1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Cluster IP hashmark target
3 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
4 * based on ideas of Fabio Olive Leite <olive@unixforge.org>
5 *
6 * Development of this code funded by SuSE Linux AG, https://www.suse.com/
7 */
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/module.h>
10 #include <linux/proc_fs.h>
11 #include <linux/jhash.h>
12 #include <linux/bitops.h>
13 #include <linux/skbuff.h>
14 #include <linux/slab.h>
15 #include <linux/ip.h>
16 #include <linux/tcp.h>
17 #include <linux/udp.h>
18 #include <linux/icmp.h>
19 #include <linux/if_arp.h>
20 #include <linux/seq_file.h>
21 #include <linux/refcount.h>
22 #include <linux/netfilter_arp.h>
23 #include <linux/netfilter/x_tables.h>
24 #include <linux/netfilter_ipv4/ip_tables.h>
25 #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h>
26 #include <net/netfilter/nf_conntrack.h>
27 #include <net/net_namespace.h>
28 #include <net/netns/generic.h>
29 #include <net/checksum.h>
30 #include <net/ip.h>
31
32 #define CLUSTERIP_VERSION "0.8"
33
34 MODULE_LICENSE("GPL");
35 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
36 MODULE_DESCRIPTION("Xtables: CLUSTERIP target");
37
38 struct clusterip_config {
39 struct list_head list; /* list of all configs */
40 refcount_t refcount; /* reference count */
41 refcount_t entries; /* number of entries/rules
42 * referencing us */
43
44 __be32 clusterip; /* the IP address */
45 u_int8_t clustermac[ETH_ALEN]; /* the MAC address */
46 int ifindex; /* device ifindex */
47 u_int16_t num_total_nodes; /* total number of nodes */
48 unsigned long local_nodes; /* node number array */
49
50 #ifdef CONFIG_PROC_FS
51 struct proc_dir_entry *pde; /* proc dir entry */
52 #endif
53 enum clusterip_hashmode hash_mode; /* which hashing mode */
54 u_int32_t hash_initval; /* hash initialization */
55 struct rcu_head rcu; /* for call_rcu */
56 struct net *net; /* netns for pernet list */
57 char ifname[IFNAMSIZ]; /* device ifname */
58 };
59
60 #ifdef CONFIG_PROC_FS
61 static const struct proc_ops clusterip_proc_ops;
62 #endif
63
64 struct clusterip_net {
65 struct list_head configs;
66 /* lock protects the configs list */
67 spinlock_t lock;
68
69 #ifdef CONFIG_PROC_FS
70 struct proc_dir_entry *procdir;
71 /* mutex protects the config->pde*/
72 struct mutex mutex;
73 #endif
74 };
75
76 static unsigned int clusterip_net_id __read_mostly;
clusterip_pernet(struct net * net)77 static inline struct clusterip_net *clusterip_pernet(struct net *net)
78 {
79 return net_generic(net, clusterip_net_id);
80 }
81
82 static inline void
clusterip_config_get(struct clusterip_config * c)83 clusterip_config_get(struct clusterip_config *c)
84 {
85 refcount_inc(&c->refcount);
86 }
87
clusterip_config_rcu_free(struct rcu_head * head)88 static void clusterip_config_rcu_free(struct rcu_head *head)
89 {
90 struct clusterip_config *config;
91 struct net_device *dev;
92
93 config = container_of(head, struct clusterip_config, rcu);
94 dev = dev_get_by_name(config->net, config->ifname);
95 if (dev) {
96 dev_mc_del(dev, config->clustermac);
97 dev_put(dev);
98 }
99 kfree(config);
100 }
101
102 static inline void
clusterip_config_put(struct clusterip_config * c)103 clusterip_config_put(struct clusterip_config *c)
104 {
105 if (refcount_dec_and_test(&c->refcount))
106 call_rcu(&c->rcu, clusterip_config_rcu_free);
107 }
108
109 /* decrease the count of entries using/referencing this config. If last
110 * entry(rule) is removed, remove the config from lists, but don't free it
111 * yet, since proc-files could still be holding references */
112 static inline void
clusterip_config_entry_put(struct clusterip_config * c)113 clusterip_config_entry_put(struct clusterip_config *c)
114 {
115 struct clusterip_net *cn = clusterip_pernet(c->net);
116
117 local_bh_disable();
