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 | */ | |
ff67e4e4 | 12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
1da177e4 | 13 | #include <linux/module.h> |
1da177e4 LT |
14 | #include <linux/proc_fs.h> |
15 | #include <linux/jhash.h> | |
136e92bb | 16 | #include <linux/bitops.h> |
1da177e4 | 17 | #include <linux/skbuff.h> |
5a0e3ad6 | 18 | #include <linux/slab.h> |
1da177e4 LT |
19 | #include <linux/ip.h> |
20 | #include <linux/tcp.h> | |
21 | #include <linux/udp.h> | |
22 | #include <linux/icmp.h> | |
23 | #include <linux/if_arp.h> | |
1da177e4 | 24 | #include <linux/seq_file.h> |
1da177e4 | 25 | #include <linux/netfilter_arp.h> |
6709dbbb | 26 | #include <linux/netfilter/x_tables.h> |
1da177e4 LT |
27 | #include <linux/netfilter_ipv4/ip_tables.h> |
28 | #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h> | |
587aa641 | 29 | #include <net/netfilter/nf_conntrack.h> |
457c4cbc | 30 | #include <net/net_namespace.h> |
ce4ff76c | 31 | #include <net/netns/generic.h> |
587aa641 | 32 | #include <net/checksum.h> |
3d04ebb6 | 33 | #include <net/ip.h> |
1da177e4 | 34 | |
136e92bb | 35 | #define CLUSTERIP_VERSION "0.8" |
1da177e4 | 36 | |
1da177e4 LT |
37 | MODULE_LICENSE("GPL"); |
38 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); | |
2ae15b64 | 39 | MODULE_DESCRIPTION("Xtables: CLUSTERIP target"); |
1da177e4 LT |
40 | |
41 | struct clusterip_config { | |
42 | struct list_head list; /* list of all configs */ | |
43 | atomic_t refcount; /* reference count */ | |
44513624 KK |
44 | atomic_t entries; /* number of entries/rules |
45 | * referencing us */ | |
1da177e4 | 46 | |
6a19d614 | 47 | __be32 clusterip; /* the IP address */ |
1da177e4 LT |
48 | u_int8_t clustermac[ETH_ALEN]; /* the MAC address */ |
49 | struct net_device *dev; /* device */ | |
50 | u_int16_t num_total_nodes; /* total number of nodes */ | |
136e92bb | 51 | unsigned long local_nodes; /* node number array */ |
1da177e4 LT |
52 | |
53 | #ifdef CONFIG_PROC_FS | |
54 | struct proc_dir_entry *pde; /* proc dir entry */ | |
55 | #endif | |
56 | enum clusterip_hashmode hash_mode; /* which hashing mode */ | |
57 | u_int32_t hash_initval; /* hash initialization */ | |
d73f33b1 | 58 | struct rcu_head rcu; |
1da177e4 LT |
59 | }; |
60 | ||
1da177e4 | 61 | #ifdef CONFIG_PROC_FS |
9a32144e | 62 | static const struct file_operations clusterip_proc_fops; |
1da177e4 | 63 | #endif |
1da177e4 | 64 | |
ce4ff76c G |
65 | static int clusterip_net_id __read_mostly; |
66 | ||
67 | struct clusterip_net { | |
26a89e43 | 68 | struct list_head configs; |
f1e8077f G |
69 | /* lock protects the configs list */ |
70 | spinlock_t lock; | |
1da177e4 LT |
71 | |
72 | #ifdef CONFIG_PROC_FS | |
ce4ff76c | 73 | struct proc_dir_entry *procdir; |
1da177e4 | 74 | #endif |
ce4ff76c | 75 | }; |
1da177e4 LT |
76 | |
77 | static inline void | |
44513624 KK |
78 | clusterip_config_get(struct clusterip_config *c) |
79 | { | |
1da177e4 LT |
80 | atomic_inc(&c->refcount); |
81 | } | |
82 | ||
d73f33b1 ED |
83 | |
84 | static void clusterip_config_rcu_free(struct rcu_head *head) | |
85 | { | |
86 | kfree(container_of(head, struct clusterip_config, rcu)); | |
87 | } | |
88 | ||
1da177e4 | 89 | static inline void |
44513624 KK |
90 | clusterip_config_put(struct clusterip_config *c) |
91 | { | |
92 | if (atomic_dec_and_test(&c->refcount)) | |
d73f33b1 | 93 | call_rcu_bh(&c->rcu, clusterip_config_rcu_free); |
44513624 KK |
94 | } |
95 | ||
44513624 KK |
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 | { | |
d86946d2 G |
102 | struct net *net = dev_net(c->dev); |
103 | struct clusterip_net *cn = net_generic(net, clusterip_net_id); | |
f1e8077f | 104 | |
d73f33b1 | 105 | local_bh_disable(); |
