net/ipv4/netfilter/ipt_CLUSTERIP.c

   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/bitops.h>
  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>
  32 #include <net/netfilter/nf_conntrack_compat.h>
  33
  34 #define CLUSTERIP_VERSION "0.8"
  35
  36 #define DEBUG_CLUSTERIP
  37
  38 #ifdef DEBUG_CLUSTERIP
  39 #define DEBUGP  printk
  40 #else
  41 #define DEBUGP
  42 #endif
  43
  44 #define ASSERT_READ_LOCK(x)
  45
  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 */
  53         atomic_t entries;                       /* number of entries/rules
  54                                                  * referencing us */
  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 */
  60         unsigned long local_nodes;              /* node number array */
  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
  71 /* clusterip_lock protects the clusterip_configs list */
  72 static DEFINE_RWLOCK(clusterip_lock);
  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
  80 clusterip_config_get(struct clusterip_config *c)
  81 {
  82         atomic_inc(&c->refcount);
  83 }
  84
  85 static inline void
  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)) {
 106                 write_lock_bh(&clusterip_lock);
 107                 list_del(&c->list);
 108                 write_unlock_bh(&clusterip_lock);
 109
 110                 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
 111                 dev_put(c->dev);
 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
 119         }
 120 }
 121
 122 static struct clusterip_config *
 123 __clusterip_config_find(u_int32_t clusterip)
 124 {
 125         struct list_head *pos;
 126
 127         ASSERT_READ_LOCK(&clusterip_lock);
 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 *
 140 clusterip_config_find_get(u_int32_t clusterip, int entry)
 141 {
 142         struct clusterip_config *c;
 143
 144         read_lock_bh(&clusterip_lock);
 145         c = __clusterip_config_find(clusterip);
 146         if (!c) {
 147                 read_unlock_bh(&clusterip_lock);
 148                 return NULL;
 149         }
 150         atomic_inc(&c->refcount);
 151         if (entry)
 152                 atomic_inc(&c->entries);
 153         read_unlock_bh(&clusterip_lock);
 154
 155         return c;
 156 }
 157
 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
 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;
 185         clusterip_config_init_nodelist(c, i);
 186         c->hash_mode = i->hash_mode;
 187         c->hash_initval = i->hash_initval;
 188         atomic_set(&c->refcount, 1);
 189         atomic_set(&c->entries, 1);
 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
 203         write_lock_bh(&clusterip_lock);
 204         list_add(&c->list, &clusterip_configs);
 205         write_unlock_bh(&clusterip_lock);
 206
 207         return c;
 208 }
 209
 210 static int
 211 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
 212 {
 213
 214         if (nodenum == 0 ||
 215             nodenum > c->num_total_nodes)
 216                 return 1;
 217
 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;
 221
 222         return 0;
 223 }
 224
 225 static int
 226 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
 227 {
 228         if (nodenum == 0 ||
 229             nodenum > c->num_total_nodes)
 230                 return 1;
 231
 232         if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
 233                 return 0;
 234
 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 {
 302         return test_bit(hash - 1, &config->local_nodes);
 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 struct xt_target *target,
 315        const void *targinfo,
 316        void *userinfo)
 317 {
 318         const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
 319         enum ip_conntrack_info ctinfo;
 320         u_int32_t *mark, hash;
 321
 322         /* don't need to clusterip_config_get() here, since refcount
 323          * is only decremented by destroy() - and ip_tables guarantees
 324          * that the ->target() function isn't called after ->destroy() */
 325
 326         mark = nf_ct_get_mark((*pskb), &ctinfo);
 327         if (mark == NULL) {
 328                 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
 329                         /* FIXME: need to drop invalid ones, since replies
 330                          * to outgoing connections of other nodes will be
 331                          * marked as INVALID */
 332                 return NF_DROP;
 333         }
 334
 335         /* special case: ICMP error handling. conntrack distinguishes between
 336          * error messages (RELATED) and information requests (see below) */
 337         if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP
 338             && (ctinfo == IP_CT_RELATED
 339                 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
 340                 return IPT_CONTINUE;
 341
 342         /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
 343          * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
 344          * on, which all have an ID field [relevant for hashing]. */
 345
 346         hash = clusterip_hashfn(*pskb, cipinfo->config);
 347
