e->ip.dst.s_addr); |
e1931b78 | 377 | return false; |
1da177e4 LT |
378 | } else { |
379 | struct net_device *dev; | |
380 | ||
381 | if (e->ip.iniface[0] == '\0') { | |
382 | printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n"); | |
e1931b78 | 383 | return false; |
1da177e4 LT |
384 | } |
385 | ||
881d966b | 386 | dev = dev_get_by_name(&init_net, e->ip.iniface); |
1da177e4 LT |
387 | if (!dev) { |
388 | printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface); | |
e1931b78 | 389 | return false; |
1da177e4 LT |
390 | } |
391 | ||
e905a9ed | 392 | config = clusterip_config_init(cipinfo, |
1da177e4 LT |
393 | e->ip.dst.s_addr, dev); |
394 | if (!config) { | |
395 | printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n"); | |
396 | dev_put(dev); | |
e1931b78 | 397 | return false; |
1da177e4 LT |
398 | } |
399 | dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0); | |
400 | } | |
401 | } | |
d3c3f424 | 402 | cipinfo->config = config; |
1da177e4 | 403 | |
af5d6dc2 | 404 | if (nf_ct_l3proto_try_module_get(par->target->family) < 0) { |
11078c37 | 405 | printk(KERN_WARNING "can't load conntrack support for " |
af5d6dc2 | 406 | "proto=%u\n", par->target->family); |
e1931b78 | 407 | return false; |
11078c37 YK |
408 | } |
409 | ||
e1931b78 | 410 | return true; |
1da177e4 LT |
411 | } |
412 | ||
413 | /* drop reference count of cluster config when rule is deleted */ | |
a2df1648 | 414 | static void clusterip_tg_destroy(const struct xt_tgdtor_param *par) |
1da177e4 | 415 | { |
a2df1648 | 416 | const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
1da177e4 | 417 | |
44513624 KK |
418 | /* if no more entries are referencing the config, remove it |
419 | * from the list and destroy the proc entry */ | |
420 | clusterip_config_entry_put(cipinfo->config); | |
421 | ||
1da177e4 | 422 | clusterip_config_put(cipinfo->config); |
11078c37 | 423 | |
a2df1648 | 424 | nf_ct_l3proto_module_put(par->target->family); |
1da177e4 LT |
425 | } |
426 | ||
d3c3f424 PM |
427 | #ifdef CONFIG_COMPAT |
428 | struct compat_ipt_clusterip_tgt_info | |
429 | { | |
430 | u_int32_t flags; | |
431 | u_int8_t clustermac[6]; | |
432 | u_int16_t num_total_nodes; | |
433 | u_int16_t num_local_nodes; | |
434 | u_int16_t local_nodes[CLUSTERIP_MAX_NODES]; | |
435 | u_int32_t hash_mode; | |
436 | u_int32_t hash_initval; | |
437 | compat_uptr_t config; | |
438 | }; | |
439 | #endif /* CONFIG_COMPAT */ | |
440 | ||
d3c5ee6d | 441 | static struct xt_target clusterip_tg_reg __read_mostly = { |
1d5cd909 | 442 | .name = "CLUSTERIP", |
ee999d8b | 443 | .family = NFPROTO_IPV4, |
d3c5ee6d JE |
444 | .target = clusterip_tg, |
445 | .checkentry = clusterip_tg_check, | |
446 | .destroy = clusterip_tg_destroy, | |
d3c3f424 PM |
447 | .targetsize = sizeof(struct ipt_clusterip_tgt_info), |
448 | #ifdef CONFIG_COMPAT | |
449 | .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info), | |
450 | #endif /* CONFIG_COMPAT */ | |
1d5cd909 | 451 | .me = THIS_MODULE |
1da177e4 LT |
452 | }; |
453 | ||
454 | ||
e905a9ed YH |
455 | /*********************************************************************** |
456 | * ARP MANGLING CODE | |
1da177e4 LT |
457 | ***********************************************************************/ |
458 | ||
459 | /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */ | |
460 | struct arp_payload { | |
