Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 16 | #include <linux/bitops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/skbuff.h> |
| 18 | #include <linux/ip.h> |
| 19 | #include <linux/tcp.h> |
| 20 | #include <linux/udp.h> |
| 21 | #include <linux/icmp.h> |
| 22 | #include <linux/if_arp.h> |
| 23 | #include <linux/proc_fs.h> |
| 24 | #include <linux/seq_file.h> |
| 25 | |
| 26 | #include <net/checksum.h> |
| 27 | |
| 28 | #include <linux/netfilter_arp.h> |
| 29 | |
| 30 | #include <linux/netfilter_ipv4/ip_tables.h> |
| 31 | #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h> |
Yasuyuki Kozakai | 9fb9cbb | 2005-11-09 16:38:16 -0800 | [diff] [blame] | 32 | #include <net/netfilter/nf_conntrack_compat.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 34 | #define CLUSTERIP_VERSION "0.8" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | |
| 36 | #define DEBUG_CLUSTERIP |
| 37 | |
| 38 | #ifdef DEBUG_CLUSTERIP |
| 39 | #define DEBUGP printk |
| 40 | #else |
| 41 | #define DEBUGP |
| 42 | #endif |
| 43 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 44 | #define ASSERT_READ_LOCK(x) |
| 45 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | MODULE_LICENSE("GPL"); |
| 47 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); |
| 48 | MODULE_DESCRIPTION("iptables target for CLUSTERIP"); |
| 49 | |
| 50 | struct clusterip_config { |
| 51 | struct list_head list; /* list of all configs */ |
| 52 | atomic_t refcount; /* reference count */ |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 53 | atomic_t entries; /* number of entries/rules |
| 54 | * referencing us */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | |
| 56 | u_int32_t clusterip; /* the IP address */ |
| 57 | u_int8_t clustermac[ETH_ALEN]; /* the MAC address */ |
| 58 | struct net_device *dev; /* device */ |
| 59 | u_int16_t num_total_nodes; /* total number of nodes */ |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 60 | unsigned long local_nodes; /* node number array */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
| 62 | #ifdef CONFIG_PROC_FS |
| 63 | struct proc_dir_entry *pde; /* proc dir entry */ |
| 64 | #endif |
| 65 | enum clusterip_hashmode hash_mode; /* which hashing mode */ |
| 66 | u_int32_t hash_initval; /* hash initialization */ |
| 67 | }; |
| 68 | |
| 69 | static LIST_HEAD(clusterip_configs); |
| 70 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 71 | /* clusterip_lock protects the clusterip_configs list */ |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 72 | static DEFINE_RWLOCK(clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | |
| 74 | #ifdef CONFIG_PROC_FS |
| 75 | static struct file_operations clusterip_proc_fops; |
| 76 | static struct proc_dir_entry *clusterip_procdir; |
| 77 | #endif |
| 78 | |
| 79 | static inline void |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 80 | clusterip_config_get(struct clusterip_config *c) |
| 81 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | atomic_inc(&c->refcount); |
| 83 | } |
| 84 | |
| 85 | static inline void |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 86 | clusterip_config_put(struct clusterip_config *c) |
| 87 | { |
| 88 | if (atomic_dec_and_test(&c->refcount)) |
| 89 | kfree(c); |
| 90 | } |
| 91 | |
| 92 | /* increase the count of entries(rules) using/referencing this config */ |
| 93 | static inline void |
| 94 | clusterip_config_entry_get(struct clusterip_config *c) |
| 95 | { |
| 96 | atomic_inc(&c->entries); |
| 97 | } |
| 98 | |
| 99 | /* decrease the count of entries using/referencing this config. If last |
| 100 | * entry(rule) is removed, remove the config from lists, but don't free it |
| 101 | * yet, since proc-files could still be holding references */ |
