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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/module.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/slab.h>
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>
24 #include <linux/seq_file.h>
25 #include <linux/netfilter_arp.h>
26 #include <linux/netfilter/x_tables.h>
27 #include <linux/netfilter_ipv4/ip_tables.h>
28 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/net_namespace.h>
31 #include <net/netns/generic.h>
32 #include <net/checksum.h>
33 #include <net/ip.h>
34
35 #define CLUSTERIP_VERSION "0.8"
36
37 MODULE_LICENSE("GPL");
38 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
39 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
40
41 struct clusterip_config {
42 struct list_head list; /* list of all configs */
43 atomic_t refcount; /* reference count */
44 atomic_t entries; /* number of entries/rules
45 * referencing us */
46
47 __be32 clusterip; /* the IP address */
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 */
51 unsigned long local_nodes; /* node number array */
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 */
58 struct rcu_head rcu;
59 };
60
61 #ifdef CONFIG_PROC_FS
62 static const struct file_operations clusterip_proc_fops;
63 #endif
64
65 static int clusterip_net_id __read_mostly;
66
67 struct clusterip_net {
68 struct list_head configs;
69 /* lock protects the configs list */
70 spinlock_t lock;
71
72 #ifdef CONFIG_PROC_FS
73 struct proc_dir_entry *procdir;
74 #endif
75 };
76
77 static inline void
clusterip_config_get(struct clusterip_config * c)78 clusterip_config_get(struct clusterip_config *c)
79 {
80 atomic_inc(&c->refcount);
81 }
82
83
clusterip_config_rcu_free(struct rcu_head * head)84 static void clusterip_config_rcu_free(struct rcu_head *head)
85 {
86 kfree(container_of(head, struct clusterip_config, rcu));
87 }
88
89 static inline void
clusterip_config_put(struct clusterip_config * c)90 clusterip_config_put(struct clusterip_config *c)
91 {
92 if (atomic_dec_and_test(&c->refcount))
93 call_rcu_bh(&c->rcu, clusterip_config_rcu_free);
94 }
95
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
clusterip_config_entry_put(struct clusterip_config * c)100 clusterip_config_entry_put(struct clusterip_config *c)
101 {
102 struct net *net = dev_net(c->dev);
103 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
104
105 local_bh_disable();
106 if (atomic_dec_and_lock(&c->entries, &cn->lock)) {
107 list_del_rcu(&c->list);
108 spin_unlock(&cn->lock);
109 local_bh_enable();
110
111 dev_mc_del(c->dev, c->clustermac);
112 dev_put(c->dev);
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
118 proc_remove(c->pde);
119 #endif
120 return;
121 }
122 local_bh_enable();
123 }
124
125 static struct clusterip_config *
__clusterip_config_find(struct net * net,__be32 clusterip)126 __clusterip_config_find(struct net *net, __be32 clusterip)
127 {
128 struct clusterip_config *c;
129 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
130
131 list_for_each_entry_rcu(c, &cn->configs, list) {
132 if (c->clusterip == clusterip)
133 return c;
134 }
135
136 return NULL;
137 }
138
139 static inline struct clusterip_config *
clusterip_config_find_get(struct net * net,__be32 clusterip,int entry)140 clusterip_config_find_get(struct net *net, __be32 clusterip, int entry)
141 {
142 struct clusterip_config *c;
143
144 rcu_read_lock_bh();
145 c = __clusterip_config_find(net, clusterip);
146 if (c) {
147 if (unlikely(!atomic_inc_not_zero(&c->refcount)))
148 c = NULL;
149 else if (entry)
150 atomic_inc(&c->entries);
151 }
152 rcu_read_unlock_bh();
153
154 return c;
155 }
156
157 static void
clusterip_config_init_nodelist(struct clusterip_config * c,const struct ipt_clusterip_tgt_info * i)158 clusterip_config_init_nodelist(struct clusterip_config *c,
159 const struct ipt_clusterip_tgt_info *i)