f1e8077f | 106 | if (atomic_dec_and_lock(&c->entries, &cn->lock)) { |
d73f33b1 | 107 | list_del_rcu(&c->list); |
f1e8077f | 108 | spin_unlock(&cn->lock); |
d73f33b1 | 109 | local_bh_enable(); |
44513624 | 110 | |
22bedad3 | 111 | dev_mc_del(c->dev, c->clustermac); |
1da177e4 | 112 | dev_put(c->dev); |
44513624 KK |
113 | |
114 | /* In case anyone still accesses the file, the open/close | |
115 | * functions are also incrementing the refcount on their own, | |
116 | * so it's safe to remove the entry even if it's in use. */ | |
117 | #ifdef CONFIG_PROC_FS | |
a8ca16ea | 118 | proc_remove(c->pde); |
44513624 | 119 | #endif |
4dee9597 | 120 | return; |
1da177e4 | 121 | } |
d73f33b1 | 122 | local_bh_enable(); |
1da177e4 LT |
123 | } |
124 | ||
1da177e4 | 125 | static struct clusterip_config * |
b5ef0f85 | 126 | __clusterip_config_find(struct net *net, __be32 clusterip) |
1da177e4 | 127 | { |
4c610979 | 128 | struct clusterip_config *c; |
b5ef0f85 | 129 | struct clusterip_net *cn = net_generic(net, clusterip_net_id); |
1da177e4 | 130 | |
26a89e43 | 131 | list_for_each_entry_rcu(c, &cn->configs, list) { |
7c4e36bc | 132 | if (c->clusterip == clusterip) |
1da177e4 | 133 | return c; |
1da177e4 LT |
134 | } |
135 | ||
136 | return NULL; | |
137 | } | |
138 | ||
139 | static inline struct clusterip_config * | |
b5ef0f85 | 140 | clusterip_config_find_get(struct net *net, __be32 clusterip, int entry) |
1da177e4 LT |
141 | { |
142 | struct clusterip_config *c; | |
143 | ||
d73f33b1 | 144 | rcu_read_lock_bh(); |
b5ef0f85 | 145 | c = __clusterip_config_find(net, clusterip); |
d73f33b1 ED |
146 | if (c) { |
147 | if (unlikely(!atomic_inc_not_zero(&c->refcount))) | |
148 | c = NULL; | |
149 | else if (entry) | |
150 | atomic_inc(&c->entries); | |
1da177e4 | 151 | } |
d73f33b1 | 152 | rcu_read_unlock_bh(); |
1da177e4 LT |
153 | |
154 | return c; | |
155 | } | |
156 | ||
136e92bb KK |
157 | static void |
158 | clusterip_config_init_nodelist(struct clusterip_config *c, | |
159 | const struct ipt_clusterip_tgt_info *i) | |
160 | { | |
161 | int n; | |
162 | ||
7c4e36bc | 163 | for (n = 0; n < i->num_local_nodes; n++) |
136e92bb | 164 | set_bit(i->local_nodes[n] - 1, &c->local_nodes); |
136e92bb KK |
165 | } |
166 | ||
1da177e4 | 167 | static struct clusterip_config * |
3cf93c96 | 168 | clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip, |
1da177e4 LT |
169 | struct net_device *dev) |
170 | { | |
171 | struct clusterip_config *c; | |
f58d7866 | 172 | struct clusterip_net *cn = net_generic(dev_net(dev), clusterip_net_id); |
1da177e4 | 173 | |
0da974f4 | 174 | c = kzalloc(sizeof(*c), GFP_ATOMIC); |
1da177e4 LT |
175 | if (!c) |
176 | return NULL; | |
177 | ||
1da177e4 LT |
178 | c->dev = dev; |
179 | c->clusterip = ip; | |
180 | memcpy(&c->clustermac, &i->clustermac, ETH_ALEN); | |
181 | c->num_total_nodes = i->num_total_nodes; | |
136e92bb | 182 | clusterip_config_init_nodelist(c, i); |
1da177e4 LT |
183 | c->hash_mode = i->hash_mode; |
184 | c->hash_initval = i->hash_initval; | |
185 | atomic_set(&c->refcount, 1); | |
44513624 | 186 | atomic_set(&c->entries, 1); |
1da177e4 LT |
187 | |
188 | #ifdef CONFIG_PROC_FS | |
76592584 PM |
189 | { |
190 | char buffer[16]; | |
191 | ||
192 | /* create proc dir entry */ | |
cffee385 | 193 | sprintf(buffer, "%pI4", &ip); |
6e79d85d | 194 | c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR, |
ce4ff76c | 195 | cn->procdir, |
6e79d85d | 196 | &clusterip_proc_fops, c); |
76592584 PM |
197 | if (!c->pde) { |
198 | kfree(c); | |
199 | return NULL; | |
200 | } | |
1da177e4 | 201 | } |
1da177e4 LT |
202 | #endif |
203 | ||
f1e8077f | 204 | spin_lock_bh(&cn->lock); |
26a89e43 | 205 | list_add_rcu(&c->list, &cn->configs); |
