[POWERPC] Make soft_enabled irqs preempt safe
[usb.git] / net / ipv4 / netfilter / ipt_CLUSTERIP.c
blob0983650622347bf2a6af923c9791a1ed575be5c5
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>
5 * Development of this code funded by SuSE Linux AG, http://www.suse.com/
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.
12 #include <linux/module.h>
13 #include <linux/proc_fs.h>
14 #include <linux/jhash.h>
15 #include <linux/bitops.h>
16 #include <linux/skbuff.h>
17 #include <linux/ip.h>
18 #include <linux/tcp.h>
19 #include <linux/udp.h>
20 #include <linux/icmp.h>
21 #include <linux/if_arp.h>
22 #include <linux/proc_fs.h>
23 #include <linux/seq_file.h>
25 #include <net/checksum.h>
27 #include <linux/netfilter_arp.h>
29 #include <linux/netfilter_ipv4/ip_tables.h>
30 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
31 #include <net/netfilter/nf_conntrack_compat.h>
33 #define CLUSTERIP_VERSION "0.8"
35 #define DEBUG_CLUSTERIP
37 #ifdef DEBUG_CLUSTERIP
38 #define DEBUGP printk
39 #else
40 #define DEBUGP
41 #endif
43 MODULE_LICENSE("GPL");
44 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
45 MODULE_DESCRIPTION("iptables target for CLUSTERIP");
47 struct clusterip_config {
48 struct list_head list; /* list of all configs */
49 atomic_t refcount; /* reference count */
50 atomic_t entries; /* number of entries/rules
51 * referencing us */
53 __be32 clusterip; /* the IP address */
54 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
55 struct net_device *dev; /* device */
56 u_int16_t num_total_nodes; /* total number of nodes */
57 unsigned long local_nodes; /* node number array */
59 #ifdef CONFIG_PROC_FS
60 struct proc_dir_entry *pde; /* proc dir entry */
61 #endif
62 enum clusterip_hashmode hash_mode; /* which hashing mode */
63 u_int32_t hash_initval; /* hash initialization */
66 static LIST_HEAD(clusterip_configs);
68 /* clusterip_lock protects the clusterip_configs list */
69 static DEFINE_RWLOCK(clusterip_lock);
71 #ifdef CONFIG_PROC_FS
72 static struct file_operations clusterip_proc_fops;
73 static struct proc_dir_entry *clusterip_procdir;
74 #endif
76 static inline void
77 clusterip_config_get(struct clusterip_config *c)
79 atomic_inc(&c->refcount);
82 static inline void
83 clusterip_config_put(struct clusterip_config *c)
85 if (atomic_dec_and_test(&c->refcount))
86 kfree(c);
89 /* increase the count of entries(rules) using/referencing this config */
90 static inline void
91 clusterip_config_entry_get(struct clusterip_config *c)
93 atomic_inc(&c->entries);
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)
102 if (atomic_dec_and_test(&c->entries)) {
103 write_lock_bh(&clusterip_lock);
104 list_del(&c->list);
105 write_unlock_bh(&clusterip_lock);
107 dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0);
108 dev_put(c->dev);
110 /* In case anyone still accesses the file, the open/close
111 * functions are also incrementing the refcount on their own,
112 * so it's safe to remove the entry even if it's in use. */
113 #ifdef CONFIG_PROC_FS
114 remove_proc_entry(c->pde->name, c->pde->parent);
115 #endif
119 static struct clusterip_config *
120 __clusterip_config_find(__be32 clusterip)
122 struct list_head *pos;
124 list_for_each(pos, &clusterip_configs) {
125 struct clusterip_config *c = list_entry(pos,
126 struct clusterip_config, list);
127 if (c->clusterip == clusterip) {
128 return c;
132 return NULL;
135 static inline struct clusterip_config *
136 clusterip_config_find_get(__be32 clusterip, int entry)
138 struct clusterip_config *c;
140 read_lock_bh(&clusterip_lock);
141 c = __clusterip_config_find(clusterip);
142 if (!c) {
143 read_unlock_bh(&clusterip_lock);
144 return NULL;
146 atomic_inc(&c->refcount);
147 if (entry)
148 atomic_inc(&c->entries);
149 read_unlock_bh(&clusterip_lock);
151 return c;
154 static void
155 clusterip_config_init_nodelist(struct clusterip_config *c,
