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> | |
13 | #include <linux/config.h> | |
14 | #include <linux/proc_fs.h> | |
15 | #include <linux/jhash.h> | |
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> | |
31 | #include <linux/netfilter_ipv4/ip_conntrack.h> | |
1da177e4 | 32 | |
4095ebf1 | 33 | #define CLUSTERIP_VERSION "0.7" |
1da177e4 LT |
34 | |
35 | #define DEBUG_CLUSTERIP | |
36 | ||
37 | #ifdef DEBUG_CLUSTERIP | |
38 | #define DEBUGP printk | |
39 | #else | |
40 | #define DEBUGP | |
41 | #endif | |
42 | ||
e45b1be8 PM |
43 | #define ASSERT_READ_LOCK(x) |
44 | ||
1da177e4 LT |
45 | MODULE_LICENSE("GPL"); |
46 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); | |
47 | MODULE_DESCRIPTION("iptables target for CLUSTERIP"); | |
48 | ||
49 | struct clusterip_config { | |
50 | struct list_head list; /* list of all configs */ | |
51 | atomic_t refcount; /* reference count */ | |
52 | ||
53 | u_int32_t clusterip; /* the IP address */ | |
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 */ | |
57 | u_int16_t num_local_nodes; /* number of local nodes */ | |
58 | u_int16_t local_nodes[CLUSTERIP_MAX_NODES]; /* node number array */ | |
59 | ||
60 | #ifdef CONFIG_PROC_FS | |
61 | struct proc_dir_entry *pde; /* proc dir entry */ | |
62 | #endif | |
63 | enum clusterip_hashmode hash_mode; /* which hashing mode */ | |
64 | u_int32_t hash_initval; /* hash initialization */ | |
65 | }; | |
66 | ||
67 | static LIST_HEAD(clusterip_configs); | |
68 | ||
69 | /* clusterip_lock protects the clusterip_configs list _AND_ the configurable | |
70 | * data within all structurses (num_local_nodes, local_nodes[]) */ | |
e45b1be8 | 71 | static DEFINE_RWLOCK(clusterip_lock); |
1da177e4 LT |
72 | |
73 | #ifdef CONFIG_PROC_FS | |
74 | static struct file_operations clusterip_proc_fops; | |
75 | static struct proc_dir_entry *clusterip_procdir; | |
76 | #endif | |
77 | ||
78 | static inline void | |
79 | clusterip_config_get(struct clusterip_config *c) { | |
80 | atomic_inc(&c->refcount); | |
81 | } | |
82 | ||
83 | static inline void | |
84 | clusterip_config_put(struct clusterip_config *c) { | |
85 | if (atomic_dec_and_test(&c->refcount)) { | |
e45b1be8 | 86 | write_lock_bh(&clusterip_lock); |
1da177e4 | 87 | list_del(&c->list); |
e45b1be8 | 88 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
89 | dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0); |
90 | dev_put(c->dev); | |
91 | kfree(c); | |
92 | } | |
93 | } | |
94 | ||
95 | ||
96 | static struct clusterip_config * | |
97 | __clusterip_config_find(u_int32_t clusterip) | |
98 | { | |
99 | struct list_head *pos; | |
100 | ||
e45b1be8 | 101 | ASSERT_READ_LOCK(&clusterip_lock); |
1da177e4 LT |
102 | list_for_each(pos, &clusterip_configs) { |
103 | struct clusterip_config *c = list_entry(pos, | |
104 | struct clusterip_config, list); | |
105 | if (c->clusterip == clusterip) { | |
106 | return c; | |
107 | } | |
108 | } | |
109 | ||
110 | return NULL; | |
111 | } | |
112 | ||
113 | static inline struct clusterip_config * | |
114 | clusterip_config_find_get(u_int32_t clusterip) | |
115 | { | |
116 | struct clusterip_config *c; | |
117 | ||
e45b1be8 | 118 | read_lock_bh(&clusterip_lock); |
1da177e4 LT |
119 | c = __clusterip_config_find(clusterip); |
120 | if (!c) { | |
e45b1be8 | 121 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
122 | return NULL; |
123 | } | |
124 | atomic_inc(&c->refcount); | |
e45b1be8 | 125 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
126 | |
127 | return c; | |
128 | } | |
129 | ||