f1e8077f | 206 | spin_unlock_bh(&cn->lock); |
1da177e4 LT |
207 | |
208 | return c; | |
209 | } | |
210 | ||
76592584 | 211 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
212 | static int |
213 | clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum) | |
214 | { | |
1da177e4 | 215 | |
136e92bb KK |
216 | if (nodenum == 0 || |
217 | nodenum > c->num_total_nodes) | |
1da177e4 | 218 | return 1; |
1da177e4 | 219 | |
136e92bb KK |
220 | /* check if we already have this number in our bitfield */ |
221 | if (test_and_set_bit(nodenum - 1, &c->local_nodes)) | |
222 | return 1; | |
1da177e4 | 223 | |
1da177e4 LT |
224 | return 0; |
225 | } | |
226 | ||
e1931b78 | 227 | static bool |
1da177e4 LT |
228 | clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum) |
229 | { | |
136e92bb KK |
230 | if (nodenum == 0 || |
231 | nodenum > c->num_total_nodes) | |
e1931b78 | 232 | return true; |
e905a9ed | 233 | |
136e92bb | 234 | if (test_and_clear_bit(nodenum - 1, &c->local_nodes)) |
e1931b78 | 235 | return false; |
1da177e4 | 236 | |
e1931b78 | 237 | return true; |
1da177e4 | 238 | } |
76592584 | 239 | #endif |
1da177e4 LT |
240 | |
241 | static inline u_int32_t | |
a47362a2 JE |
242 | clusterip_hashfn(const struct sk_buff *skb, |
243 | const struct clusterip_config *config) | |
1da177e4 | 244 | { |
a47362a2 | 245 | const struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 246 | unsigned long hashval; |
3d04ebb6 CG |
247 | u_int16_t sport = 0, dport = 0; |
248 | int poff; | |
249 | ||
250 | poff = proto_ports_offset(iph->protocol); | |
251 | if (poff >= 0) { | |
252 | const u_int16_t *ports; | |
253 | u16 _ports[2]; | |
254 | ||
255 | ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports); | |
256 | if (ports) { | |
257 | sport = ports[0]; | |
258 | dport = ports[1]; | |
259 | } | |
260 | } else { | |
e87cc472 | 261 | net_info_ratelimited("unknown protocol %u\n", iph->protocol); |
1da177e4 LT |
262 | } |
263 | ||
264 | switch (config->hash_mode) { | |
265 | case CLUSTERIP_HASHMODE_SIP: | |
266 | hashval = jhash_1word(ntohl(iph->saddr), | |
267 | config->hash_initval); | |
268 | break; | |
269 | case CLUSTERIP_HASHMODE_SIP_SPT: | |
e905a9ed | 270 | hashval = jhash_2words(ntohl(iph->saddr), sport, |
1da177e4 LT |
271 | config->hash_initval); |
272 | break; | |
273 | case CLUSTERIP_HASHMODE_SIP_SPT_DPT: | |
274 | hashval = jhash_3words(ntohl(iph->saddr), sport, dport, | |
275 | config->hash_initval); | |
276 | break; | |
277 | default: | |
278 | /* to make gcc happy */ | |
279 | hashval = 0; | |
280 | /* This cannot happen, unless the check function wasn't called | |
281 | * at rule load time */ | |
ff67e4e4 | 282 | pr_info("unknown mode %u\n", config->hash_mode); |
1da177e4 LT |
283 | BUG(); |
284 | break; | |
285 | } | |
286 | ||
287 | /* node numbers are 1..n, not 0..n */ | |
8fc54f68 | 288 | return reciprocal_scale(hashval, config->num_total_nodes) + 1; |
1da177e4 LT |
289 | } |
290 | ||
291 | static inline int | |
a47362a2 | 292 | clusterip_responsible(const struct clusterip_config *config, u_int32_t hash) |
1da177e4 | 293 | { |
136e92bb | 294 | return test_bit(hash - 1, &config->local_nodes); |
1da177e4 LT |
295 | } |
296 | ||
e905a9ed YH |
297 | /*********************************************************************** |
298 | * IPTABLES TARGET | |
1da177e4 LT |
299 | ***********************************************************************/ |
300 | ||
301 | static unsigned int | |
4b560b44 | 302 | clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par) |
1da177e4 | 303 | { |
7eb35586 | 304 | const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
587aa641 | 305 | struct nf_conn *ct; |