156 const struct ipt_clusterip_tgt_info *i)
158 int n;
160 for (n = 0; n < i->num_local_nodes; n++) {
161 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
165 static struct clusterip_config *
166 clusterip_config_init(struct ipt_clusterip_tgt_info *i, __be32 ip,
167 struct net_device *dev)
169 struct clusterip_config *c;
171 c = kzalloc(sizeof(*c), GFP_ATOMIC);
172 if (!c)
173 return NULL;
175 c->dev = dev;
176 c->clusterip = ip;
177 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
178 c->num_total_nodes = i->num_total_nodes;
179 clusterip_config_init_nodelist(c, i);
180 c->hash_mode = i->hash_mode;
181 c->hash_initval = i->hash_initval;
182 atomic_set(&c->refcount, 1);
183 atomic_set(&c->entries, 1);
185 #ifdef CONFIG_PROC_FS
187 char buffer[16];
189 /* create proc dir entry */
190 sprintf(buffer, "%u.%u.%u.%u", NIPQUAD(ip));
191 c->pde = create_proc_entry(buffer, S_IWUSR|S_IRUSR,
192 clusterip_procdir);
193 if (!c->pde) {
194 kfree(c);
195 return NULL;
198 c->pde->proc_fops = &clusterip_proc_fops;
199 c->pde->data = c;
200 #endif
202 write_lock_bh(&clusterip_lock);
203 list_add(&c->list, &clusterip_configs);
204 write_unlock_bh(&clusterip_lock);
206 return c;
209 #ifdef CONFIG_PROC_FS
210 static int
211 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
214 if (nodenum == 0 ||
215 nodenum > c->num_total_nodes)
216 return 1;
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;
222 return 0;
225 static int
226 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
228 if (nodenum == 0 ||
229 nodenum > c->num_total_nodes)
230 return 1;
232 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
233 return 0;
235 return 1;
237 #endif
239 static inline u_int32_t
240 clusterip_hashfn(struct sk_buff *skb, struct clusterip_config *config)
242 struct iphdr *iph = skb->nh.iph;
243 unsigned long hashval;
244 u_int16_t sport, dport;
245 u_int16_t *ports;
247 switch (iph->protocol) {
248 case IPPROTO_TCP:
249 case IPPROTO_UDP:
250 case IPPROTO_SCTP:
251 case IPPROTO_DCCP:
252 case IPPROTO_ICMP:
253 ports = (void *)iph+iph->ihl*4;
254 sport = ports[0];
255 dport = ports[1];
256 break;
257 default:
258 if (net_ratelimit()) {
259 printk(KERN_NOTICE "CLUSTERIP: unknown protocol `%u'\n",
260 iph->protocol);
262 sport = dport = 0;
265 switch (config->hash_mode) {
266 case CLUSTERIP_HASHMODE_SIP:
267 hashval = jhash_1word(ntohl(iph->saddr),
268 config->hash_initval);
269 break;
270 case CLUSTERIP_HASHMODE_SIP_SPT:
271 hashval = jhash_2words(ntohl(iph->saddr), sport,
272 config->hash_initval);
273 break;
274 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
275 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
276 config->hash_initval);
277 break;
278 default:
279 /* to make gcc happy */
280 hashval = 0;
281 /* This cannot happen, unless the check function wasn't called
282 * at rule load time */
283 printk("CLUSTERIP: unknown mode `%u'\n", config->hash_mode);
284 BUG();
285 break;
288 /* node numbers are 1..n, not 0..n */
289 return ((hashval % config->num_total_nodes)+1);
292 static inline int
293 clusterip_responsible(struct clusterip_config *config, u_int32_t hash)
295 return test_bit(hash - 1, &config->local_nodes);
298 /***********************************************************************
299 * IPTABLES TARGET
300 ***********************************************************************/
302 static unsigned int
303 target(struct sk_buff **pskb,
304 const struct net_device *in,
305 const struct net_device *out,
306 unsigned int hooknum,
307 const struct xt_target *target,
308 const void *targinfo)
310 const struct ipt_clusterip_tgt_info *cipinfo = targinfo;
311 enum ip_conntrack_info ctinfo;
312 u_int32_t *mark, hash;
314 /* don't need to clusterip_config_get() here, since refcount
315 * is only decremented by destroy() - and ip_tables guarantees
316 * that the ->target() function isn't called after ->destroy() */