| 102 | static inline void |
| 103 | clusterip_config_entry_put(struct clusterip_config *c) |
| 104 | { |
| 105 | if (atomic_dec_and_test(&c->entries)) { |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 106 | write_lock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | list_del(&c->list); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 108 | write_unlock_bh(&clusterip_lock); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 109 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0); |
| 111 | dev_put(c->dev); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 112 | |
| 113 | /* In case anyone still accesses the file, the open/close |
| 114 | * functions are also incrementing the refcount on their own, |
| 115 | * so it's safe to remove the entry even if it's in use. */ |
| 116 | #ifdef CONFIG_PROC_FS |
| 117 | remove_proc_entry(c->pde->name, c->pde->parent); |
| 118 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | } |
| 120 | } |
| 121 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | static struct clusterip_config * |
| 123 | __clusterip_config_find(u_int32_t clusterip) |
| 124 | { |
| 125 | struct list_head *pos; |
| 126 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 127 | ASSERT_READ_LOCK(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | list_for_each(pos, &clusterip_configs) { |
| 129 | struct clusterip_config *c = list_entry(pos, |
| 130 | struct clusterip_config, list); |
| 131 | if (c->clusterip == clusterip) { |
| 132 | return c; |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | return NULL; |
| 137 | } |
| 138 | |
| 139 | static inline struct clusterip_config * |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 140 | clusterip_config_find_get(u_int32_t clusterip, int entry) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | { |
| 142 | struct clusterip_config *c; |
| 143 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 144 | read_lock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 145 | c = __clusterip_config_find(clusterip); |
| 146 | if (!c) { |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 147 | read_unlock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | return NULL; |
| 149 | } |
| 150 | atomic_inc(&c->refcount); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 151 | if (entry) |
| 152 | atomic_inc(&c->entries); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 153 | read_unlock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | |
| 155 | return c; |
| 156 | } |
| 157 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 158 | static void |
| 159 | clusterip_config_init_nodelist(struct clusterip_config *c, |
| 160 | const struct ipt_clusterip_tgt_info *i) |
| 161 | { |
| 162 | int n; |
| 163 | |
| 164 | for (n = 0; n < i->num_local_nodes; n++) { |
| 165 | set_bit(i->local_nodes[n] - 1, &c->local_nodes); |
| 166 | } |
| 167 | } |
| 168 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 169 | static struct clusterip_config * |
| 170 | clusterip_config_init(struct ipt_clusterip_tgt_info *i, u_int32_t ip, |
| 171 | struct net_device *dev) |
| 172 | { |
| 173 | struct clusterip_config *c; |
| 174 | char buffer[16]; |
| 175 | |
| 176 | c = kmalloc(sizeof(*c), GFP_ATOMIC); |
| 177 | if (!c) |
| 178 | return NULL; |
| 179 | |
| 180 | memset(c, 0, sizeof(*c)); |
| 181 | c->dev = dev; |
| 182 | c->clusterip = ip; |
| 183 | memcpy(&c->clustermac, &i->clustermac, ETH_ALEN); |
| 184 | c->num_total_nodes = i->num_total_nodes; |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 185 | clusterip_config_init_nodelist(c, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | c->hash_mode = i->hash_mode; |
| 187 | c->hash_initval = i->hash_initval; |