1da177e4 | 306 | enum ip_conntrack_info ctinfo; |
587aa641 | 307 | u_int32_t hash; |
1da177e4 LT |
308 | |
309 | /* don't need to clusterip_config_get() here, since refcount | |
310 | * is only decremented by destroy() - and ip_tables guarantees | |
311 | * that the ->target() function isn't called after ->destroy() */ | |
312 | ||
3db05fea | 313 | ct = nf_ct_get(skb, &ctinfo); |
94d117a1 | 314 | if (ct == NULL) |
1da177e4 | 315 | return NF_DROP; |
1da177e4 LT |
316 | |
317 | /* special case: ICMP error handling. conntrack distinguishes between | |
318 | * error messages (RELATED) and information requests (see below) */ | |
3666ed1c JP |
319 | if (ip_hdr(skb)->protocol == IPPROTO_ICMP && |
320 | (ctinfo == IP_CT_RELATED || | |
fb048833 | 321 | ctinfo == IP_CT_RELATED_REPLY)) |
6709dbbb | 322 | return XT_CONTINUE; |
1da177e4 | 323 | |
e905a9ed | 324 | /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO, |
1da177e4 LT |
325 | * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here |
326 | * on, which all have an ID field [relevant for hashing]. */ | |
327 | ||
3db05fea | 328 | hash = clusterip_hashfn(skb, cipinfo->config); |
1da177e4 LT |
329 | |
330 | switch (ctinfo) { | |
181b1e9c JP |
331 | case IP_CT_NEW: |
332 | ct->mark = hash; | |
333 | break; | |
334 | case IP_CT_RELATED: | |
335 | case IP_CT_RELATED_REPLY: | |
336 | /* FIXME: we don't handle expectations at the moment. | |
337 | * They can arrive on a different node than | |
338 | * the master connection (e.g. FTP passive mode) */ | |
339 | case IP_CT_ESTABLISHED: | |
340 | case IP_CT_ESTABLISHED_REPLY: | |
341 | break; | |
342 | default: /* Prevent gcc warnings */ | |
343 | break; | |
1da177e4 LT |
344 | } |
345 | ||
0d53778e | 346 | #ifdef DEBUG |
3c9fba65 | 347 | nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
1da177e4 | 348 | #endif |
0d53778e | 349 | pr_debug("hash=%u ct_hash=%u ", hash, ct->mark); |
1da177e4 | 350 | if (!clusterip_responsible(cipinfo->config, hash)) { |
0d53778e | 351 | pr_debug("not responsible\n"); |
1da177e4 LT |
352 | return NF_DROP; |
353 | } | |
0d53778e | 354 | pr_debug("responsible\n"); |
1da177e4 LT |
355 | |
356 | /* despite being received via linklayer multicast, this is | |
357 | * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */ | |
3db05fea | 358 | skb->pkt_type = PACKET_HOST; |
1da177e4 | 359 | |
6709dbbb | 360 | return XT_CONTINUE; |
1da177e4 LT |
361 | } |
362 | ||
135367b8 | 363 | static int clusterip_tg_check(const struct xt_tgchk_param *par) |
1da177e4 | 364 | { |
af5d6dc2 JE |
365 | struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
366 | const struct ipt_entry *e = par->entryinfo; | |
1da177e4 | 367 | struct clusterip_config *config; |
4a5a5c73 | 368 | int ret; |
1da177e4 | 369 | |
55917a21 PNA |
370 | if (par->nft_compat) { |
371 | pr_err("cannot use CLUSTERIP target from nftables compat\n"); | |
372 | return -EOPNOTSUPP; | |
373 | } | |
374 | ||
1da177e4 LT |
375 | if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP && |
376 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT && | |
377 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) { | |
ff67e4e4 | 378 | pr_info("unknown mode %u\n", cipinfo->hash_mode); |
d6b00a53 | 379 | return -EINVAL; |
1da177e4 LT |
380 | |
381 | } | |
3666ed1c JP |
382 | if (e->ip.dmsk.s_addr != htonl(0xffffffff) || |
383 | e->ip.dst.s_addr == 0) { | |
ff67e4e4 | 384 | pr_info("Please specify destination IP\n"); |
d6b00a53 | 385 | return -EINVAL; |
1da177e4 LT |
386 | } |
387 | ||
388 | /* FIXME: further sanity checks */ | |
389 | ||
d86946d2 | 390 | config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1); |
d3c3f424 | 391 | if (!config) { |