130 | static struct clusterip_config * | |
131 | clusterip_config_init(struct ipt_clusterip_tgt_info *i, u_int32_t ip, | |
132 | struct net_device *dev) | |
133 | { | |
134 | struct clusterip_config *c; | |
135 | char buffer[16]; | |
136 | ||
137 | c = kmalloc(sizeof(*c), GFP_ATOMIC); | |
138 | if (!c) | |
139 | return NULL; | |
140 | ||
141 | memset(c, 0, sizeof(*c)); | |
142 | c->dev = dev; | |
143 | c->clusterip = ip; | |
144 | memcpy(&c->clustermac, &i->clustermac, ETH_ALEN); | |
145 | c->num_total_nodes = i->num_total_nodes; | |
146 | c->num_local_nodes = i->num_local_nodes; | |
5170dbeb | 147 | memcpy(&c->local_nodes, &i->local_nodes, sizeof(c->local_nodes)); |
1da177e4 LT |
148 | c->hash_mode = i->hash_mode; |
149 | c->hash_initval = i->hash_initval; | |
150 | atomic_set(&c->refcount, 1); | |
151 | ||
152 | #ifdef CONFIG_PROC_FS | |
153 | /* create proc dir entry */ | |
154 | sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip)); | |
155 | c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR, clusterip_procdir); | |
156 | if (!c->pde) { | |
157 | kfree(c); | |
158 | return NULL; | |
159 | } | |
160 | c->pde->proc_fops = &clusterip_proc_fops; | |
161 | c->pde->data = c; | |
162 | #endif | |
163 | ||
e45b1be8 | 164 | write_lock_bh(&clusterip_lock); |
1da177e4 | 165 | list_add(&c->list, &clusterip_configs); |
e45b1be8 | 166 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
167 | |
168 | return c; | |
169 | } | |
170 | ||
171 | static int | |
172 | clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum) | |
173 | { | |
174 | int i; | |
175 | ||
e45b1be8 | 176 | write_lock_bh(&clusterip_lock); |
1da177e4 LT |
177 | |
178 | if (c->num_local_nodes >= CLUSTERIP_MAX_NODES | |
179 | || nodenum > CLUSTERIP_MAX_NODES) { | |
e45b1be8 | 180 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
181 | return 1; |
182 | } | |
183 | ||
184 | /* check if we alrady have this number in our array */ | |
185 | for (i = 0; i < c->num_local_nodes; i++) { | |
186 | if (c->local_nodes[i] == nodenum) { | |
e45b1be8 | 187 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
188 | return 1; |
189 | } | |
190 | } | |
191 | ||
192 | c->local_nodes[c->num_local_nodes++] = nodenum; | |
193 | ||
e45b1be8 | 194 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
195 | return 0; |
196 | } | |
197 | ||
198 | static int | |
199 | clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum) | |
200 | { | |
201 | int i; | |
202 | ||
e45b1be8 | 203 | write_lock_bh(&clusterip_lock); |
1da177e4 LT |
204 | |
205 | if (c->num_local_nodes <= 1 || nodenum > CLUSTERIP_MAX_NODES) { | |
e45b1be8 | 206 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
207 | return 1; |
208 | } | |
209 | ||
210 | for (i = 0; i < c->num_local_nodes; i++) { | |
211 | if (c->local_nodes[i] == nodenum) { | |
212 | int size = sizeof(u_int16_t)*(c->num_local_nodes-(i+1)); | |
213 | memmove(&c->local_nodes[i], &c->local_nodes[i+1], size); | |
214 | c->num_local_nodes--; | |
e45b1be8 | 215 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
216 | return 0; |
217 | } | |
218 | } | |
219 | ||
e45b1be8 | 220 | write_unlock_bh(&clusterip_lock); |
1da177e4 LT |
221 | return 1; |
222 | } | |
223 | ||
224 | static inline u_int32_t | |
225 | clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config) | |
226 | { | |
227 | struct iphdr *iph = skb->nh.iph; | |
228 | unsigned long hashval; | |
229 | u_int16_t sport, dport; | |
230 | struct tcphdr *th; | |
231 | struct udphdr *uh; | |
232 | struct icmphdr *ih; | |
233 | ||
234 | switch (iph->protocol) { | |