118 if (refcount_dec_and_lock(&c->entries, &cn->lock)) {
119 list_del_rcu(&c->list);
120 spin_unlock(&cn->lock);
121 local_bh_enable();
122 /* In case anyone still accesses the file, the open/close
123 * functions are also incrementing the refcount on their own,
124 * so it's safe to remove the entry even if it's in use. */
125 #ifdef CONFIG_PROC_FS
126 mutex_lock(&cn->mutex);
127 if (cn->procdir)
128 proc_remove(c->pde);
129 mutex_unlock(&cn->mutex);
130 #endif
131 return;
132 }
133 local_bh_enable();
134 }
135
136 static struct clusterip_config *
__clusterip_config_find(struct net * net,__be32 clusterip)137 __clusterip_config_find(struct net *net, __be32 clusterip)
138 {
139 struct clusterip_config *c;
140 struct clusterip_net *cn = clusterip_pernet(net);
141
142 list_for_each_entry_rcu(c, &cn->configs, list) {
143 if (c->clusterip == clusterip)
144 return c;
145 }
146
147 return NULL;
148 }
149
150 static inline struct clusterip_config *
clusterip_config_find_get(struct net * net,__be32 clusterip,int entry)151 clusterip_config_find_get(struct net *net, __be32 clusterip, int entry)
152 {
153 struct clusterip_config *c;
154
155 rcu_read_lock_bh();
156 c = __clusterip_config_find(net, clusterip);
157 if (c) {
158 #ifdef CONFIG_PROC_FS
159 if (!c->pde)
160 c = NULL;
161 else
162 #endif
163 if (unlikely(!refcount_inc_not_zero(&c->refcount)))
164 c = NULL;
165 else if (entry) {
166 if (unlikely(!refcount_inc_not_zero(&c->entries))) {
167 clusterip_config_put(c);
168 c = NULL;
169 }
170 }
171 }
172 rcu_read_unlock_bh();
173
174 return c;
175 }
176
177 static void
clusterip_config_init_nodelist(struct clusterip_config * c,const struct ipt_clusterip_tgt_info * i)178 clusterip_config_init_nodelist(struct clusterip_config *c,
179 const struct ipt_clusterip_tgt_info *i)
180 {
181 int n;
182
183 for (n = 0; n < i->num_local_nodes; n++)
184 set_bit(i->local_nodes[n] - 1, &c->local_nodes);
185 }
186
187 static int
clusterip_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)188 clusterip_netdev_event(struct notifier_block *this, unsigned long event,
189 void *ptr)
190 {
191 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
192 struct net *net = dev_net(dev);
193 struct clusterip_net *cn = clusterip_pernet(net);
194 struct clusterip_config *c;
195
196 spin_lock_bh(&cn->lock);
197 list_for_each_entry_rcu(c, &cn->configs, list) {
198 switch (event) {
199 case NETDEV_REGISTER:
200 if (!strcmp(dev->name, c->ifname)) {
201 c->ifindex = dev->ifindex;
202 dev_mc_add(dev, c->clustermac);
203 }
204 break;
205 case NETDEV_UNREGISTER:
206 if (dev->ifindex == c->ifindex) {
207 dev_mc_del(dev, c->clustermac);
208 c->ifindex = -1;
209 }
210 break;
211 case NETDEV_CHANGENAME:
212 if (!strcmp(dev->name, c->ifname)) {
213 c->ifindex = dev->ifindex;
214 dev_mc_add(dev, c->clustermac);
215 } else if (dev->ifindex == c->ifindex) {
216 dev_mc_del(dev, c->clustermac);
217 c->ifindex = -1;
218 }
219 break;
220 }
221 }
222 spin_unlock_bh(&cn->lock);
223
224 return NOTIFY_DONE;
225 }
226
227 static struct clusterip_config *
clusterip_config_init(struct net * net,const struct ipt_clusterip_tgt_info * i,__be32 ip,const char * iniface)228 clusterip_config_init(struct net *net, const struct ipt_clusterip_tgt_info *i,
229 __be32 ip, const char *iniface)
230 {
231 struct clusterip_net *cn = clusterip_pernet(net);
232 struct clusterip_config *c;
233 struct net_device *dev;
234 int err;
235
236 if (iniface[0] == '\0') {
237 pr_info("Please specify an interface name\n");
238 return ERR_PTR(-EINVAL);
239 }
240
241 c = kzalloc(sizeof(*c), GFP_ATOMIC);
242 if (!c)
243 return ERR_PTR(-ENOMEM);
244
245 dev = dev_get_by_name(net, iniface);
246 if (!dev) {
247 pr_info("no such interface %s\n", iniface);
248 kfree(c);
249 return ERR_PTR(-ENOENT);
250 }
251 c->ifindex = dev->ifindex;
252 strcpy(c->ifname, dev->name);
253 memcpy(&c->clustermac, &i->clustermac, ETH_ALEN);