| 188 | atomic_set(&c->refcount, 1); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 189 | atomic_set(&c->entries, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
| 191 | #ifdef CONFIG_PROC_FS |
| 192 | /* create proc dir entry */ |
| 193 | sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip)); |
| 194 | c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR, clusterip_procdir); |
| 195 | if (!c->pde) { |
| 196 | kfree(c); |
| 197 | return NULL; |
| 198 | } |
| 199 | c->pde->proc_fops = &clusterip_proc_fops; |
| 200 | c->pde->data = c; |
| 201 | #endif |
| 202 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 203 | write_lock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | list_add(&c->list, &clusterip_configs); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 205 | write_unlock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | |
| 207 | return c; |
| 208 | } |
| 209 | |
| 210 | static int |
| 211 | clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum) |
| 212 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 214 | if (nodenum == 0 || |
| 215 | nodenum > c->num_total_nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 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; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 222 | return 0; |
| 223 | } |
| 224 | |
| 225 | static int |
| 226 | clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum) |
| 227 | { |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 228 | if (nodenum == 0 || |
| 229 | nodenum > c->num_total_nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 232 | if (test_and_clear_bit(nodenum - 1, &c->local_nodes)) |
| 233 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 234 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | return 1; |
| 236 | } |
| 237 | |
| 238 | static inline u_int32_t |
| 239 | clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config) |
| 240 | { |
| 241 | struct iphdr *iph = skb->nh.iph; |
| 242 | unsigned long hashval; |
| 243 | u_int16_t sport, dport; |
| 244 | struct tcphdr *th; |
| 245 | struct udphdr *uh; |
| 246 | struct icmphdr *ih; |
| 247 | |
| 248 | switch (iph->protocol) { |
| 249 | case IPPROTO_TCP: |
| 250 | th = (void *)iph+iph->ihl*4; |
| 251 | sport = ntohs(th->source); |
| 252 | dport = ntohs(th->dest); |
| 253 | break; |
| 254 | case IPPROTO_UDP: |
| 255 | uh = (void *)iph+iph->ihl*4; |
| 256 | sport = ntohs(uh->source); |
| 257 | dport = ntohs(uh->dest); |
| 258 | break; |
| 259 | case IPPROTO_ICMP: |
| 260 | ih = (void *)iph+iph->ihl*4; |
| 261 | sport = ntohs(ih->un.echo.id); |
| 262 | dport = (ih->type<<8)|ih->code; |
| 263 | break; |
| 264 | default: |
| 265 | if (net_ratelimit()) { |
| 266 | printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n", |
| 267 | iph->protocol); |
| 268 | } |
| 269 | sport = dport = 0; |
| 270 | } |
| 271 | |
| 272 | switch (config->hash_mode) { |
| 273 | case CLUSTERIP_HASHMODE_SIP: |
| 274 | hashval = jhash_1word(ntohl(iph->saddr), |
| 275 | config->hash_initval); |
| 276 | break; |
| 277 | case CLUSTERIP_HASHMODE_SIP_SPT: |
| 278 | hashval = jhash_2words(ntohl(iph->saddr), sport, |
| 279 | config->hash_initval); |
| 280 | break; |
| 281 | case CLUSTERIP_HASHMODE_SIP_SPT_DPT: |
| 282 | hashval = jhash_3words(ntohl(iph->saddr), sport, dport, |
| 283 | config->hash_initval); |
| 284 | break; |
| 285 | default: |
| 286 | /* to make gcc happy */ |
| 287 | hashval = 0; |
| 288 | /* This cannot happen, unless the check function wasn't called |
| 289 | * at rule load time */ |
| 290 | printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode); |
| 291 | BUG(); |
| 292 | break; |
| 293 | } |
| 294 | |
| 295 | /* node numbers are 1..n, not 0..n */ |
| 296 | return ((hashval % config->num_total_nodes)+1); |
| 297 | } |
| 298 | |