1da177e4 | 392 | if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) { |
ff67e4e4 JE |
393 | pr_info("no config found for %pI4, need 'new'\n", |
394 | &e->ip.dst.s_addr); | |
d6b00a53 | 395 | return -EINVAL; |
1da177e4 LT |
396 | } else { |
397 | struct net_device *dev; | |
398 | ||
399 | if (e->ip.iniface[0] == '\0') { | |
ff67e4e4 | 400 | pr_info("Please specify an interface name\n"); |
d6b00a53 | 401 | return -EINVAL; |
1da177e4 LT |
402 | } |
403 | ||
d86946d2 | 404 | dev = dev_get_by_name(par->net, e->ip.iniface); |
1da177e4 | 405 | if (!dev) { |
ff67e4e4 JE |
406 | pr_info("no such interface %s\n", |
407 | e->ip.iniface); | |
4a5a5c73 | 408 | return -ENOENT; |
1da177e4 LT |
409 | } |
410 | ||
e905a9ed | 411 | config = clusterip_config_init(cipinfo, |
1da177e4 LT |
412 | e->ip.dst.s_addr, dev); |
413 | if (!config) { | |
1da177e4 | 414 | dev_put(dev); |
4a5a5c73 | 415 | return -ENOMEM; |
1da177e4 | 416 | } |
22bedad3 | 417 | dev_mc_add(config->dev, config->clustermac); |
1da177e4 LT |
418 | } |
419 | } | |
d3c3f424 | 420 | cipinfo->config = config; |
1da177e4 | 421 | |
4a5a5c73 | 422 | ret = nf_ct_l3proto_try_module_get(par->family); |
f95c74e3 | 423 | if (ret < 0) |
ff67e4e4 JE |
424 | pr_info("cannot load conntrack support for proto=%u\n", |
425 | par->family); | |
43270b1b PNA |
426 | |
427 | if (!par->net->xt.clusterip_deprecated_warning) { | |
428 | pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, " | |
429 | "use xt_cluster instead\n"); | |
430 | par->net->xt.clusterip_deprecated_warning = true; | |
431 | } | |
432 | ||
f95c74e3 | 433 | return ret; |
1da177e4 LT |
434 | } |
435 | ||
436 | /* drop reference count of cluster config when rule is deleted */ | |
a2df1648 | 437 | static void clusterip_tg_destroy(const struct xt_tgdtor_param *par) |
1da177e4 | 438 | { |
a2df1648 | 439 | const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
1da177e4 | 440 | |
44513624 KK |
441 | /* if no more entries are referencing the config, remove it |
442 | * from the list and destroy the proc entry */ | |
443 | clusterip_config_entry_put(cipinfo->config); | |
444 | ||
1da177e4 | 445 | clusterip_config_put(cipinfo->config); |
11078c37 | 446 | |
0d345455 | 447 | nf_ct_l3proto_module_put(par->family); |
1da177e4 LT |
448 | } |
449 | ||
d3c3f424 PM |
450 | #ifdef CONFIG_COMPAT |
451 | struct compat_ipt_clusterip_tgt_info | |
452 | { | |
453 | u_int32_t flags; | |
454 | u_int8_t clustermac[6]; | |
455 | u_int16_t num_total_nodes; | |
456 | u_int16_t num_local_nodes; | |
457 | u_int16_t local_nodes[CLUSTERIP_MAX_NODES]; | |
458 | u_int32_t hash_mode; | |
459 | u_int32_t hash_initval; | |
460 | compat_uptr_t config; | |
461 | }; | |
462 | #endif /* CONFIG_COMPAT */ | |
463 | ||
d3c5ee6d | 464 | static struct xt_target clusterip_tg_reg __read_mostly = { |
1d5cd909 | 465 | .name = "CLUSTERIP", |
ee999d8b | 466 | .family = NFPROTO_IPV4, |
d3c5ee6d JE |
467 | .target = clusterip_tg, |
468 | .checkentry = clusterip_tg_check, | |
469 | .destroy = clusterip_tg_destroy, | |
d3c3f424 PM |
470 | .targetsize = sizeof(struct ipt_clusterip_tgt_info), |
471 | #ifdef CONFIG_COMPAT | |
472 | .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info), | |
473 | #endif /* CONFIG_COMPAT */ | |
1d5cd909 | 474 | .me = THIS_MODULE |
1da177e4 LT |
475 | }; |
476 | ||
477 | ||
e905a9ed YH |
478 | /*********************************************************************** |
479 | * ARP MANGLING CODE | |
1da177e4 LT |
480 | ***********************************************************************/ |
481 | ||
482 | /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */ | |
483 | struct arp_payload { | |
484 | u_int8_t src_hw[ETH_ALEN]; | |
6a19d614 | 485 | __be32 src_ip; |