235 | case IPPROTO_TCP: | |
236 | th = (void *)iph+iph->ihl*4; | |
237 | sport = ntohs(th->source); | |
238 | dport = ntohs(th->dest); | |
239 | break; | |
240 | case IPPROTO_UDP: | |
241 | uh = (void *)iph+iph->ihl*4; | |
242 | sport = ntohs(uh->source); | |
243 | dport = ntohs(uh->dest); | |
244 | break; | |
245 | case IPPROTO_ICMP: | |
246 | ih = (void *)iph+iph->ihl*4; | |
247 | sport = ntohs(ih->un.echo.id); | |
248 | dport = (ih->type<<8)|ih->code; | |
249 | break; | |
250 | default: | |
251 | if (net_ratelimit()) { | |
252 | printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n", | |
253 | iph->protocol); | |
254 | } | |
255 | sport = dport = 0; | |
256 | } | |
257 | ||
258 | switch (config->hash_mode) { | |
259 | case CLUSTERIP_HASHMODE_SIP: | |
260 | hashval = jhash_1word(ntohl(iph->saddr), | |
261 | config->hash_initval); | |
262 | break; | |
263 | case CLUSTERIP_HASHMODE_SIP_SPT: | |
264 | hashval = jhash_2words(ntohl(iph->saddr), sport, | |
265 | config->hash_initval); | |
266 | break; | |
267 | case CLUSTERIP_HASHMODE_SIP_SPT_DPT: | |
268 | hashval = jhash_3words(ntohl(iph->saddr), sport, dport, | |
269 | config->hash_initval); | |
270 | break; | |
271 | default: | |
272 | /* to make gcc happy */ | |
273 | hashval = 0; | |
274 | /* This cannot happen, unless the check function wasn't called | |
275 | * at rule load time */ | |
276 | printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode); | |
277 | BUG(); | |
278 | break; | |
279 | } | |
280 | ||
281 | /* node numbers are 1..n, not 0..n */ | |
282 | return ((hashval % config->num_total_nodes)+1); | |
283 | } | |
284 | ||
285 | static inline int | |
286 | clusterip_responsible(struct clusterip_config *config, u_int32_t hash) | |
287 | { | |
288 | int i; | |
289 | ||
e45b1be8 | 290 | read_lock_bh(&clusterip_lock); |
1da177e4 LT |
291 | |
292 | if (config->num_local_nodes == 0) { | |
e45b1be8 | 293 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
294 | return 0; |
295 | } | |
296 | ||
297 | for (i = 0; i < config->num_local_nodes; i++) { | |
298 | if (config->local_nodes[i] == hash) { | |
e45b1be8 | 299 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
300 | return 1; |
301 | } | |
302 | } | |
303 | ||
e45b1be8 | 304 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
305 | |
306 | return 0; | |
307 | } | |
308 | ||
309 | /*********************************************************************** | |
310 | * IPTABLES TARGET | |
311 | ***********************************************************************/ | |
312 | ||
313 | static unsigned int | |
314 | target(struct sk_buff **pskb, | |
315 | const struct net_device *in, | |
316 | const struct net_device *out, | |
317 | unsigned int hooknum, | |
318 | const void *targinfo, | |
319 | void *userinfo) | |
320 | { | |
321 | const struct ipt_clusterip_tgt_info *cipinfo = targinfo; | |
322 | enum ip_conntrack_info ctinfo; | |
323 | struct ip_conntrack *ct = ip_conntrack_get((*pskb), &ctinfo); | |
324 | u_int32_t hash; | |
325 | ||
326 | /* don't need to clusterip_config_get() here, since refcount | |
327 | * is only decremented by destroy() - and ip_tables guarantees | |
328 | * that the ->target() function isn't called after ->destroy() */ | |
329 | ||
330 | if (!ct) { | |
331 | printk(KERN_ERR "CLUSTERIP: no conntrack!\n"); | |
332 | /* FIXME: need to drop invalid ones, since replies | |
333 | * to outgoing connections of other nodes will be | |
334 | * marked as INVALID */ | |
335 | return NF_DROP; | |
336 | } | |
337 | ||
338 | /* special case: ICMP error handling. conntrack distinguishes between | |