| 299 | static inline int |
| 300 | clusterip_responsible(struct clusterip_config *config, u_int32_t hash) |
| 301 | { |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 302 | return test_bit(hash - 1, &config->local_nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | } |
| 304 | |
| 305 | /*********************************************************************** |
| 306 | * IPTABLES TARGET |
| 307 | ***********************************************************************/ |
| 308 | |
| 309 | static unsigned int |
| 310 | target(struct sk_buff **pskb, |
| 311 | const struct net_device *in, |
| 312 | const struct net_device *out, |
| 313 | unsigned int hooknum, |
| 314 | const void *targinfo, |
| 315 | void *userinfo) |
| 316 | { |
| 317 | const struct ipt_clusterip_tgt_info *cipinfo = targinfo; |
| 318 | enum ip_conntrack_info ctinfo; |
Yasuyuki Kozakai | 9fb9cbb | 2005-11-09 16:38:16 -0800 | [diff] [blame] | 319 | u_int32_t *mark, hash; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | |
| 321 | /* don't need to clusterip_config_get() here, since refcount |
| 322 | * is only decremented by destroy() - and ip_tables guarantees |
| 323 | * that the ->target() function isn't called after ->destroy() */ |
| 324 | |
Yasuyuki Kozakai | 9fb9cbb | 2005-11-09 16:38:16 -0800 | [diff] [blame] | 325 | mark = nf_ct_get_mark((*pskb), &ctinfo); |
| 326 | if (mark == NULL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 327 | printk(KERN_ERR "CLUSTERIP: no conntrack!\n"); |
| 328 | /* FIXME: need to drop invalid ones, since replies |
| 329 | * to outgoing connections of other nodes will be |
| 330 | * marked as INVALID */ |
| 331 | return NF_DROP; |
| 332 | } |
| 333 | |
| 334 | /* special case: ICMP error handling. conntrack distinguishes between |
| 335 | * error messages (RELATED) and information requests (see below) */ |
| 336 | if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP |
| 337 | && (ctinfo == IP_CT_RELATED |
Harald Welte | 5d927eb | 2005-06-22 12:37:50 -0700 | [diff] [blame] | 338 | || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | return IPT_CONTINUE; |
| 340 | |
| 341 | /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO, |
| 342 | * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here |
| 343 | * on, which all have an ID field [relevant for hashing]. */ |
| 344 | |
| 345 | hash = clusterip_hashfn(*pskb, cipinfo->config); |
| 346 | |
| 347 | switch (ctinfo) { |
| 348 | case IP_CT_NEW: |
Yasuyuki Kozakai | 9fb9cbb | 2005-11-09 16:38:16 -0800 | [diff] [blame] | 349 | *mark = hash; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | break; |
| 351 | case IP_CT_RELATED: |
| 352 | case IP_CT_RELATED+IP_CT_IS_REPLY: |
| 353 | /* FIXME: we don't handle expectations at the |
| 354 | * moment. they can arrive on a different node than |
| 355 | * the master connection (e.g. FTP passive mode) */ |
| 356 | case IP_CT_ESTABLISHED: |
| 357 | case IP_CT_ESTABLISHED+IP_CT_IS_REPLY: |
| 358 | break; |
| 359 | default: |
| 360 | break; |
| 361 | } |
| 362 | |
| 363 | #ifdef DEBUG_CLUSTERP |
| 364 | DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
| 365 | #endif |
Yasuyuki Kozakai | 9fb9cbb | 2005-11-09 16:38:16 -0800 | [diff] [blame] | 366 | DEBUGP("hash=%u ct_hash=%u ", hash, *mark); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | if (!clusterip_responsible(cipinfo->config, hash)) { |
| 368 | DEBUGP("not responsible\n"); |
| 369 | return NF_DROP; |
| 370 | } |
| 371 | DEBUGP("responsible\n"); |
| 372 | |
| 373 | /* despite being received via linklayer multicast, this is |
| 374 | * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */ |
| 375 | (*pskb)->pkt_type = PACKET_HOST; |
| 376 | |
| 377 | return IPT_CONTINUE; |
| 378 | } |
| 379 | |
| 380 | static int |