1da177e4 | 486 | u_int8_t dst_hw[ETH_ALEN]; |
6a19d614 | 487 | __be32 dst_ip; |
3f30fc15 | 488 | } __packed; |
1da177e4 | 489 | |
0d53778e | 490 | #ifdef DEBUG |
e905a9ed | 491 | static void arp_print(struct arp_payload *payload) |
1da177e4 LT |
492 | { |
493 | #define HBUFFERLEN 30 | |
494 | char hbuffer[HBUFFERLEN]; | |
c8d71d08 | 495 | int j, k; |
1da177e4 | 496 | |
c8d71d08 | 497 | for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) { |
6a8341b6 HH |
498 | hbuffer[k++] = hex_asc_hi(payload->src_hw[j]); |
499 | hbuffer[k++] = hex_asc_lo(payload->src_hw[j]); | |
c8d71d08 | 500 | hbuffer[k++] = ':'; |
1da177e4 | 501 | } |
c8d71d08 | 502 | hbuffer[--k] = '\0'; |
1da177e4 | 503 | |
ff67e4e4 JE |
504 | pr_debug("src %pI4@%s, dst %pI4\n", |
505 | &payload->src_ip, hbuffer, &payload->dst_ip); | |
1da177e4 LT |
506 | } |
507 | #endif | |
508 | ||
509 | static unsigned int | |
06198b34 | 510 | arp_mangle(void *priv, |
3db05fea | 511 | struct sk_buff *skb, |
238e54c9 | 512 | const struct nf_hook_state *state) |
1da177e4 | 513 | { |
3db05fea | 514 | struct arphdr *arp = arp_hdr(skb); |
1da177e4 LT |
515 | struct arp_payload *payload; |
516 | struct clusterip_config *c; | |
9dff2c96 | 517 | struct net *net = state->net; |
1da177e4 LT |
518 | |
519 | /* we don't care about non-ethernet and non-ipv4 ARP */ | |
3666ed1c JP |
520 | if (arp->ar_hrd != htons(ARPHRD_ETHER) || |
521 | arp->ar_pro != htons(ETH_P_IP) || | |
522 | arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN) | |
1da177e4 LT |
523 | return NF_ACCEPT; |
524 | ||
4095ebf1 | 525 | /* we only want to mangle arp requests and replies */ |
3666ed1c JP |
526 | if (arp->ar_op != htons(ARPOP_REPLY) && |
527 | arp->ar_op != htons(ARPOP_REQUEST)) | |
1da177e4 LT |
528 | return NF_ACCEPT; |
529 | ||
530 | payload = (void *)(arp+1); | |
531 | ||
e905a9ed | 532 | /* if there is no clusterip configuration for the arp reply's |
1da177e4 | 533 | * source ip, we don't want to mangle it */ |
d86946d2 | 534 | c = clusterip_config_find_get(net, payload->src_ip, 0); |
1da177e4 LT |
535 | if (!c) |
536 | return NF_ACCEPT; | |
537 | ||
e905a9ed | 538 | /* normally the linux kernel always replies to arp queries of |
1da177e4 LT |
539 | * addresses on different interfacs. However, in the CLUSTERIP case |
540 | * this wouldn't work, since we didn't subscribe the mcast group on | |
541 | * other interfaces */ | |
238e54c9 | 542 | if (c->dev != state->out) { |
ff67e4e4 | 543 | pr_debug("not mangling arp reply on different " |
0d53778e | 544 | "interface: cip'%s'-skb'%s'\n", |
238e54c9 | 545 | c->dev->name, state->out->name); |
1da177e4 LT |
546 | clusterip_config_put(c); |
547 | return NF_ACCEPT; | |
548 | } | |
549 | ||
550 | /* mangle reply hardware address */ | |
551 | memcpy(payload->src_hw, c->clustermac, arp->ar_hln); | |
552 | ||
0d53778e | 553 | #ifdef DEBUG |
ff67e4e4 | 554 | pr_debug("mangled arp reply: "); |
1da177e4 LT |
555 | arp_print(payload); |
556 | #endif | |
557 | ||
558 | clusterip_config_put(c); | |
559 | ||
560 | return NF_ACCEPT; | |
561 | } | |
562 | ||
1999414a | 563 | static struct nf_hook_ops cip_arp_ops __read_mostly = { |
1da177e4 | 564 | .hook = arp_mangle, |
ee999d8b | 565 | .pf = NFPROTO_ARP, |
1da177e4 LT |
566 | .hooknum = NF_ARP_OUT, |
567 | .priority = -1 | |
568 | }; | |
569 | ||
e905a9ed YH |
570 | /*********************************************************************** |
571 | * PROC DIR HANDLING | |
1da177e4 LT |
572 | ***********************************************************************/ |
573 | ||
574 | #ifdef CONFIG_PROC_FS | |
575 | ||
136e92bb KK |
576 | struct clusterip_seq_position { |
577 | unsigned int pos; /* position */ | |
578 | unsigned int weight; /* number of bits set == size */ | |