339 | * error messages (RELATED) and information requests (see below) */ | |
340 | if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP | |
341 | && (ctinfo == IP_CT_RELATED | |
5d927eb0 | 342 | || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY)) |
1da177e4 LT |
343 | return IPT_CONTINUE; |
344 | ||
345 | /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO, | |
346 | * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here | |
347 | * on, which all have an ID field [relevant for hashing]. */ | |
348 | ||
349 | hash = clusterip_hashfn(*pskb, cipinfo->config); | |
350 | ||
351 | switch (ctinfo) { | |
352 | case IP_CT_NEW: | |
353 | ct->mark = hash; | |
354 | break; | |
355 | case IP_CT_RELATED: | |
356 | case IP_CT_RELATED+IP_CT_IS_REPLY: | |
357 | /* FIXME: we don't handle expectations at the | |
358 | * moment. they can arrive on a different node than | |
359 | * the master connection (e.g. FTP passive mode) */ | |
360 | case IP_CT_ESTABLISHED: | |
361 | case IP_CT_ESTABLISHED+IP_CT_IS_REPLY: | |
362 | break; | |
363 | default: | |
364 | break; | |
365 | } | |
366 | ||
367 | #ifdef DEBUG_CLUSTERP | |
368 | DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); | |
369 | #endif | |
bf3a46aa | 370 | DEBUGP("hash=%u ct_hash=%u ", hash, ct->mark); |
1da177e4 LT |
371 | if (!clusterip_responsible(cipinfo->config, hash)) { |
372 | DEBUGP("not responsible\n"); | |
373 | return NF_DROP; | |
374 | } | |
375 | DEBUGP("responsible\n"); | |
376 | ||
377 | /* despite being received via linklayer multicast, this is | |
378 | * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */ | |
379 | (*pskb)->pkt_type = PACKET_HOST; | |
380 | ||
381 | return IPT_CONTINUE; | |
382 | } | |
383 | ||
384 | static int | |
385 | checkentry(const char *tablename, | |
386 | const struct ipt_entry *e, | |
387 | void *targinfo, | |
388 | unsigned int targinfosize, | |
389 | unsigned int hook_mask) | |
390 | { | |
391 | struct ipt_clusterip_tgt_info *cipinfo = targinfo; | |
392 | ||
393 | struct clusterip_config *config; | |
394 | ||
395 | if (targinfosize != IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info))) { | |
396 | printk(KERN_WARNING "CLUSTERIP: targinfosize %u != %Zu\n", | |
397 | targinfosize, | |
398 | IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info))); | |
399 | return 0; | |
400 | } | |
401 | ||
402 | if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP && | |
403 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT && | |
404 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) { | |
405 | printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n", | |
406 | cipinfo->hash_mode); | |
407 | return 0; | |
408 | ||
409 | } | |
410 | if (e->ip.dmsk.s_addr != 0xffffffff | |
411 | || e->ip.dst.s_addr == 0) { | |
412 | printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n"); | |
413 | return 0; | |
414 | } | |
415 | ||
416 | /* FIXME: further sanity checks */ | |
417 | ||
418 | config = clusterip_config_find_get(e->ip.dst.s_addr); | |
419 | if (!config) { | |
420 | if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) { | |
421 | printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr)); | |
422 | return 0; | |
423 | } else { | |
424 | struct net_device *dev; | |
425 | ||
426 | if (e->ip.iniface[0] == '\0') { | |
427 | printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n"); | |
428 | return 0; | |
429 | } | |
430 | ||
431 | dev = dev_get_by_name(e->ip.iniface); | |
432 | if (!dev) { | |