461 | u_int8_t src_hw[ETH_ALEN]; | |
6a19d614 | 462 | __be32 src_ip; |
1da177e4 | 463 | u_int8_t dst_hw[ETH_ALEN]; |
6a19d614 | 464 | __be32 dst_ip; |
1da177e4 LT |
465 | } __attribute__ ((packed)); |
466 | ||
0d53778e | 467 | #ifdef DEBUG |
e905a9ed | 468 | static void arp_print(struct arp_payload *payload) |
1da177e4 LT |
469 | { |
470 | #define HBUFFERLEN 30 | |
471 | char hbuffer[HBUFFERLEN]; | |
472 | int j,k; | |
1da177e4 LT |
473 | |
474 | for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) { | |
6a8341b6 HH |
475 | hbuffer[k++] = hex_asc_hi(payload->src_hw[j]); |
476 | hbuffer[k++] = hex_asc_lo(payload->src_hw[j]); | |
1da177e4 LT |
477 | hbuffer[k++]=':'; |
478 | } | |
479 | hbuffer[--k]='\0'; | |
480 | ||
cffee385 HH |
481 | printk("src %pI4@%s, dst %pI4\n", |
482 | &payload->src_ip, hbuffer, &payload->dst_ip); | |
1da177e4 LT |
483 | } |
484 | #endif | |
485 | ||
486 | static unsigned int | |
487 | arp_mangle(unsigned int hook, | |
3db05fea | 488 | struct sk_buff *skb, |
1da177e4 LT |
489 | const struct net_device *in, |
490 | const struct net_device *out, | |
491 | int (*okfn)(struct sk_buff *)) | |
492 | { | |
3db05fea | 493 | struct arphdr *arp = arp_hdr(skb); |
1da177e4 LT |
494 | struct arp_payload *payload; |
495 | struct clusterip_config *c; | |
496 | ||
497 | /* we don't care about non-ethernet and non-ipv4 ARP */ | |
498 | if (arp->ar_hrd != htons(ARPHRD_ETHER) | |
499 | || arp->ar_pro != htons(ETH_P_IP) | |
500 | || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN) | |
501 | return NF_ACCEPT; | |
502 | ||
4095ebf1 HW |
503 | /* we only want to mangle arp requests and replies */ |
504 | if (arp->ar_op != htons(ARPOP_REPLY) | |
505 | && arp->ar_op != htons(ARPOP_REQUEST)) | |
1da177e4 LT |
506 | return NF_ACCEPT; |
507 | ||
508 | payload = (void *)(arp+1); | |
509 | ||
e905a9ed | 510 | /* if there is no clusterip configuration for the arp reply's |
1da177e4 | 511 | * source ip, we don't want to mangle it */ |
44513624 | 512 | c = clusterip_config_find_get(payload->src_ip, 0); |
1da177e4 LT |
513 | if (!c) |
514 | return NF_ACCEPT; | |
515 | ||
e905a9ed | 516 | /* normally the linux kernel always replies to arp queries of |
1da177e4 LT |
517 | * addresses on different interfacs. However, in the CLUSTERIP case |
518 | * this wouldn't work, since we didn't subscribe the mcast group on | |
519 | * other interfaces */ | |
520 | if (c->dev != out) { | |
0d53778e PM |
521 | pr_debug("CLUSTERIP: not mangling arp reply on different " |
522 | "interface: cip'%s'-skb'%s'\n", | |
523 | c->dev->name, out->name); | |
1da177e4 LT |
524 | clusterip_config_put(c); |
525 | return NF_ACCEPT; | |
526 | } | |
527 | ||
528 | /* mangle reply hardware address */ | |
529 | memcpy(payload->src_hw, c->clustermac, arp->ar_hln); | |
530 | ||
0d53778e PM |
531 | #ifdef DEBUG |
532 | pr_debug(KERN_DEBUG "CLUSTERIP mangled arp reply: "); | |
1da177e4 LT |
533 | arp_print(payload); |
534 | #endif | |
535 | ||
536 | clusterip_config_put(c); | |
537 | ||
538 | return NF_ACCEPT; | |
539 | } | |
540 | ||
1999414a | 541 | static struct nf_hook_ops cip_arp_ops __read_mostly = { |
1da177e4 | 542 | .hook = arp_mangle, |
ee999d8b | 543 | .pf = NFPROTO_ARP, |
1da177e4 LT |
544 | .hooknum = NF_ARP_OUT, |
545 | .priority = -1 | |
546 | }; | |
547 | ||
e905a9ed YH |