 348         switch (ctinfo) {
 349                 case IP_CT_NEW:
 350                         *mark = hash;
 351                         break;
 352                 case IP_CT_RELATED:
 353                 case IP_CT_RELATED+IP_CT_IS_REPLY:
 354                         /* FIXME: we don't handle expectations at the
 355                          * moment.  they can arrive on a different node than
 356                          * the master connection (e.g. FTP passive mode) */
 357                 case IP_CT_ESTABLISHED:
 358                 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
 359                         break;
 360                 default:
 361                         break;
 362         }
 363
 364 #ifdef DEBUG_CLUSTERP
 365         DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 366 #endif
 367         DEBUGP("hash=%u ct_hash=%u ", hash, *mark);
 368         if (!clusterip_responsible(cipinfo->config, hash)) {
 369                 DEBUGP("not responsible\n");
 370                 return NF_DROP;
 371         }
 372         DEBUGP("responsible\n");
 373
 374         /* despite being received via linklayer multicast, this is
 375          * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
 376         (*pskb)->pkt_type = PACKET_HOST;
 377
 378         return IPT_CONTINUE;
 379 }
 380
 381 static int
 382 checkentry(const char *tablename,
 383            const void *e_void,
 384            const struct xt_target *target,
 385            void *targinfo,
 386            unsigned int targinfosize,
 387            unsigned int hook_mask)
 388 {
 389         struct ipt_clusterip_tgt_info *cipinfo = targinfo;
 390         const struct ipt_entry *e = e_void;
 391
 392         struct clusterip_config *config;
 393
 394         if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
 395             cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
 396             cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
 397                 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
 398                         cipinfo->hash_mode);
 399                 return 0;
 400
 401         }
 402         if (e->ip.dmsk.s_addr != 0xffffffff
 403             || e->ip.dst.s_addr == 0) {
 404                 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
 405                 return 0;
 406         }
 407
 408         /* FIXME: further sanity checks */
 409
 410         config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
 411         if (config) {
 412                 if (cipinfo->config != NULL) {
 413                         /* Case A: This is an entry that gets reloaded, since
 414                          * it still has a cipinfo->config pointer. Simply
 415                          * increase the entry refcount and return */
 416                         if (cipinfo->config != config) {
 417                                 printk(KERN_ERR "CLUSTERIP: Reloaded entry "
 418                                        "has invalid config pointer!\n");
 419                                 return 0;
 420                         }
 421                         clusterip_config_entry_get(cipinfo->config);
 422                 } else {
 423                         /* Case B: This is a new rule referring to an existing
 424                          * clusterip config. */
 425                         cipinfo->config = config;
 426                         clusterip_config_entry_get(cipinfo->config);
 427                 }
 428         } else {
 429                 /* Case C: This is a completely new clusterip config */
 430                 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
 431                         printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
 432                         return 0;
 433                 } else {
 434                         struct net_device *dev;
 435
 436                         if (e->ip.iniface[0] == '\0') {
 437                                 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
 438                                 return 0;
 439                         }
 440
 441                         dev = dev_get_by_name(e->ip.iniface);
 442                         if (!dev) {
 443                                 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
 444                                 return 0;
 445                         }
 446
 447                         config = clusterip_config_init(cipinfo,
 448                                                         e->ip.dst.s_addr, dev);
 449                         if (!config) {
 450                                 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
 451                                 dev_put(dev);
 452                                 return 0;
 453                         }
 454                         dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
 455                 }
 456                 cipinfo->config = config;
 457         }
 458
 459         return 1;
 460 }
 461
 462 /* drop reference count of cluster config when rule is deleted */
 463 static void destroy(const struct xt_target *target, void *targinfo,
 464                     unsigned int targinfosize)
 465 {