254 dev_mc_add(dev, c->clustermac);
255 dev_put(dev);
256
257 c->clusterip = ip;
258 c->num_total_nodes = i->num_total_nodes;
259 clusterip_config_init_nodelist(c, i);
260 c->hash_mode = i->hash_mode;
261 c->hash_initval = i->hash_initval;
262 c->net = net;
263 refcount_set(&c->refcount, 1);
264
265 spin_lock_bh(&cn->lock);
266 if (__clusterip_config_find(net, ip)) {
267 err = -EBUSY;
268 goto out_config_put;
269 }
270
271 list_add_rcu(&c->list, &cn->configs);
272 spin_unlock_bh(&cn->lock);
273
274 #ifdef CONFIG_PROC_FS
275 {
276 char buffer[16];
277
278 /* create proc dir entry */
279 sprintf(buffer, "%pI4", &ip);
280 mutex_lock(&cn->mutex);
281 c->pde = proc_create_data(buffer, 0600,
282 cn->procdir,
283 &clusterip_proc_ops, c);
284 mutex_unlock(&cn->mutex);
285 if (!c->pde) {
286 err = -ENOMEM;
287 goto err;
288 }
289 }
290 #endif
291
292 refcount_set(&c->entries, 1);
293 return c;
294
295 #ifdef CONFIG_PROC_FS
296 err:
297 #endif
298 spin_lock_bh(&cn->lock);
299 list_del_rcu(&c->list);
300 out_config_put:
301 spin_unlock_bh(&cn->lock);
302 clusterip_config_put(c);
303 return ERR_PTR(err);
304 }
305
306 #ifdef CONFIG_PROC_FS
307 static int
clusterip_add_node(struct clusterip_config * c,u_int16_t nodenum)308 clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum)
309 {
310
311 if (nodenum == 0 ||
312 nodenum > c->num_total_nodes)
313 return 1;
314
315 /* check if we already have this number in our bitfield */
316 if (test_and_set_bit(nodenum - 1, &c->local_nodes))
317 return 1;
318
319 return 0;
320 }
321
322 static bool
clusterip_del_node(struct clusterip_config * c,u_int16_t nodenum)323 clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum)
324 {
325 if (nodenum == 0 ||
326 nodenum > c->num_total_nodes)
327 return true;
328
329 if (test_and_clear_bit(nodenum - 1, &c->local_nodes))
330 return false;
331
332 return true;
333 }
334 #endif
335
336 static inline u_int32_t
clusterip_hashfn(const struct sk_buff * skb,const struct clusterip_config * config)337 clusterip_hashfn(const struct sk_buff *skb,
338 const struct clusterip_config *config)
339 {
340 const struct iphdr *iph = ip_hdr(skb);
341 unsigned long hashval;
342 u_int16_t sport = 0, dport = 0;
343 int poff;
344
345 poff = proto_ports_offset(iph->protocol);
346 if (poff >= 0) {
347 const u_int16_t *ports;
348 u16 _ports[2];
349
350 ports = skb_header_pointer(skb, iph->ihl * 4 + poff, 4, _ports);
351 if (ports) {
352 sport = ports[0];
353 dport = ports[1];
354 }
355 } else {
356 net_info_ratelimited("unknown protocol %u\n", iph->protocol);
357 }
358
359 switch (config->hash_mode) {
360 case CLUSTERIP_HASHMODE_SIP:
361 hashval = jhash_1word(ntohl(iph->saddr),
362 config->hash_initval);
363 break;
364 case CLUSTERIP_HASHMODE_SIP_SPT:
365 hashval = jhash_2words(ntohl(iph->saddr), sport,
366 config->hash_initval);
367 break;
368 case CLUSTERIP_HASHMODE_SIP_SPT_DPT:
369 hashval = jhash_3words(ntohl(iph->saddr), sport, dport,
370 config->hash_initval);
371 break;
372 default:
373 /* to make gcc happy */
374 hashval = 0;
375 /* This cannot happen, unless the check function wasn't called
376 * at rule load time */
377 pr_info("unknown mode %u\n", config->hash_mode);
378 BUG();
379 break;
380 }
381
382 /* node numbers are 1..n, not 0..n */
383 return reciprocal_scale(hashval, config->num_total_nodes) + 1;
384 }
385
386 static inline int
clusterip_responsible(const struct clusterip_config * config,u_int32_t hash)387 clusterip_responsible(const struct clusterip_config *config, u_int32_t hash)
388 {
389 return test_bit(hash - 1, &config->local_nodes);
390 }
391
392 /***********************************************************************
393 * IPTABLES TARGET
394 ***********************************************************************/
395
396 static unsigned int