433 | printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface); | |
434 | return 0; | |
435 | } | |
436 | ||
437 | config = clusterip_config_init(cipinfo, | |
438 | e->ip.dst.s_addr, dev); | |
439 | if (!config) { | |
440 | printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n"); | |
441 | dev_put(dev); | |
442 | return 0; | |
443 | } | |
444 | dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0); | |
445 | } | |
446 | } | |
447 | ||
448 | cipinfo->config = config; | |
449 | ||
450 | return 1; | |
451 | } | |
452 | ||
453 | /* drop reference count of cluster config when rule is deleted */ | |
454 | static void destroy(void *matchinfo, unsigned int matchinfosize) | |
455 | { | |
456 | struct ipt_clusterip_tgt_info *cipinfo = matchinfo; | |
457 | ||
458 | /* we first remove the proc entry and then drop the reference | |
459 | * count. In case anyone still accesses the file, the open/close | |
460 | * functions are also incrementing the refcount on their own */ | |
461 | #ifdef CONFIG_PROC_FS | |
462 | remove_proc_entry(cipinfo->config->pde->name, | |
463 | cipinfo->config->pde->parent); | |
464 | #endif | |
465 | clusterip_config_put(cipinfo->config); | |
466 | } | |
467 | ||
468 | static struct ipt_target clusterip_tgt = { | |
469 | .name = "CLUSTERIP", | |
470 | .target = &target, | |
471 | .checkentry = &checkentry, | |
472 | .destroy = &destroy, | |
473 | .me = THIS_MODULE | |
474 | }; | |
475 | ||
476 | ||
477 | /*********************************************************************** | |
478 | * ARP MANGLING CODE | |
479 | ***********************************************************************/ | |
480 | ||
481 | /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */ | |
482 | struct arp_payload { | |
483 | u_int8_t src_hw[ETH_ALEN]; | |
484 | u_int32_t src_ip; | |
485 | u_int8_t dst_hw[ETH_ALEN]; | |
486 | u_int32_t dst_ip; | |
487 | } __attribute__ ((packed)); | |
488 | ||
489 | #ifdef CLUSTERIP_DEBUG | |
490 | static void arp_print(struct arp_payload *payload) | |
491 | { | |
492 | #define HBUFFERLEN 30 | |
493 | char hbuffer[HBUFFERLEN]; | |
494 | int j,k; | |
495 | const char hexbuf[]= "0123456789abcdef"; | |
496 | ||
497 | for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) { | |
498 | hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15]; | |
499 | hbuffer[k++]=hexbuf[payload->src_hw[j]&15]; | |
500 | hbuffer[k++]=':'; | |
501 | } | |
502 | hbuffer[--k]='\0'; | |
503 | ||
504 | printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n", | |
505 | NIPQUAD(payload->src_ip), hbuffer, | |
506 | NIPQUAD(payload->dst_ip)); | |
507 | } | |
508 | #endif | |
509 | ||
510 | static unsigned int | |
511 | arp_mangle(unsigned int hook, | |
512 | struct sk_buff **pskb, | |
513 | const struct net_device *in, | |
514 | const struct net_device *out, | |
515 | int (*okfn)(struct sk_buff *)) | |
516 | { | |
517 | struct arphdr *arp = (*pskb)->nh.arph; | |
518 | struct arp_payload *payload; | |
519 | struct clusterip_config *c; | |
520 | ||
521 | /* we don't care about non-ethernet and non-ipv4 ARP */ | |
522 | if (arp->ar_hrd != htons(ARPHRD_ETHER) | |
523 | || arp->ar_pro != htons(ETH_P_IP) | |
524 | || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN) | |
525 | return NF_ACCEPT; | |
526 | ||
4095ebf1 HW |
527 | /* we only want to mangle arp requests and replies */ |
528 | if (arp->ar_op != htons(ARPOP_REPLY) | |
529 | && arp->ar_op != htons(ARPOP_REQUEST)) | |
1da177e4 LT |
530 | return NF_ACCEPT; |
531 | ||
532 | payload = (void *)(arp+1); | |
533 | ||
534 | /* if there is no clusterip configuration for the arp reply's | |