548 | /*********************************************************************** |
549 | * PROC DIR HANDLING | |
1da177e4 LT |
550 | ***********************************************************************/ |
551 | ||
552 | #ifdef CONFIG_PROC_FS | |
553 | ||
136e92bb KK |
554 | struct clusterip_seq_position { |
555 | unsigned int pos; /* position */ | |
556 | unsigned int weight; /* number of bits set == size */ | |
557 | unsigned int bit; /* current bit */ | |
558 | unsigned long val; /* current value */ | |
559 | }; | |
560 | ||
1da177e4 LT |
561 | static void *clusterip_seq_start(struct seq_file *s, loff_t *pos) |
562 | { | |
3cf93c96 | 563 | const struct proc_dir_entry *pde = s->private; |
1da177e4 | 564 | struct clusterip_config *c = pde->data; |
136e92bb KK |
565 | unsigned int weight; |
566 | u_int32_t local_nodes; | |
567 | struct clusterip_seq_position *idx; | |
568 | ||
569 | /* FIXME: possible race */ | |
570 | local_nodes = c->local_nodes; | |
571 | weight = hweight32(local_nodes); | |
572 | if (*pos >= weight) | |
1da177e4 LT |
573 | return NULL; |
574 | ||
136e92bb KK |
575 | idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL); |
576 | if (!idx) | |
1da177e4 LT |
577 | return ERR_PTR(-ENOMEM); |
578 | ||
136e92bb KK |
579 | idx->pos = *pos; |
580 | idx->weight = weight; | |
581 | idx->bit = ffs(local_nodes); | |
582 | idx->val = local_nodes; | |
583 | clear_bit(idx->bit - 1, &idx->val); | |
584 | ||
585 | return idx; | |
1da177e4 LT |
586 | } |
587 | ||
588 | static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos) | |
589 | { | |
3cf93c96 | 590 | struct clusterip_seq_position *idx = v; |
1da177e4 | 591 | |
136e92bb KK |
592 | *pos = ++idx->pos; |
593 | if (*pos >= idx->weight) { | |
1da177e4 LT |
594 | kfree(v); |
595 | return NULL; | |
596 | } | |
136e92bb KK |
597 | idx->bit = ffs(idx->val); |
598 | clear_bit(idx->bit - 1, &idx->val); | |
599 | return idx; | |
1da177e4 LT |
600 | } |
601 | ||
602 | static void clusterip_seq_stop(struct seq_file *s, void *v) | |
603 | { | |
604 | kfree(v); | |
1da177e4 LT |
605 | } |
606 | ||
607 | static int clusterip_seq_show(struct seq_file *s, void *v) | |
608 | { | |
3cf93c96 | 609 | struct clusterip_seq_position *idx = v; |
1da177e4 | 610 | |
e905a9ed | 611 | if (idx->pos != 0) |
1da177e4 | 612 | seq_putc(s, ','); |
1da177e4 | 613 | |
136e92bb KK |
614 | seq_printf(s, "%u", idx->bit); |
615 | ||
616 | if (idx->pos == idx->weight - 1) | |
1da177e4 LT |
617 | seq_putc(s, '\n'); |
618 | ||
619 | return 0; | |
620 | } | |
621 | ||
56b3d975 | 622 | static const struct seq_operations clusterip_seq_ops = { |
1da177e4 LT |
623 | .start = clusterip_seq_start, |
624 | .next = clusterip_seq_next, | |
625 | .stop = clusterip_seq_stop, | |
626 | .show = clusterip_seq_show, | |
627 | }; | |
628 | ||
629 | static int clusterip_proc_open(struct inode *inode, struct file *file) | |
630 | { | |
631 | int ret = seq_open(file, &clusterip_seq_ops); | |
632 | ||
633 | if (!ret) { | |
634 | struct seq_file *sf = file->private_data; | |
635 | struct proc_dir_entry *pde = PDE(inode); | |
636 | struct clusterip_config *c = pde->data; | |
637 | ||
638 | sf->private = pde; | |
639 | ||
640 | clusterip_config_get(c); | |
641 | } | |
642 | ||
643 | return ret; | |
644 | } | |
645 | ||
646 | static int clusterip_proc_release(struct inode *inode, struct file *file) | |
647 | { | |