160 {
161 int n;
162
163 for (n = 0; n < i->num_local_nodes; n++)
164 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
165 }
166
167 static struct clusterip_config *
clusterip_config_init(const struct ipt_clusterip_tgt_info * i,__be32 ip,struct net_device * dev)168 clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip,
169 struct net_device *dev)
170 {
171 struct clusterip_config *c;
172 struct clusterip_net *cn = net_generic(dev_net(dev), clusterip_net_id);
173
174 c = kzalloc(sizeof(*c), GFP_ATOMIC);
175 if (!c)
176 return NULL;
177
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;
182 clusterip_config_init_nodelist(c, i);
183 c->hash_mode = i->hash_mode;
184 c->hash_initval = i->hash_initval;
185 atomic_set(&c->refcount, 1);
186 atomic_set(&c->entries, 1);
187
188 #ifdef CONFIG_PROC_FS
189 {
190 char buffer[16];
191
192 /* create proc dir entry */
193 sprintf(buffer, "%pI4", &ip);
194 c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR,
195 cn->procdir,
196 &clusterip_proc_fops, c);
197 if (!c->pde) {
198 kfree(c);
199 return NULL;
200 }
201 }
202 #endif
203
204 spin_lock_bh(&cn->lock);
205 list_add_rcu(&c->list, &cn->configs);
206 spin_unlock_bh(&cn->lock);
207
208 return c;
209 }
210
211 #ifdef CONFIG_PROC_FS
212 static int
clusterip_add_node(struct clusterip_config * c,u_int16_t nodenum)213 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
214 {
215
216 if (nodenum == 0 ||
217 nodenum > c->num_total_nodes)
218 return 1;
219
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;
223
224 return 0;
225 }
226
227 static bool
clusterip_del_node(struct clusterip_config * c,u_int16_t nodenum)228 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
229 {
230 if (nodenum == 0 ||
231 nodenum > c->num_total_nodes)
232 return true;
233
234 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
235 return false;
236
237 return true;
238 }
239 #endif
240
241 static inline u_int32_t
clusterip_hashfn(const struct sk_buff * skb,const struct clusterip_config * config)242 clusterip_hashfn(const struct sk_buff *skb,
243 const struct clusterip_config *config)
244 {
245 const struct iphdr *iph = ip_hdr(skb);
246 unsigned long hashval;
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 {
261 net_info_ratelimited("unknown protocol %u\n", iph->protocol);
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:
270 hashval = jhash_2words(ntohl(iph->saddr), sport,
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 */
282 pr_info("unknown mode %u\n", config->hash_mode);
283 BUG();
284 break;
285 }
286
287 /* node numbers are 1..n, not 0..n */
288 return reciprocal_scale(hashval, config->num_total_nodes) + 1;
289 }
290
291 static inline int
clusterip_responsible(const struct clusterip_config * config,u_int32_t hash)292 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
293 {
294 return test_bit(hash - 1, &config->local_nodes);
295 }
296
297 /***********************************************************************
298 * IPTABLES TARGET
299 ***********************************************************************/
300
301 static unsigned int
clusterip_tg(struct sk_buff * skb,const struct xt_action_param * par)302 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
303 {
304 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
305 struct nf_conn *ct;
306 enum ip_conntrack_info ctinfo;
307 u_int32_t hash;
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
313 ct = nf_ct_get(skb, &ctinfo);
314 if (ct == NULL)
315 return NF_DROP;
316
317 /* special case: ICMP error handling. conntrack distinguishes between
318 * error messages (RELATED) and information requests (see below) */
319 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
320 (ctinfo == IP_CT_RELATED ||
321 ctinfo == IP_CT_RELATED_REPLY))
322 return XT_CONTINUE;
323
324 /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO,