| 381 | checkentry(const char *tablename, |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame^] | 382 | const void *e_void, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | void *targinfo, |
| 384 | unsigned int targinfosize, |
| 385 | unsigned int hook_mask) |
| 386 | { |
| 387 | struct ipt_clusterip_tgt_info *cipinfo = targinfo; |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame^] | 388 | const struct ipt_entry *e = e_void; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | |
| 390 | struct clusterip_config *config; |
| 391 | |
| 392 | if (targinfosize != IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info))) { |
| 393 | printk(KERN_WARNING "CLUSTERIP: targinfosize %u != %Zu\n", |
| 394 | targinfosize, |
| 395 | IPT_ALIGN(sizeof(struct ipt_clusterip_tgt_info))); |
| 396 | return 0; |
| 397 | } |
| 398 | |
| 399 | if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP && |
| 400 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT && |
| 401 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) { |
| 402 | printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n", |
| 403 | cipinfo->hash_mode); |
| 404 | return 0; |
| 405 | |
| 406 | } |
| 407 | if (e->ip.dmsk.s_addr != 0xffffffff |
| 408 | || e->ip.dst.s_addr == 0) { |
| 409 | printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n"); |
| 410 | return 0; |
| 411 | } |
| 412 | |
| 413 | /* FIXME: further sanity checks */ |
| 414 | |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 415 | config = clusterip_config_find_get(e->ip.dst.s_addr, 1); |
| 416 | if (config) { |
| 417 | if (cipinfo->config != NULL) { |
| 418 | /* Case A: This is an entry that gets reloaded, since |
| 419 | * it still has a cipinfo->config pointer. Simply |
| 420 | * increase the entry refcount and return */ |
| 421 | if (cipinfo->config != config) { |
| 422 | printk(KERN_ERR "CLUSTERIP: Reloaded entry " |
| 423 | "has invalid config pointer!\n"); |
| 424 | return 0; |
| 425 | } |
| 426 | clusterip_config_entry_get(cipinfo->config); |
| 427 | } else { |
| 428 | /* Case B: This is a new rule referring to an existing |
| 429 | * clusterip config. */ |
| 430 | cipinfo->config = config; |
| 431 | clusterip_config_entry_get(cipinfo->config); |
| 432 | } |
| 433 | } else { |
| 434 | /* Case C: This is a completely new clusterip config */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 435 | if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) { |
| 436 | printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr)); |
| 437 | return 0; |
| 438 | } else { |
| 439 | struct net_device *dev; |
| 440 | |
| 441 | if (e->ip.iniface[0] == '\0') { |
| 442 | printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n"); |
| 443 | return 0; |
| 444 | } |
| 445 | |
| 446 | dev = dev_get_by_name(e->ip.iniface); |
| 447 | if (!dev) { |
| 448 | printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface); |
| 449 | return 0; |
| 450 | } |
| 451 | |
| 452 | config = clusterip_config_init(cipinfo, |
| 453 | e->ip.dst.s_addr, dev); |
| 454 | if (!config) { |
| 455 | printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n"); |
| 456 | dev_put(dev); |
| 457 | return 0; |
| 458 | } |
| 459 | dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0); |
| 460 | } |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 461 | cipinfo->config = config; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | } |
| 463 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | return 1; |
| 465 | } |
| 466 | |
| 467 | /* drop reference count of cluster config when rule is deleted */ |
| 468 | static void destroy(void *matchinfo, unsigned int matchinfosize) |
| 469 | { |
| 470 | struct ipt_clusterip_tgt_info *cipinfo = matchinfo; |