318 mark = nf_ct_get_mark((*pskb), &ctinfo);
319 if (mark == NULL) {
320 printk(KERN_ERR "CLUSTERIP: no conntrack!\n");
321 /* FIXME: need to drop invalid ones, since replies
322 * to outgoing connections of other nodes will be
323 * marked as INVALID */
324 return NF_DROP;
327 /* special case: ICMP error handling. conntrack distinguishes between
328 * error messages (RELATED) and information requests (see below) */
329 if ((*pskb)->nh.iph->protocol == IPPROTO_ICMP
330 && (ctinfo == IP_CT_RELATED
331 || ctinfo == IP_CT_RELATED+IP_CT_IS_REPLY))
332 return IPT_CONTINUE;
334 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
335 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
336 * on, which all have an ID field [relevant for hashing]. */
338 hash = clusterip_hashfn(*pskb, cipinfo->config);
340 switch (ctinfo) {
341 case IP_CT_NEW:
342 *mark = hash;
343 break;
344 case IP_CT_RELATED:
345 case IP_CT_RELATED+IP_CT_IS_REPLY:
346 /* FIXME: we don't handle expectations at the
347 * moment. they can arrive on a different node than
348 * the master connection (e.g. FTP passive mode) */
349 case IP_CT_ESTABLISHED:
350 case IP_CT_ESTABLISHED+IP_CT_IS_REPLY:
351 break;
352 default:
353 break;
356 #ifdef DEBUG_CLUSTERP
357 DUMP_TUPLE(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
358 #endif
359 DEBUGP("hash=%u ct_hash=%u ", hash, *mark);
360 if (!clusterip_responsible(cipinfo->config, hash)) {
361 DEBUGP("not responsible\n");
362 return NF_DROP;
364 DEBUGP("responsible\n");
366 /* despite being received via linklayer multicast, this is
367 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
368 (*pskb)->pkt_type = PACKET_HOST;
370 return IPT_CONTINUE;
373 static int
374 checkentry(const char *tablename,
375 const void *e_void,
376 const struct xt_target *target,
377 void *targinfo,
378 unsigned int hook_mask)
380 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
381 const struct ipt_entry *e = e_void;
383 struct clusterip_config *config;
385 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
386 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
387 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
388 printk(KERN_WARNING "CLUSTERIP: unknown mode `%u'\n",
389 cipinfo->hash_mode);
390 return 0;
393 if (e->ip.dmsk.s_addr != htonl(0xffffffff)
394 || e->ip.dst.s_addr == 0) {
395 printk(KERN_ERR "CLUSTERIP: Please specify destination IP\n");
396 return 0;
399 /* FIXME: further sanity checks */
401 config = clusterip_config_find_get(e->ip.dst.s_addr, 1);
402 if (config) {
403 if (cipinfo->config != NULL) {
404 /* Case A: This is an entry that gets reloaded, since
405 * it still has a cipinfo->config pointer. Simply
406 * increase the entry refcount and return */
407 if (cipinfo->config != config) {
408 printk(KERN_ERR "CLUSTERIP: Reloaded entry "
409 "has invalid config pointer!\n");
410 return 0;
412 clusterip_config_entry_get(cipinfo->config);
413 } else {
414 /* Case B: This is a new rule referring to an existing
415 * clusterip config. */
416 cipinfo->config = config;
417 clusterip_config_entry_get(cipinfo->config);
419 } else {
420 /* Case C: This is a completely new clusterip config */
421 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
422 printk(KERN_WARNING "CLUSTERIP: no config found for %u.%u.%u.%u, need 'new'\n", NIPQUAD(e->ip.dst.s_addr));
423 return 0;
424 } else {
425 struct net_device *dev;
427 if (e->ip.iniface[0] == '\0') {
428 printk(KERN_WARNING "CLUSTERIP: Please specify an interface name\n");
429 return 0;
432 dev = dev_get_by_name(e->ip.iniface);
433 if (!dev) {
434 printk(KERN_WARNING "CLUSTERIP: no such interface %s\n", e->ip.iniface);
435 return 0;
438 config = clusterip_config_init(cipinfo,
439 e->ip.dst.s_addr, dev);
440 if (!config) {
441 printk(KERN_WARNING "CLUSTERIP: cannot allocate config\n");
442 dev_put(dev);
443 return 0;