325 * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here
326 * on, which all have an ID field [relevant for hashing]. */
327
328 hash = clusterip_hashfn(skb, cipinfo->config);
329
330 switch (ctinfo) {
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;
344 }
345
346 #ifdef DEBUG
347 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
348 #endif
349 pr_debug("hash=%u ct_hash=%u ", hash, ct->mark);
350 if (!clusterip_responsible(cipinfo->config, hash)) {
351 pr_debug("not responsible\n");
352 return NF_DROP;
353 }
354 pr_debug("responsible\n");
355
356 /* despite being received via linklayer multicast, this is
357 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
358 skb->pkt_type = PACKET_HOST;
359
360 return XT_CONTINUE;
361 }
362
clusterip_tg_check(const struct xt_tgchk_param * par)363 static int clusterip_tg_check(const struct xt_tgchk_param *par)
364 {
365 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
366 const struct ipt_entry *e = par->entryinfo;
367 struct clusterip_config *config;
368 int ret, i;
369
370 if (par->nft_compat) {
371 pr_err("cannot use CLUSTERIP target from nftables compat\n");
372 return -EOPNOTSUPP;
373 }
374
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) {
378 pr_info("unknown mode %u\n", cipinfo->hash_mode);
379 return -EINVAL;
380
381 }
382 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
383 e->ip.dst.s_addr == 0) {
384 pr_info("Please specify destination IP\n");
385 return -EINVAL;
386 }
387 if (cipinfo->num_local_nodes > ARRAY_SIZE(cipinfo->local_nodes)) {
388 pr_info("bad num_local_nodes %u\n", cipinfo->num_local_nodes);
389 return -EINVAL;
390 }
391 for (i = 0; i < cipinfo->num_local_nodes; i++) {
392 if (cipinfo->local_nodes[i] - 1 >=
393 sizeof(config->local_nodes) * 8) {
394 pr_info("bad local_nodes[%d] %u\n",
395 i, cipinfo->local_nodes[i]);
396 return -EINVAL;
397 }
398 }
399
400 config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1);
401 if (!config) {
402 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
403 pr_info("no config found for %pI4, need 'new'\n",
404 &e->ip.dst.s_addr);
405 return -EINVAL;
406 } else {
407 struct net_device *dev;
408
409 if (e->ip.iniface[0] == '\0') {
410 pr_info("Please specify an interface name\n");
411 return -EINVAL;
412 }
413
414 dev = dev_get_by_name(par->net, e->ip.iniface);
415 if (!dev) {
416 pr_info("no such interface %s\n",
417 e->ip.iniface);
418 return -ENOENT;
419 }
420
421 config = clusterip_config_init(cipinfo,
422 e->ip.dst.s_addr, dev);
423 if (!config) {
424 dev_put(dev);
425 return -ENOMEM;
426 }
427 dev_mc_add(config->dev, config->clustermac);
428 }
429 }
430 cipinfo->config = config;
431
432 ret = nf_ct_l3proto_try_module_get(par->family);
433 if (ret < 0)
434 pr_info("cannot load conntrack support for proto=%u\n",
435 par->family);
436
437 if (!par->net->xt.clusterip_deprecated_warning) {
438 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
439 "use xt_cluster instead\n");
440 par->net->xt.clusterip_deprecated_warning = true;
441 }
442
443 return ret;
444 }
445
446 /* drop reference count of cluster config when rule is deleted */
clusterip_tg_destroy(const struct xt_tgdtor_param * par)447 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
448 {
449 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
450
451 /* if no more entries are referencing the config, remove it
452 * from the list and destroy the proc entry */
453 clusterip_config_entry_put(cipinfo->config);
454
455 clusterip_config_put(cipinfo->config);
456
457 nf_ct_l3proto_module_put(par->family);
458 }
459
460 #ifdef CONFIG_COMPAT
461 struct compat_ipt_clusterip_tgt_info
462 {
463 u_int32_t flags;
464 u_int8_t clustermac[6];
465 u_int16_t num_total_nodes;
466 u_int16_t num_local_nodes;
467 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
468 u_int32_t hash_mode;