| 471 | |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 472 | /* if no more entries are referencing the config, remove it |
| 473 | * from the list and destroy the proc entry */ |
| 474 | clusterip_config_entry_put(cipinfo->config); |
| 475 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | clusterip_config_put(cipinfo->config); |
| 477 | } |
| 478 | |
| 479 | static struct ipt_target clusterip_tgt = { |
| 480 | .name = "CLUSTERIP", |
| 481 | .target = &target, |
| 482 | .checkentry = &checkentry, |
| 483 | .destroy = &destroy, |
| 484 | .me = THIS_MODULE |
| 485 | }; |
| 486 | |
| 487 | |
| 488 | /*********************************************************************** |
| 489 | * ARP MANGLING CODE |
| 490 | ***********************************************************************/ |
| 491 | |
| 492 | /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */ |
| 493 | struct arp_payload { |
| 494 | u_int8_t src_hw[ETH_ALEN]; |
| 495 | u_int32_t src_ip; |
| 496 | u_int8_t dst_hw[ETH_ALEN]; |
| 497 | u_int32_t dst_ip; |
| 498 | } __attribute__ ((packed)); |
| 499 | |
| 500 | #ifdef CLUSTERIP_DEBUG |
| 501 | static void arp_print(struct arp_payload *payload) |
| 502 | { |
| 503 | #define HBUFFERLEN 30 |
| 504 | char hbuffer[HBUFFERLEN]; |
| 505 | int j,k; |
| 506 | const char hexbuf[]= "0123456789abcdef"; |
| 507 | |
| 508 | for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) { |
| 509 | hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15]; |
| 510 | hbuffer[k++]=hexbuf[payload->src_hw[j]&15]; |
| 511 | hbuffer[k++]=':'; |
| 512 | } |
| 513 | hbuffer[--k]='\0'; |
| 514 | |
| 515 | printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n", |
| 516 | NIPQUAD(payload->src_ip), hbuffer, |
| 517 | NIPQUAD(payload->dst_ip)); |
| 518 | } |
| 519 | #endif |
| 520 | |
| 521 | static unsigned int |
| 522 | arp_mangle(unsigned int hook, |
| 523 | struct sk_buff **pskb, |
| 524 | const struct net_device *in, |
| 525 | const struct net_device *out, |
| 526 | int (*okfn)(struct sk_buff *)) |
| 527 | { |
| 528 | struct arphdr *arp = (*pskb)->nh.arph; |
| 529 | struct arp_payload *payload; |
| 530 | struct clusterip_config *c; |
| 531 | |
| 532 | /* we don't care about non-ethernet and non-ipv4 ARP */ |
| 533 | if (arp->ar_hrd != htons(ARPHRD_ETHER) |
| 534 | || arp->ar_pro != htons(ETH_P_IP) |
| 535 | || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN) |
| 536 | return NF_ACCEPT; |
| 537 | |
Harald Welte | 4095ebf | 2005-06-28 12:49:30 -0700 | [diff] [blame] | 538 | /* we only want to mangle arp requests and replies */ |
| 539 | if (arp->ar_op != htons(ARPOP_REPLY) |
| 540 | && arp->ar_op != htons(ARPOP_REQUEST)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | return NF_ACCEPT; |
| 542 | |
| 543 | payload = (void *)(arp+1); |
| 544 | |
| 545 | /* if there is no clusterip configuration for the arp reply's |
| 546 | * source ip, we don't want to mangle it */ |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 547 | c = clusterip_config_find_get(payload->src_ip, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | if (!c) |
| 549 | return NF_ACCEPT; |
| 550 | |
| 551 | /* normally the linux kernel always replies to arp queries of |
| 552 | * addresses on different interfacs. However, in the CLUSTERIP case |
| 553 | * this wouldn't work, since we didn't subscribe the mcast group on |
| 554 | * other interfaces */ |
| 555 | if (c->dev != out) { |
| 556 | DEBUGP("CLUSTERIP: not mangling arp reply on different " |
| 557 | "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name); |
| 558 | clusterip_config_put(c); |
| 559 | return NF_ACCEPT; |
| 560 | } |
| 561 | |
| 562 | /* mangle reply hardware address */ |
| 563 | memcpy(payload->src_hw, c->clustermac, arp->ar_hln); |
| 564 | |
| 565 | #ifdef CLUSTERIP_DEBUG |