648 | struct proc_dir_entry *pde = PDE(inode); | |
649 | struct clusterip_config *c = pde->data; | |
650 | int ret; | |
651 | ||
652 | ret = seq_release(inode, file); | |
653 | ||
654 | if (!ret) | |
655 | clusterip_config_put(c); | |
656 | ||
657 | return ret; | |
658 | } | |
659 | ||
660 | static ssize_t clusterip_proc_write(struct file *file, const char __user *input, | |
661 | size_t size, loff_t *ofs) | |
662 | { | |
663 | #define PROC_WRITELEN 10 | |
664 | char buffer[PROC_WRITELEN+1]; | |
3cf93c96 | 665 | const struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode); |
1da177e4 LT |
666 | struct clusterip_config *c = pde->data; |
667 | unsigned long nodenum; | |
668 | ||
669 | if (copy_from_user(buffer, input, PROC_WRITELEN)) | |
670 | return -EFAULT; | |
671 | ||
672 | if (*buffer == '+') { | |
673 | nodenum = simple_strtoul(buffer+1, NULL, 10); | |
674 | if (clusterip_add_node(c, nodenum)) | |
675 | return -ENOMEM; | |
676 | } else if (*buffer == '-') { | |
677 | nodenum = simple_strtoul(buffer+1, NULL,10); | |
678 | if (clusterip_del_node(c, nodenum)) | |
679 | return -ENOENT; | |
680 | } else | |
681 | return -EIO; | |
682 | ||
683 | return size; | |
684 | } | |
685 | ||
9a32144e | 686 | static const struct file_operations clusterip_proc_fops = { |
1da177e4 LT |
687 | .owner = THIS_MODULE, |
688 | .open = clusterip_proc_open, | |
689 | .read = seq_read, | |
690 | .write = clusterip_proc_write, | |
691 | .llseek = seq_lseek, | |
692 | .release = clusterip_proc_release, | |
693 | }; | |
694 | ||
695 | #endif /* CONFIG_PROC_FS */ | |
696 | ||
d3c5ee6d | 697 | static int __init clusterip_tg_init(void) |
1da177e4 LT |
698 | { |
699 | int ret; | |
700 | ||
d3c5ee6d | 701 | ret = xt_register_target(&clusterip_tg_reg); |
32292a7f PM |
702 | if (ret < 0) |
703 | return ret; | |
1da177e4 | 704 | |
32292a7f PM |
705 | ret = nf_register_hook(&cip_arp_ops); |
706 | if (ret < 0) | |
1da177e4 | 707 | goto cleanup_target; |
1da177e4 LT |
708 | |
709 | #ifdef CONFIG_PROC_FS | |
457c4cbc | 710 | clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", init_net.proc_net); |
1da177e4 LT |
711 | if (!clusterip_procdir) { |
712 | printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n"); | |
713 | ret = -ENOMEM; | |
714 | goto cleanup_hook; | |
715 | } | |
716 | #endif /* CONFIG_PROC_FS */ | |
717 | ||
718 | printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n", | |
719 | CLUSTERIP_VERSION); | |
1da177e4 LT |
720 | return 0; |
721 | ||
76592584 | 722 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
723 | cleanup_hook: |
724 | nf_unregister_hook(&cip_arp_ops); | |
76592584 | 725 | #endif /* CONFIG_PROC_FS */ |
1da177e4 | 726 | cleanup_target: |
d3c5ee6d | 727 | xt_unregister_target(&clusterip_tg_reg); |
32292a7f | 728 | return ret; |
1da177e4 LT |
729 | } |
730 | ||
d3c5ee6d | 731 | static void __exit clusterip_tg_exit(void) |
1da177e4 | 732 | { |
32292a7f PM |
733 | printk(KERN_NOTICE "ClusterIP Version %s unloading\n", |
734 | CLUSTERIP_VERSION); | |
735 | #ifdef CONFIG_PROC_FS | |
736 | remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent); | |
737 | #endif | |
738 | nf_unregister_hook(&cip_arp_ops); | |
d3c5ee6d | 739 | xt_unregister_target(&clusterip_tg_reg); |
1da177e4 LT |
740 | } |
741 | ||
d3c5ee6d JE |
742 | module_init(clusterip_tg_init); |
743 | module_exit(clusterip_tg_exit); |