469 u_int32_t hash_initval;
470 compat_uptr_t config;
471 };
472 #endif /* CONFIG_COMPAT */
473
474 static struct xt_target clusterip_tg_reg __read_mostly = {
475 .name = "CLUSTERIP",
476 .family = NFPROTO_IPV4,
477 .target = clusterip_tg,
478 .checkentry = clusterip_tg_check,
479 .destroy = clusterip_tg_destroy,
480 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
481 #ifdef CONFIG_COMPAT
482 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
483 #endif /* CONFIG_COMPAT */
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 __be32 src_ip;
496 u_int8_t dst_hw[ETH_ALEN];
497 __be32 dst_ip;
498 } __packed;
499
500 #ifdef DEBUG
arp_print(struct arp_payload * payload)501 static void arp_print(struct arp_payload *payload)
502 {
503 #define HBUFFERLEN 30
504 char hbuffer[HBUFFERLEN];
505 int j, k;
506
507 for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) {
508 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
509 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
510 hbuffer[k++] = ':';
511 }
512 hbuffer[--k] = '\0';
513
514 pr_debug("src %pI4@%s, dst %pI4\n",
515 &payload->src_ip, hbuffer, &payload->dst_ip);
516 }
517 #endif
518
519 static unsigned int
arp_mangle(void * priv,struct sk_buff * skb,const struct nf_hook_state * state)520 arp_mangle(void *priv,
521 struct sk_buff *skb,
522 const struct nf_hook_state *state)
523 {
524 struct arphdr *arp = arp_hdr(skb);
525 struct arp_payload *payload;
526 struct clusterip_config *c;
527 struct net *net = state->net;
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(net, 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 != state->out) {
553 pr_debug("not mangling arp reply on different "
554 "interface: cip'%s'-skb'%s'\n",
555 c->dev->name, state->out->name);
556 clusterip_config_put(c);
557 return NF_ACCEPT;
558 }
559
560 /* mangle reply hardware address */
561 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
562
563 #ifdef DEBUG
564 pr_debug("mangled arp reply: ");
565 arp_print(payload);
566 #endif
567
568 clusterip_config_put(c);
569
570 return NF_ACCEPT;
571 }
572
573 static struct nf_hook_ops cip_arp_ops __read_mostly = {
574 .hook = arp_mangle,
575 .pf = NFPROTO_ARP,
576 .hooknum = NF_ARP_OUT,
577 .priority = -1
578 };
579
580 /***********************************************************************
581 * PROC DIR HANDLING
582 ***********************************************************************/
583
584 #ifdef CONFIG_PROC_FS
585
586 struct clusterip_seq_position {
587 unsigned int pos; /* position */
588 unsigned int weight; /* number of bits set == size */
589 unsigned int bit; /* current bit */
590 unsigned long val; /* current value */
591 };
592
clusterip_seq_start(struct seq_file * s,loff_t * pos)593 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
594 {
595 struct clusterip_config *c = s->private;
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
clusterip_seq_next(struct seq_file * s,void * v,loff_t * pos)619 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
620 {
621 struct clusterip_seq_position *idx = 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
clusterip_seq_stop(struct seq_file * s,void * v)633 static void clusterip_seq_stop(struct seq_file *s, void *v)
634 {
635 if (!IS_ERR(v))
636 kfree(v);
637 }
638
clusterip_seq_show(struct seq_file * s,void * v)639 static int clusterip_seq_show(struct seq_file *s, void *v)
640 {
641 struct clusterip_seq_position *idx = v;
642
643 if (idx->pos != 0)
644 seq_putc(s, ',');
645
646 seq_printf(s, "%u", idx->bit);
647
648 if (idx->pos == idx->weight - 1)
649 seq_putc(s, '\n');
650
651 return 0;
652 }
653
654 static const struct seq_operations clusterip_seq_ops = {
655 .start = clusterip_seq_start,
656 .next = clusterip_seq_next,
657 .stop = clusterip_seq_stop,
658 .show = clusterip_seq_show,
659 };
660