| 566 | DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: "); |
| 567 | arp_print(payload); |
| 568 | #endif |
| 569 | |
| 570 | clusterip_config_put(c); |
| 571 | |
| 572 | return NF_ACCEPT; |
| 573 | } |
| 574 | |
| 575 | static struct nf_hook_ops cip_arp_ops = { |
| 576 | .hook = arp_mangle, |
| 577 | .pf = NF_ARP, |
| 578 | .hooknum = NF_ARP_OUT, |
| 579 | .priority = -1 |
| 580 | }; |
| 581 | |
| 582 | /*********************************************************************** |
| 583 | * PROC DIR HANDLING |
| 584 | ***********************************************************************/ |
| 585 | |
| 586 | #ifdef CONFIG_PROC_FS |
| 587 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 588 | struct clusterip_seq_position { |
| 589 | unsigned int pos; /* position */ |
| 590 | unsigned int weight; /* number of bits set == size */ |
| 591 | unsigned int bit; /* current bit */ |
| 592 | unsigned long val; /* current value */ |
| 593 | }; |
| 594 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | static void *clusterip_seq_start(struct seq_file *s, loff_t *pos) |
| 596 | { |
| 597 | struct proc_dir_entry *pde = s->private; |
| 598 | struct clusterip_config *c = pde->data; |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 599 | unsigned int weight; |
| 600 | u_int32_t local_nodes; |
| 601 | struct clusterip_seq_position *idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 602 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 603 | /* FIXME: possible race */ |
| 604 | local_nodes = c->local_nodes; |
| 605 | weight = hweight32(local_nodes); |
| 606 | if (*pos >= weight) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | return NULL; |
| 608 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 609 | idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL); |
| 610 | if (!idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | return ERR_PTR(-ENOMEM); |
| 612 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 613 | idx->pos = *pos; |
| 614 | idx->weight = weight; |
| 615 | idx->bit = ffs(local_nodes); |
| 616 | idx->val = local_nodes; |
| 617 | clear_bit(idx->bit - 1, &idx->val); |
| 618 | |
| 619 | return idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | } |
| 621 | |
| 622 | static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos) |
| 623 | { |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 624 | struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 626 | *pos = ++idx->pos; |
| 627 | if (*pos >= idx->weight) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | kfree(v); |
| 629 | return NULL; |
| 630 | } |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 631 | idx->bit = ffs(idx->val); |
| 632 | clear_bit(idx->bit - 1, &idx->val); |
| 633 | return idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | } |
| 635 | |
| 636 | static void clusterip_seq_stop(struct seq_file *s, void *v) |
| 637 | { |
| 638 | kfree(v); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | } |
| 640 | |
| 641 | static int clusterip_seq_show(struct seq_file *s, void *v) |
| 642 | { |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 643 | struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 645 | if (idx->pos != 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | seq_putc(s, ','); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 648 | seq_printf(s, "%u", idx->bit); |
| 649 | |
| 650 | if (idx->pos == idx->weight - 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 651 | seq_putc(s, '\n'); |
| 652 | |
| 653 | return 0; |
| 654 | } |
| 655 | |
| 656 | static struct seq_operations clusterip_seq_ops = { |
| 657 | .start = clusterip_seq_start, |
| 658 | .next = clusterip_seq_next, |
| 659 | .stop = clusterip_seq_stop, |