579 | unsigned int bit; /* current bit */ | |
580 | unsigned long val; /* current value */ | |
581 | }; | |
582 | ||
1da177e4 LT |
583 | static void *clusterip_seq_start(struct seq_file *s, loff_t *pos) |
584 | { | |
47778147 | 585 | struct clusterip_config *c = s->private; |
136e92bb KK |
586 | unsigned int weight; |
587 | u_int32_t local_nodes; | |
588 | struct clusterip_seq_position *idx; | |
589 | ||
590 | /* FIXME: possible race */ | |
591 | local_nodes = c->local_nodes; | |
592 | weight = hweight32(local_nodes); | |
593 | if (*pos >= weight) | |
1da177e4 LT |
594 | return NULL; |
595 | ||
136e92bb KK |
596 | idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL); |
597 | if (!idx) | |
1da177e4 LT |
598 | return ERR_PTR(-ENOMEM); |
599 | ||
136e92bb KK |
600 | idx->pos = *pos; |
601 | idx->weight = weight; | |
602 | idx->bit = ffs(local_nodes); | |
603 | idx->val = local_nodes; | |
604 | clear_bit(idx->bit - 1, &idx->val); | |
605 | ||
606 | return idx; | |
1da177e4 LT |
607 | } |
608 | ||
609 | static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos) | |
610 | { | |
3cf93c96 | 611 | struct clusterip_seq_position *idx = v; |
1da177e4 | 612 | |
136e92bb KK |
613 | *pos = ++idx->pos; |
614 | if (*pos >= idx->weight) { | |
1da177e4 LT |
615 | kfree(v); |
616 | return NULL; | |
617 | } | |
136e92bb KK |
618 | idx->bit = ffs(idx->val); |
619 | clear_bit(idx->bit - 1, &idx->val); | |
620 | return idx; | |
1da177e4 LT |
621 | } |
622 | ||
623 | static void clusterip_seq_stop(struct seq_file *s, void *v) | |
624 | { | |
902a3dd5 ED |
625 | if (!IS_ERR(v)) |
626 | kfree(v); | |
1da177e4 LT |
627 | } |
628 | ||
629 | static int clusterip_seq_show(struct seq_file *s, void *v) | |
630 | { | |
3cf93c96 | 631 | struct clusterip_seq_position *idx = v; |
1da177e4 | 632 | |
e905a9ed | 633 | if (idx->pos != 0) |
1da177e4 | 634 | seq_putc(s, ','); |
1da177e4 | 635 | |
136e92bb KK |
636 | seq_printf(s, "%u", idx->bit); |
637 | ||
638 | if (idx->pos == idx->weight - 1) | |
1da177e4 LT |
639 | seq_putc(s, '\n'); |
640 | ||
641 | return 0; | |
642 | } | |
643 | ||
56b3d975 | 644 | static const struct seq_operations clusterip_seq_ops = { |
1da177e4 LT |
645 | .start = clusterip_seq_start, |
646 | .next = clusterip_seq_next, | |
647 | .stop = clusterip_seq_stop, | |
648 | .show = clusterip_seq_show, | |
649 | }; | |
650 | ||
651 | static int clusterip_proc_open(struct inode *inode, struct file *file) | |
652 | { | |
653 | int ret = seq_open(file, &clusterip_seq_ops); | |
654 | ||
655 | if (!ret) { | |
656 | struct seq_file *sf = file->private_data; | |
d9dda78b | 657 | struct clusterip_config *c = PDE_DATA(inode); |
1da177e4 | 658 | |
47778147 | 659 | sf->private = c; |
1da177e4 LT |
660 | |
661 | clusterip_config_get(c); | |
662 | } | |
663 | ||
664 | return ret; | |
665 | } | |
666 | ||
667 | static int clusterip_proc_release(struct inode *inode, struct file *file) | |
668 | { | |
d9dda78b | 669 | struct clusterip_config *c = PDE_DATA(inode); |
1da177e4 LT |
670 | int ret; |
671 | ||
672 | ret = seq_release(inode, file); | |
673 | ||
674 | if (!ret) | |
675 | clusterip_config_put(c); | |
676 | ||
677 | return ret; | |
678 | } | |
679 | ||
680 | static ssize_t clusterip_proc_write(struct file *file, const char __user *input, | |
681 | size_t size, loff_t *ofs) | |
682 | { | |
d9dda78b | 683 | struct clusterip_config *c = PDE_DATA(file_inode(file)); |
1da177e4 LT |
684 | #define PROC_WRITELEN 10 |
685 | char buffer[PROC_WRITELEN+1]; | |
1da177e4 | 686 | unsigned long nodenum; |
4b5511eb | 687 | int rc; |
1da177e4 | 688 | |
961ed183 VK |
689 | if (size > PROC_WRITELEN) |
690 | return -EIO; | |