 466         struct ipt_clusterip_tgt_info *cipinfo = targinfo;
 467
 468         /* if no more entries are referencing the config, remove it
 469          * from the list and destroy the proc entry */
 470         clusterip_config_entry_put(cipinfo->config);
 471
 472         clusterip_config_put(cipinfo->config);
 473 }
 474
 475 static struct ipt_target clusterip_tgt = {
 476         .name           = "CLUSTERIP",
 477         .target         = target,
 478         .targetsize     = sizeof(struct ipt_clusterip_tgt_info),
 479         .checkentry     = checkentry,
 480         .destroy        = destroy,
 481         .me             = THIS_MODULE
 482 };
 483
 484
 485 /***********************************************************************
 486  * ARP MANGLING CODE
 487  ***********************************************************************/
 488
 489 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
 490 struct arp_payload {
 491         u_int8_t src_hw[ETH_ALEN];
 492         u_int32_t src_ip;
 493         u_int8_t dst_hw[ETH_ALEN];
 494         u_int32_t dst_ip;
 495 } __attribute__ ((packed));
 496
 497 #ifdef CLUSTERIP_DEBUG
 498 static void arp_print(struct arp_payload *payload)
 499 {
 500 #define HBUFFERLEN 30
 501         char hbuffer[HBUFFERLEN];
 502         int j,k;
 503         const char hexbuf[]= "0123456789abcdef";
 504
 505         for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
 506                 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
 507                 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
 508                 hbuffer[k++]=':';
 509         }
 510         hbuffer[--k]='\0';
 511
 512         printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
 513                 NIPQUAD(payload->src_ip), hbuffer,
 514                 NIPQUAD(payload->dst_ip));
 515 }
 516 #endif
 517
 518 static unsigned int
 519 arp_mangle(unsigned int hook,
 520            struct sk_buff **pskb,
 521            const struct net_device *in,
 522            const struct net_device *out,
 523            int (*okfn)(struct sk_buff *))
 524 {
 525         struct arphdr *arp = (*pskb)->nh.arph;
 526         struct arp_payload *payload;
 527         struct clusterip_config *c;
 528
 529         /* we don't care about non-ethernet and non-ipv4 ARP */
 530         if (arp->ar_hrd != htons(ARPHRD_ETHER)
 531             || arp->ar_pro != htons(ETH_P_IP)
 532             || arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
 533                 return NF_ACCEPT;
 534
 535         /* we only want to mangle arp requests and replies */
 536         if (arp->ar_op != htons(ARPOP_REPLY)
 537             && arp->ar_op != htons(ARPOP_REQUEST))
 538                 return NF_ACCEPT;
 539
 540         payload = (void *)(arp+1);
 541
 542         /* if there is no clusterip configuration for the arp reply's
 543          * source ip, we don't want to mangle it */
 544         c = clusterip_config_find_get(payload->src_ip, 0);
 545         if (!c)
 546                 return NF_ACCEPT;
 547
 548         /* normally the linux kernel always replies to arp queries of
 549          * addresses on different interfacs.  However, in the CLUSTERIP case
 550          * this wouldn't work, since we didn't subscribe the mcast group on
 551          * other interfaces */
 552         if (c->dev != out) {
 553                 DEBUGP("CLUSTERIP: not mangling arp reply on different "
 554                        "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name);
 555                 clusterip_config_put(c);
 556                 return NF_ACCEPT;
 557         }
 558
 559         /* mangle reply hardware address */
 560         memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
 561
 562 #ifdef CLUSTERIP_DEBUG
 563         DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
 564         arp_print(payload);
 565 #endif
 566
 567         clusterip_config_put(c);
 568
 569         return NF_ACCEPT;
 570 }
 571
 572 static struct nf_hook_ops cip_arp_ops = {
 573         .hook = arp_mangle,
 574         .pf = NF_ARP,
 575         .hooknum = NF_ARP_OUT,
 576         .priority = -1
 577 };
 578
 579 /***********************************************************************
 580  * PROC DIR HANDLING
 581  ***********************************************************************/
 582
 583 #ifdef CONFIG_PROC_FS
 584
 585 struct clusterip_seq_position {
 586         unsigned int pos;       /* position */
 587         unsigned int weight;    /* number of bits set == size */
 588         unsigned int bit;       /* current bit */
 589         unsigned long val;      /* current value */
 590 };
 591
 592 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
 593 {
 594         struct proc_dir_entry *pde = s->private;
 595         struct clusterip_config *c = pde->data;
 596         unsigned int weight;
 597         u_int32_t local_nodes;
 598         struct clusterip_seq_position *idx;
 599
 600         /* FIXME: possible race */
 601         local_nodes = c->local_nodes;
 602         weight = hweight32(local_nodes);
 603         if (*pos >= weight)
 604                 return NULL;
 605
 606         idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
 607         if (!idx)