445 dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0);
447 cipinfo->config = config;
450 return 1;
453 /* drop reference count of cluster config when rule is deleted */
454 static void destroy(const struct xt_target *target, void *targinfo)
456 struct ipt_clusterip_tgt_info *cipinfo = targinfo;
458 /* if no more entries are referencing the config, remove it
459 * from the list and destroy the proc entry */
460 clusterip_config_entry_put(cipinfo->config);
462 clusterip_config_put(cipinfo->config);
465 static struct ipt_target clusterip_tgt = {
466 .name = "CLUSTERIP",
467 .target = target,
468 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
469 .checkentry = checkentry,
470 .destroy = destroy,
471 .me = THIS_MODULE
475 /***********************************************************************
476 * ARP MANGLING CODE
477 ***********************************************************************/
479 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
480 struct arp_payload {
481 u_int8_t src_hw[ETH_ALEN];
482 __be32 src_ip;
483 u_int8_t dst_hw[ETH_ALEN];
484 __be32 dst_ip;
485 } __attribute__ ((packed));
487 #ifdef CLUSTERIP_DEBUG
488 static void arp_print(struct arp_payload *payload)
490 #define HBUFFERLEN 30
491 char hbuffer[HBUFFERLEN];
492 int j,k;
493 const char hexbuf[]= "0123456789abcdef";
495 for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) {
496 hbuffer[k++]=hexbuf[(payload->src_hw[j]>>4)&15];
497 hbuffer[k++]=hexbuf[payload->src_hw[j]&15];
498 hbuffer[k++]=':';
500 hbuffer[--k]='\0';
502 printk("src %u.%u.%u.%u@%s, dst %u.%u.%u.%u\n",
503 NIPQUAD(payload->src_ip), hbuffer,
504 NIPQUAD(payload->dst_ip));
506 #endif
508 static unsigned int
509 arp_mangle(unsigned int hook,
510 struct sk_buff **pskb,
511 const struct net_device *in,
512 const struct net_device *out,
513 int (*okfn)(struct sk_buff *))
515 struct arphdr *arp = (*pskb)->nh.arph;
516 struct arp_payload *payload;
517 struct clusterip_config *c;
519 /* we don't care about non-ethernet and non-ipv4 ARP */
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)
523 return NF_ACCEPT;
525 /* we only want to mangle arp requests and replies */
526 if (arp->ar_op != htons(ARPOP_REPLY)
527 && arp->ar_op != htons(ARPOP_REQUEST))
528 return NF_ACCEPT;
530 payload = (void *)(arp+1);
532 /* if there is no clusterip configuration for the arp reply's
533 * source ip, we don't want to mangle it */
534 c = clusterip_config_find_get(payload->src_ip, 0);
535 if (!c)
536 return NF_ACCEPT;
538 /* normally the linux kernel always replies to arp queries of
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 */
542 if (c->dev != out) {
543 DEBUGP("CLUSTERIP: not mangling arp reply on different "
544 "interface: cip'%s'-skb'%s'\n", c->dev->name, out->name);
545 clusterip_config_put(c);
546 return NF_ACCEPT;
549 /* mangle reply hardware address */
550 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
552 #ifdef CLUSTERIP_DEBUG
553 DEBUGP(KERN_DEBUG "CLUSTERIP mangled arp reply: ");
554 arp_print(payload);
555 #endif
557 clusterip_config_put(c);
559 return NF_ACCEPT;
562 static struct nf_hook_ops cip_arp_ops = {
563 .hook = arp_mangle,
564 .pf = NF_ARP,
565 .hooknum = NF_ARP_OUT,
566 .priority = -1
569 /***********************************************************************
570 * PROC DIR HANDLING
571 ***********************************************************************/
573 #ifdef CONFIG_PROC_FS
575 struct clusterip_seq_position {
576 unsigned int pos; /* position */
577 unsigned int weight; /* number of bits set == size */
578 unsigned int bit; /* current bit */
579 unsigned long val; /* current value */
582 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
584 struct proc_dir_entry *pde = s->private;
585 struct clusterip_config *c = pde->data;
586 unsigned int weight;
587 u_int32_t local_nodes;
588 struct clusterip_seq_position *idx;
590 /* FIXME: possible race */