535 | * source ip, we don't want to mangle it */ | |
536 | c = clusterip_config_find_get(payload->src_ip); | |
537 | if (!c) | |
538 | return NF_ACCEPT; | |
539 | ||
540 | /* normally the linux kernel always replies to arp queries of | |
541 | * addresses on different interfacs. However, in the CLUSTERIP case | |
542 | * this wouldn't work, since we didn't subscribe the mcast group on | |
543 | * other interfaces */ | |
544 | if (c->dev != out) { | |
545 | DEBUGP("CLUSTERIP: not mangling arp reply on different " | |
546 | "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name); | |
547 | clusterip_config_put(c); | |
548 | return NF_ACCEPT; | |
549 | } | |
550 | ||
551 | /* mangle reply hardware address */ | |
552 | memcpy(payload->src_hw, c->clustermac, arp->ar_hln); | |
553 | ||
554 | #ifdef CLUSTERIP_DEBUG | |
555 | DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: "); | |
556 | arp_print(payload); | |
557 | #endif | |
558 | ||
559 | clusterip_config_put(c); | |
560 | ||
561 | return NF_ACCEPT; | |
562 | } | |
563 | ||
564 | static struct nf_hook_ops cip_arp_ops = { | |
565 | .hook = arp_mangle, | |
566 | .pf = NF_ARP, | |
567 | .hooknum = NF_ARP_OUT, | |
568 | .priority = -1 | |
569 | }; | |
570 | ||
571 | /*********************************************************************** | |
572 | * PROC DIR HANDLING | |
573 | ***********************************************************************/ | |
574 | ||
575 | #ifdef CONFIG_PROC_FS | |
576 | ||
577 | static void *clusterip_seq_start(struct seq_file *s, loff_t *pos) | |
578 | { | |
579 | struct proc_dir_entry *pde = s->private; | |
580 | struct clusterip_config *c = pde->data; | |
581 | unsigned int *nodeidx; | |
582 | ||
e45b1be8 | 583 | read_lock_bh(&clusterip_lock); |
1da177e4 LT |
584 | if (*pos >= c->num_local_nodes) |
585 | return NULL; | |
586 | ||
587 | nodeidx = kmalloc(sizeof(unsigned int), GFP_KERNEL); | |
588 | if (!nodeidx) | |
589 | return ERR_PTR(-ENOMEM); | |
590 | ||
591 | *nodeidx = *pos; | |
592 | return nodeidx; | |
593 | } | |
594 | ||
595 | static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos) | |
596 | { | |
597 | struct proc_dir_entry *pde = s->private; | |
598 | struct clusterip_config *c = pde->data; | |
599 | unsigned int *nodeidx = (unsigned int *)v; | |
600 | ||
601 | *pos = ++(*nodeidx); | |
602 | if (*pos >= c->num_local_nodes) { | |
603 | kfree(v); | |
604 | return NULL; | |
605 | } | |
606 | return nodeidx; | |
607 | } | |
608 | ||
609 | static void clusterip_seq_stop(struct seq_file *s, void *v) | |
610 | { | |
611 | kfree(v); | |
612 | ||
e45b1be8 | 613 | read_unlock_bh(&clusterip_lock); |
1da177e4 LT |
614 | } |
615 | ||
616 | static int clusterip_seq_show(struct seq_file *s, void *v) | |
617 | { | |
618 | struct proc_dir_entry *pde = s->private; | |
619 | struct clusterip_config *c = pde->data; | |
620 | unsigned int *nodeidx = (unsigned int *)v; | |
621 | ||
622 | if (*nodeidx != 0) | |
623 | seq_putc(s, ','); | |
624 | seq_printf(s, "%u", c->local_nodes[*nodeidx]); | |
625 | ||
626 | if (*nodeidx == c->num_local_nodes-1) | |
627 | seq_putc(s, '\n'); | |
628 | ||
629 | return 0; | |
630 | } | |
631 | ||
632 | static struct seq_operations clusterip_seq_ops = { | |
633 | .start = clusterip_seq_start, | |
634 | .next = clusterip_seq_next, | |
635 | .stop = clusterip_seq_stop, | |
636 | .show = clusterip_seq_show, | |
637 | }; | |
638 | ||
639 | static int clusterip_proc_open(struct inode *inode, struct file *file) | |
640 | { | |
641 | int ret = seq_open(file, &clusterip_seq_ops); | |