clusterip_proc_open(struct inode * inode,struct file * file)661 static int clusterip_proc_open(struct inode *inode, struct file *file)
662 {
663 int ret = seq_open(file, &clusterip_seq_ops);
664
665 if (!ret) {
666 struct seq_file *sf = file->private_data;
667 struct clusterip_config *c = PDE_DATA(inode);
668
669 sf->private = c;
670
671 clusterip_config_get(c);
672 }
673
674 return ret;
675 }
676
clusterip_proc_release(struct inode * inode,struct file * file)677 static int clusterip_proc_release(struct inode *inode, struct file *file)
678 {
679 struct clusterip_config *c = PDE_DATA(inode);
680 int ret;
681
682 ret = seq_release(inode, file);
683
684 if (!ret)
685 clusterip_config_put(c);
686
687 return ret;
688 }
689
clusterip_proc_write(struct file * file,const char __user * input,size_t size,loff_t * ofs)690 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
691 size_t size, loff_t *ofs)
692 {
693 struct clusterip_config *c = PDE_DATA(file_inode(file));
694 #define PROC_WRITELEN 10
695 char buffer[PROC_WRITELEN+1];
696 unsigned long nodenum;
697 int rc;
698
699 if (size > PROC_WRITELEN)
700 return -EIO;
701 if (copy_from_user(buffer, input, size))
702 return -EFAULT;
703 buffer[size] = 0;
704
705 if (*buffer == '+') {
706 rc = kstrtoul(buffer+1, 10, &nodenum);
707 if (rc)
708 return rc;
709 if (clusterip_add_node(c, nodenum))
710 return -ENOMEM;
711 } else if (*buffer == '-') {
712 rc = kstrtoul(buffer+1, 10, &nodenum);
713 if (rc)
714 return rc;
715 if (clusterip_del_node(c, nodenum))
716 return -ENOENT;
717 } else
718 return -EIO;
719
720 return size;
721 }
722
723 static const struct file_operations clusterip_proc_fops = {
724 .owner = THIS_MODULE,
725 .open = clusterip_proc_open,
726 .read = seq_read,
727 .write = clusterip_proc_write,
728 .llseek = seq_lseek,
729 .release = clusterip_proc_release,
730 };
731
732 #endif /* CONFIG_PROC_FS */
733
clusterip_net_init(struct net * net)734 static int clusterip_net_init(struct net *net)
735 {
736 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
737
738 INIT_LIST_HEAD(&cn->configs);
739
740 spin_lock_init(&cn->lock);
741
742 #ifdef CONFIG_PROC_FS
743 cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net);
744 if (!cn->procdir) {
745 pr_err("Unable to proc dir entry\n");
746 return -ENOMEM;
747 }
748 #endif /* CONFIG_PROC_FS */
749
750 return 0;
751 }
752
clusterip_net_exit(struct net * net)753 static void clusterip_net_exit(struct net *net)
754 {
755 #ifdef CONFIG_PROC_FS
756 struct clusterip_net *cn = net_generic(net, clusterip_net_id);
757 proc_remove(cn->procdir);
758 #endif
759 }
760
761 static struct pernet_operations clusterip_net_ops = {
762 .init = clusterip_net_init,
763 .exit = clusterip_net_exit,
764 .id = &clusterip_net_id,
765 .size = sizeof(struct clusterip_net),
766 };
767
clusterip_tg_init(void)768 static int __init clusterip_tg_init(void)
769 {
770 int ret;
771
772 ret = register_pernet_subsys(&clusterip_net_ops);
773 if (ret < 0)
774 return ret;
775
776 ret = xt_register_target(&clusterip_tg_reg);
777 if (ret < 0)
778 goto cleanup_subsys;
779
780 ret = nf_register_hook(&cip_arp_ops);
781 if (ret < 0)
782 goto cleanup_target;
783
784 pr_info("ClusterIP Version %s loaded successfully\n",
785 CLUSTERIP_VERSION);
786
787 return 0;
788
789 cleanup_target:
790 xt_unregister_target(&clusterip_tg_reg);
791 cleanup_subsys:
792 unregister_pernet_subsys(&clusterip_net_ops);
793 return ret;
794 }
795
clusterip_tg_exit(void)796 static void __exit clusterip_tg_exit(void)
797 {
798 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
799
800 nf_unregister_hook(&cip_arp_ops);
801 xt_unregister_target(&clusterip_tg_reg);
802 unregister_pernet_subsys(&clusterip_net_ops);
803
804 /* Wait for completion of call_rcu_bh()'s (clusterip_config_rcu_free) */
805 rcu_barrier_bh();
806 }
807
808 module_init(clusterip_tg_init);
809 module_exit(clusterip_tg_exit);
810