 608                 return ERR_PTR(-ENOMEM);
 609
 610         idx->pos = *pos;
 611         idx->weight = weight;
 612         idx->bit = ffs(local_nodes);
 613         idx->val = local_nodes;
 614         clear_bit(idx->bit - 1, &idx->val);
 615
 616         return idx;
 617 }
 618
 619 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
 620 {
 621         struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
 622
 623         *pos = ++idx->pos;
 624         if (*pos >= idx->weight) {
 625                 kfree(v);
 626                 return NULL;
 627         }
 628         idx->bit = ffs(idx->val);
 629         clear_bit(idx->bit - 1, &idx->val);
 630         return idx;
 631 }
 632
 633 static void clusterip_seq_stop(struct seq_file *s, void *v)
 634 {
 635         kfree(v);
 636 }
 637
 638 static int clusterip_seq_show(struct seq_file *s, void *v)
 639 {
 640         struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
 641
 642         if (idx->pos != 0)
 643                 seq_putc(s, ',');
 644
 645         seq_printf(s, "%u", idx->bit);
 646
 647         if (idx->pos == idx->weight - 1)
 648                 seq_putc(s, '\n');
 649
 650         return 0;
 651 }
 652
 653 static struct seq_operations clusterip_seq_ops = {
 654         .start  = clusterip_seq_start,
 655         .next   = clusterip_seq_next,
 656         .stop   = clusterip_seq_stop,
 657         .show   = clusterip_seq_show,
 658 };
 659
 660 static int clusterip_proc_open(struct inode *inode, struct file *file)
 661 {
 662         int ret = seq_open(file, &clusterip_seq_ops);
 663
 664         if (!ret) {
 665                 struct seq_file *sf = file->private_data;
 666                 struct proc_dir_entry *pde = PDE(inode);
 667                 struct clusterip_config *c = pde->data;
 668
 669                 sf->private = pde;
 670
 671                 clusterip_config_get(c);
 672         }
 673
 674         return ret;
 675 }
 676
 677 static int clusterip_proc_release(struct inode *inode, struct file *file)
 678 {
 679         struct proc_dir_entry *pde = PDE(inode);
 680         struct clusterip_config *c = pde->data;
 681         int ret;
 682
 683         ret = seq_release(inode, file);
 684
 685         if (!ret)
 686                 clusterip_config_put(c);
 687
 688         return ret;
 689 }
 690
 691 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
 692                                 size_t size, loff_t *ofs)
 693 {
 694 #define PROC_WRITELEN   10
 695         char buffer[PROC_WRITELEN+1];
 696         struct proc_dir_entry *pde = PDE(file->f_dentry->d_inode);
 697         struct clusterip_config *c = pde->data;
 698         unsigned long nodenum;
 699
 700         if (copy_from_user(buffer, input, PROC_WRITELEN))
 701                 return -EFAULT;
 702
 703         if (*buffer == '+') {
 704                 nodenum = simple_strtoul(buffer+1, NULL, 10);
 705                 if (clusterip_add_node(c, nodenum))
 706                         return -ENOMEM;
 707         } else if (*buffer == '-') {
 708                 nodenum = simple_strtoul(buffer+1, NULL,10);
 709                 if (clusterip_del_node(c, nodenum))
 710                         return -ENOENT;
 711         } else
 712                 return -EIO;
 713
 714         return size;
 715 }
 716
 717 static struct file_operations clusterip_proc_fops = {
 718         .owner   = THIS_MODULE,
 719         .open    = clusterip_proc_open,
 720         .read    = seq_read,
 721         .write   = clusterip_proc_write,
 722         .llseek  = seq_lseek,
 723         .release = clusterip_proc_release,
 724 };
 725
 726 #endif /* CONFIG_PROC_FS */
 727
 728 static int init_or_cleanup(int fini)
 729 {
 730         int ret;
 731
 732         if (fini)
 733                 goto cleanup;
 734
 735         if (ipt_register_target(&clusterip_tgt)) {
 736                 ret = -EINVAL;
 737                 goto cleanup_none;
 738         }
 739
 740         if (nf_register_hook(&cip_arp_ops) < 0) {
 741                 ret = -EINVAL;
 742                 goto cleanup_target;
 743         }
 744
 745 #ifdef CONFIG_PROC_FS
 746         clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net);
 747         if (!clusterip_procdir) {
 748                 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
 749                 ret = -ENOMEM;
 750                 goto cleanup_hook;
 751         }
 752 #endif /* CONFIG_PROC_FS */
 753
 754         printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
 755                 CLUSTERIP_VERSION);
 756
 757         return 0;
 758
 759 cleanup:
 760         printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
 761                 CLUSTERIP_VERSION);
 762 #ifdef CONFIG_PROC_FS
 763         remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
 764 #endif
 765 cleanup_hook:
 766         nf_unregister_hook(&cip_arp_ops);
 767 cleanup_target:
 768         ipt_unregister_target(&clusterip_tgt);
 769 cleanup_none:
 770         return -EINVAL;
 771 }
 772
 773 static int __init init(void)
 774 {
 775         return init_or_cleanup(0);
 776 }
 777
 778 static void __exit fini(void)
 779 {
 780         init_or_cleanup(1);
 781 }
 782
 783 module_init(init);
 784 module_exit(fini);