642 | ||
643 | if (!ret) { | |
644 | struct seq_file *sf = file->private_data; | |
645 | struct proc_dir_entry *pde = PDE(inode); | |
646 | struct clusterip_config *c = pde->data; | |
647 | ||
648 | sf->private = pde; | |
649 | ||
650 | clusterip_config_get(c); | |
651 | } | |
652 | ||
653 | return ret; | |
654 | } | |
655 | ||
656 | static int clusterip_proc_release(struct inode *inode, struct file *file) | |
657 | { | |
658 | struct proc_dir_entry *pde = PDE(inode); | |
659 | struct clusterip_config *c = pde->data; | |
660 | int ret; | |
661 | ||
662 | ret = seq_release(inode, file); | |
663 | ||
664 | if (!ret) | |
665 | clusterip_config_put(c); | |
666 | ||
667 | return ret; | |
668 | } | |
669 | ||
670 | static ssize_t clusterip_proc_write(struct file *file, const char __user *input, | |
671 | size_t size, loff_t *ofs) | |
672 | { | |
673 | #define PROC_WRITELEN 10 | |
674 | char buffer[PROC_WRITELEN+1]; | |
675 | struct proc_dir_entry *pde = PDE(file->f_dentry->d_inode); | |
676 | struct clusterip_config *c = pde->data; | |
677 | unsigned long nodenum; | |
678 | ||
679 | if (copy_from_user(buffer, input, PROC_WRITELEN)) | |
680 | return -EFAULT; | |
681 | ||
682 | if (*buffer == '+') { | |
683 | nodenum = simple_strtoul(buffer+1, NULL, 10); | |
684 | if (clusterip_add_node(c, nodenum)) | |
685 | return -ENOMEM; | |
686 | } else if (*buffer == '-') { | |
687 | nodenum = simple_strtoul(buffer+1, NULL,10); | |
688 | if (clusterip_del_node(c, nodenum)) | |
689 | return -ENOENT; | |
690 | } else | |
691 | return -EIO; | |
692 | ||
693 | return size; | |
694 | } | |
695 | ||
696 | static struct file_operations clusterip_proc_fops = { | |
697 | .owner = THIS_MODULE, | |
698 | .open = clusterip_proc_open, | |
699 | .read = seq_read, | |
700 | .write = clusterip_proc_write, | |
701 | .llseek = seq_lseek, | |
702 | .release = clusterip_proc_release, | |
703 | }; | |
704 | ||
705 | #endif /* CONFIG_PROC_FS */ | |
706 | ||
707 | static int init_or_cleanup(int fini) | |
708 | { | |
709 | int ret; | |
710 | ||
711 | if (fini) | |
712 | goto cleanup; | |
713 | ||
714 | if (ipt_register_target(&clusterip_tgt)) { | |
715 | ret = -EINVAL; | |
716 | goto cleanup_none; | |
717 | } | |
718 | ||
719 | if (nf_register_hook(&cip_arp_ops) < 0) { | |
720 | ret = -EINVAL; | |
721 | goto cleanup_target; | |
722 | } | |
723 | ||
724 | #ifdef CONFIG_PROC_FS | |
725 | clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net); | |
726 | if (!clusterip_procdir) { | |
727 | printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n"); | |
728 | ret = -ENOMEM; | |
729 | goto cleanup_hook; | |
730 | } | |
731 | #endif /* CONFIG_PROC_FS */ | |
732 | ||
733 | printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n", | |
734 | CLUSTERIP_VERSION); | |
735 | ||
736 | return 0; | |
737 | ||
738 | cleanup: | |
739 | printk(KERN_NOTICE "ClusterIP Version %s unloading\n", | |
740 | CLUSTERIP_VERSION); | |
741 | #ifdef CONFIG_PROC_FS | |
742 | remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent); | |
743 | #endif | |
744 | cleanup_hook: | |
745 | nf_unregister_hook(&cip_arp_ops); | |
746 | cleanup_target: | |
747 | ipt_unregister_target(&clusterip_tgt); | |
748 | cleanup_none: | |
749 | return -EINVAL; | |
750 | } | |
751 | ||
752 | static int __init init(void) | |
753 | { | |
754 | return init_or_cleanup(0); | |
755 | } | |
756 | ||
757 | static void __exit fini(void) | |
758 | { | |
759 | init_or_cleanup(1); | |
760 | } | |
761 | ||
762 | module_init(init); | |
763 | module_exit(fini); |