| 660 | .show = clusterip_seq_show, |
| 661 | }; |
| 662 | |
| 663 | static int clusterip_proc_open(struct inode *inode, struct file *file) |
| 664 | { |
| 665 | int ret = seq_open(file, &clusterip_seq_ops); |
| 666 | |
| 667 | if (!ret) { |
| 668 | struct seq_file *sf = file->private_data; |
| 669 | struct proc_dir_entry *pde = PDE(inode); |
| 670 | struct clusterip_config *c = pde->data; |
| 671 | |
| 672 | sf->private = pde; |
| 673 | |
| 674 | clusterip_config_get(c); |
| 675 | } |
| 676 | |
| 677 | return ret; |
| 678 | } |
| 679 | |
| 680 | static int clusterip_proc_release(struct inode *inode, struct file *file) |
| 681 | { |
| 682 | struct proc_dir_entry *pde = PDE(inode); |
| 683 | struct clusterip_config *c = pde->data; |
| 684 | int ret; |
| 685 | |
| 686 | ret = seq_release(inode, file); |
| 687 | |
| 688 | if (!ret) |
| 689 | clusterip_config_put(c); |
| 690 | |
| 691 | return ret; |
| 692 | } |
| 693 | |
| 694 | static ssize_t clusterip_proc_write(struct file *file, const char __user *input, |
| 695 | size_t size, loff_t *ofs) |
| 696 | { |
| 697 | #define PROC_WRITELEN 10 |
| 698 | char buffer[PROC_WRITELEN+1]; |
| 699 | struct proc_dir_entry *pde = PDE(file->f_dentry->d_inode); |
| 700 | struct clusterip_config *c = pde->data; |
| 701 | unsigned long nodenum; |
| 702 | |
| 703 | if (copy_from_user(buffer, input, PROC_WRITELEN)) |
| 704 | return -EFAULT; |
| 705 | |
| 706 | if (*buffer == '+') { |
| 707 | nodenum = simple_strtoul(buffer+1, NULL, 10); |
| 708 | if (clusterip_add_node(c, nodenum)) |
| 709 | return -ENOMEM; |
| 710 | } else if (*buffer == '-') { |
| 711 | nodenum = simple_strtoul(buffer+1, NULL,10); |
| 712 | if (clusterip_del_node(c, nodenum)) |
| 713 | return -ENOENT; |
| 714 | } else |
| 715 | return -EIO; |
| 716 | |
| 717 | return size; |
| 718 | } |
| 719 | |
| 720 | static struct file_operations clusterip_proc_fops = { |
| 721 | .owner = THIS_MODULE, |
| 722 | .open = clusterip_proc_open, |
| 723 | .read = seq_read, |
| 724 | .write = clusterip_proc_write, |
| 725 | .llseek = seq_lseek, |
| 726 | .release = clusterip_proc_release, |
| 727 | }; |
| 728 | |
| 729 | #endif /* CONFIG_PROC_FS */ |
| 730 | |
| 731 | static int init_or_cleanup(int fini) |
| 732 | { |
| 733 | int ret; |
| 734 | |
| 735 | if (fini) |
| 736 | goto cleanup; |
| 737 | |
| 738 | if (ipt_register_target(&clusterip_tgt)) { |
| 739 | ret = -EINVAL; |
| 740 | goto cleanup_none; |
| 741 | } |
| 742 | |
| 743 | if (nf_register_hook(&cip_arp_ops) < 0) { |
| 744 | ret = -EINVAL; |
| 745 | goto cleanup_target; |
| 746 | } |
| 747 | |
| 748 | #ifdef CONFIG_PROC_FS |
| 749 | clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net); |
| 750 | if (!clusterip_procdir) { |
| 751 | printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n"); |
| 752 | ret = -ENOMEM; |
| 753 | goto cleanup_hook; |
| 754 | } |
| 755 | #endif /* CONFIG_PROC_FS */ |
| 756 | |
| 757 | printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n", |
| 758 | CLUSTERIP_VERSION); |
| 759 | |
| 760 | return 0; |
| 761 | |
| 762 | cleanup: |
| 763 | printk(KERN_NOTICE "ClusterIP Version %s unloading\n", |
| 764 | CLUSTERIP_VERSION); |
| 765 | #ifdef CONFIG_PROC_FS |
| 766 | remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent); |
| 767 | #endif |
| 768 | cleanup_hook: |
| 769 | nf_unregister_hook(&cip_arp_ops); |
| 770 | cleanup_target: |
| 771 | ipt_unregister_target(&clusterip_tgt); |
| 772 | cleanup_none: |
| 773 | return -EINVAL; |
| 774 | } |
| 775 | |
| 776 | static int __init init(void) |
| 777 | { |
| 778 | return init_or_cleanup(0); |
| 779 | } |
| 780 | |
| 781 | static void __exit fini(void) |
| 782 | { |
| 783 | init_or_cleanup(1); |
| 784 | } |
| 785 | |
| 786 | module_init(init); |
| 787 | module_exit(fini); |