691 | if (copy_from_user(buffer, input, size)) | |
1da177e4 | 692 | return -EFAULT; |
961ed183 | 693 | buffer[size] = 0; |
1da177e4 LT |
694 | |
695 | if (*buffer == '+') { | |
4b5511eb AP |
696 | rc = kstrtoul(buffer+1, 10, &nodenum); |
697 | if (rc) | |
698 | return rc; | |
1da177e4 LT |
699 | if (clusterip_add_node(c, nodenum)) |
700 | return -ENOMEM; | |
701 | } else if (*buffer == '-') { | |
4b5511eb AP |
702 | rc = kstrtoul(buffer+1, 10, &nodenum); |
703 | if (rc) | |
704 | return rc; | |
1da177e4 LT |
705 | if (clusterip_del_node(c, nodenum)) |
706 | return -ENOENT; | |
707 | } else | |
708 | return -EIO; | |
709 | ||
710 | return size; | |
711 | } | |
712 | ||
9a32144e | 713 | static const struct file_operations clusterip_proc_fops = { |
1da177e4 LT |
714 | .owner = THIS_MODULE, |
715 | .open = clusterip_proc_open, | |
716 | .read = seq_read, | |
717 | .write = clusterip_proc_write, | |
718 | .llseek = seq_lseek, | |
719 | .release = clusterip_proc_release, | |
720 | }; | |
721 | ||
722 | #endif /* CONFIG_PROC_FS */ | |
723 | ||
ce4ff76c G |
724 | static int clusterip_net_init(struct net *net) |
725 | { | |
ce4ff76c G |
726 | struct clusterip_net *cn = net_generic(net, clusterip_net_id); |
727 | ||
26a89e43 G |
728 | INIT_LIST_HEAD(&cn->configs); |
729 | ||
f1e8077f G |
730 | spin_lock_init(&cn->lock); |
731 | ||
26a89e43 | 732 | #ifdef CONFIG_PROC_FS |
ce4ff76c G |
733 | cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net); |
734 | if (!cn->procdir) { | |
735 | pr_err("Unable to proc dir entry\n"); | |
736 | return -ENOMEM; | |
737 | } | |
738 | #endif /* CONFIG_PROC_FS */ | |
739 | ||
740 | return 0; | |
741 | } | |
742 | ||
743 | static void clusterip_net_exit(struct net *net) | |
744 | { | |
745 | #ifdef CONFIG_PROC_FS | |
746 | struct clusterip_net *cn = net_generic(net, clusterip_net_id); | |
747 | proc_remove(cn->procdir); | |
748 | #endif | |
749 | } | |
750 | ||
751 | static struct pernet_operations clusterip_net_ops = { | |
752 | .init = clusterip_net_init, | |
753 | .exit = clusterip_net_exit, | |
754 | .id = &clusterip_net_id, | |
755 | .size = sizeof(struct clusterip_net), | |
756 | }; | |
757 | ||
d3c5ee6d | 758 | static int __init clusterip_tg_init(void) |
1da177e4 LT |
759 | { |
760 | int ret; | |
761 | ||
ce4ff76c | 762 | ret = register_pernet_subsys(&clusterip_net_ops); |
32292a7f PM |
763 | if (ret < 0) |
764 | return ret; | |
1da177e4 | 765 | |
ce4ff76c G |
766 | ret = xt_register_target(&clusterip_tg_reg); |
767 | if (ret < 0) | |
768 | goto cleanup_subsys; | |
769 | ||
32292a7f PM |
770 | ret = nf_register_hook(&cip_arp_ops); |
771 | if (ret < 0) | |
1da177e4 | 772 | goto cleanup_target; |
1da177e4 | 773 | |
ff67e4e4 | 774 | pr_info("ClusterIP Version %s loaded successfully\n", |
1da177e4 | 775 | CLUSTERIP_VERSION); |
ce4ff76c | 776 | |
1da177e4 LT |
777 | return 0; |
778 | ||
1da177e4 | 779 | cleanup_target: |
d3c5ee6d | 780 | xt_unregister_target(&clusterip_tg_reg); |
ce4ff76c G |
781 | cleanup_subsys: |
782 | unregister_pernet_subsys(&clusterip_net_ops); | |
32292a7f | 783 | return ret; |
1da177e4 LT |
784 | } |
785 | ||
d3c5ee6d | 786 | static void __exit clusterip_tg_exit(void) |
1da177e4 | 787 | { |
ff67e4e4 | 788 | pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION); |
ce4ff76c | 789 | |
32292a7f | 790 | nf_unregister_hook(&cip_arp_ops); |
d3c5ee6d | 791 | xt_unregister_target(&clusterip_tg_reg); |
ce4ff76c | 792 | unregister_pernet_subsys(&clusterip_net_ops); |
d73f33b1 ED |
793 | |
794 | /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */ | |
795 | rcu_barrier_bh(); | |
1da177e4 LT |
796 | } |
797 | ||
d3c5ee6d JE |
798 | module_init(clusterip_tg_init); |
799 | module_exit(clusterip_tg_exit); |