591 local_nodes = c->local_nodes;
592 weight = hweight32(local_nodes);
593 if (*pos >= weight)
594 return NULL;
596 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
597 if (!idx)
598 return ERR_PTR(-ENOMEM);
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);
606 return idx;
609 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
611 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
613 *pos = ++idx->pos;
614 if (*pos >= idx->weight) {
615 kfree(v);
616 return NULL;
618 idx->bit = ffs(idx->val);
619 clear_bit(idx->bit - 1, &idx->val);
620 return idx;
623 static void clusterip_seq_stop(struct seq_file *s, void *v)
625 kfree(v);
628 static int clusterip_seq_show(struct seq_file *s, void *v)
630 struct clusterip_seq_position *idx = (struct clusterip_seq_position *)v;
632 if (idx->pos != 0)
633 seq_putc(s, ',');
635 seq_printf(s, "%u", idx->bit);
637 if (idx->pos == idx->weight - 1)
638 seq_putc(s, '\n');
640 return 0;
643 static struct seq_operations clusterip_seq_ops = {
644 .start = clusterip_seq_start,
645 .next = clusterip_seq_next,
646 .stop = clusterip_seq_stop,
647 .show = clusterip_seq_show,
650 static int clusterip_proc_open(struct inode *inode, struct file *file)
652 int ret = seq_open(file, &clusterip_seq_ops);
654 if (!ret) {
655 struct seq_file *sf = file->private_data;
656 struct proc_dir_entry *pde = PDE(inode);
657 struct clusterip_config *c = pde->data;
659 sf->private = pde;
661 clusterip_config_get(c);
664 return ret;
667 static int clusterip_proc_release(struct inode *inode, struct file *file)
669 struct proc_dir_entry *pde = PDE(inode);
670 struct clusterip_config *c = pde->data;
671 int ret;
673 ret = seq_release(inode, file);
675 if (!ret)
676 clusterip_config_put(c);
678 return ret;
681 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
682 size_t size, loff_t *ofs)
684 #define PROC_WRITELEN 10
685 char buffer[PROC_WRITELEN+1];
686 struct proc_dir_entry *pde = PDE(file->f_dentry->d_inode);
687 struct clusterip_config *c = pde->data;
688 unsigned long nodenum;
690 if (copy_from_user(buffer, input, PROC_WRITELEN))
691 return -EFAULT;
693 if (*buffer == '+') {
694 nodenum = simple_strtoul(buffer+1, NULL, 10);
695 if (clusterip_add_node(c, nodenum))
696 return -ENOMEM;
697 } else if (*buffer == '-') {
698 nodenum = simple_strtoul(buffer+1, NULL,10);
699 if (clusterip_del_node(c, nodenum))
700 return -ENOENT;
701 } else
702 return -EIO;
704 return size;
707 static struct file_operations clusterip_proc_fops = {
708 .owner = THIS_MODULE,
709 .open = clusterip_proc_open,
710 .read = seq_read,
711 .write = clusterip_proc_write,
712 .llseek = seq_lseek,
713 .release = clusterip_proc_release,
716 #endif /* CONFIG_PROC_FS */
718 static int __init ipt_clusterip_init(void)
720 int ret;
722 ret = ipt_register_target(&clusterip_tgt);
723 if (ret < 0)
724 return ret;
726 ret = nf_register_hook(&cip_arp_ops);
727 if (ret < 0)
728 goto cleanup_target;
730 #ifdef CONFIG_PROC_FS
731 clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", proc_net);
732 if (!clusterip_procdir) {
733 printk(KERN_ERR "CLUSTERIP: Unable to proc dir entry\n");
734 ret = -ENOMEM;
735 goto cleanup_hook;
737 #endif /* CONFIG_PROC_FS */
739 printk(KERN_NOTICE "ClusterIP Version %s loaded successfully\n",
740 CLUSTERIP_VERSION);
741 return 0;
743 #ifdef CONFIG_PROC_FS
744 cleanup_hook:
745 nf_unregister_hook(&cip_arp_ops);
746 #endif /* CONFIG_PROC_FS */
747 cleanup_target:
748 ipt_unregister_target(&clusterip_tgt);
749 return ret;
752 static void __exit ipt_clusterip_fini(void)
754 printk(KERN_NOTICE "ClusterIP Version %s unloading\n",
755 CLUSTERIP_VERSION);
756 #ifdef CONFIG_PROC_FS
757 remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent);
758 #endif
759 nf_unregister_hook(&cip_arp_ops);
760 ipt_unregister_target(&clusterip_tgt);
763 module_init(ipt_clusterip_init);
764 module_exit(ipt_clusterip_fini);