clusterip_tg(struct sk_buff * skb,const struct xt_action_param * par)397 clusterip_tg(struct sk_buff *skb, const struct xt_action_param *par)
398 {
399 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
400 struct nf_conn *ct;
401 enum ip_conntrack_info ctinfo;
402 u_int32_t hash;
403
404 /* don't need to clusterip_config_get() here, since refcount
405 * is only decremented by destroy() - and ip_tables guarantees
406 * that the ->target() function isn't called after ->destroy() */
407
408 ct = nf_ct_get(skb, &ctinfo);
409 if (ct == NULL)
410 return NF_DROP;
411
412 /* special case: ICMP error handling. conntrack distinguishes between
413 * error messages (RELATED) and information requests (see below) */
414 if (ip_hdr(skb)->protocol == IPPROTO_ICMP &&
415 (ctinfo == IP_CT_RELATED ||
416 ctinfo == IP_CT_RELATED_REPLY))
417 return XT_CONTINUE;
418
419 /* nf_conntrack_proto_icmp guarantees us that we only have ICMP_ECHO,
420 * TIMESTAMP, INFO_REQUEST or ICMP_ADDRESS type icmp packets from here
421 * on, which all have an ID field [relevant for hashing]. */
422
423 hash = clusterip_hashfn(skb, cipinfo->config);
424
425 switch (ctinfo) {
426 case IP_CT_NEW:
427 WRITE_ONCE(ct->mark, hash);
428 break;
429 case IP_CT_RELATED:
430 case IP_CT_RELATED_REPLY:
431 /* FIXME: we don't handle expectations at the moment.
432 * They can arrive on a different node than
433 * the master connection (e.g. FTP passive mode) */
434 case IP_CT_ESTABLISHED:
435 case IP_CT_ESTABLISHED_REPLY:
436 break;
437 default: /* Prevent gcc warnings */
438 break;
439 }
440
441 #ifdef DEBUG
442 nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
443 #endif
444 pr_debug("hash=%u ct_hash=%u ", hash, READ_ONCE(ct->mark));
445 if (!clusterip_responsible(cipinfo->config, hash)) {
446 pr_debug("not responsible\n");
447 return NF_DROP;
448 }
449 pr_debug("responsible\n");
450
451 /* despite being received via linklayer multicast, this is
452 * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */
453 skb->pkt_type = PACKET_HOST;
454
455 return XT_CONTINUE;
456 }
457
clusterip_tg_check(const struct xt_tgchk_param * par)458 static int clusterip_tg_check(const struct xt_tgchk_param *par)
459 {
460 struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
461 const struct ipt_entry *e = par->entryinfo;
462 struct clusterip_config *config;
463 int ret, i;
464
465 if (par->nft_compat) {
466 pr_err("cannot use CLUSTERIP target from nftables compat\n");
467 return -EOPNOTSUPP;
468 }
469
470 if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP &&
471 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT &&
472 cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) {
473 pr_info("unknown mode %u\n", cipinfo->hash_mode);
474 return -EINVAL;
475
476 }
477 if (e->ip.dmsk.s_addr != htonl(0xffffffff) ||
478 e->ip.dst.s_addr == 0) {
479 pr_info("Please specify destination IP\n");
480 return -EINVAL;
481 }
482 if (cipinfo->num_local_nodes > ARRAY_SIZE(cipinfo->local_nodes)) {
483 pr_info("bad num_local_nodes %u\n", cipinfo->num_local_nodes);
484 return -EINVAL;
485 }
486 for (i = 0; i < cipinfo->num_local_nodes; i++) {
487 if (cipinfo->local_nodes[i] - 1 >=
488 sizeof(config->local_nodes) * 8) {
489 pr_info("bad local_nodes[%d] %u\n",
490 i, cipinfo->local_nodes[i]);
491 return -EINVAL;
492 }
493 }
494
495 config = clusterip_config_find_get(par->net, e->ip.dst.s_addr, 1);
496 if (!config) {
497 if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) {
498 pr_info("no config found for %pI4, need 'new'\n",
499 &e->ip.dst.s_addr);
500 return -EINVAL;
501 } else {
502 config = clusterip_config_init(par->net, cipinfo,
503 e->ip.dst.s_addr,
504 e->ip.iniface);
505 if (IS_ERR(config))
506 return PTR_ERR(config);
507 }
508 } else if (memcmp(&config->clustermac, &cipinfo->clustermac, ETH_ALEN)) {
509 clusterip_config_entry_put(config);
510 clusterip_config_put(config);
511 return -EINVAL;
512 }
513
514 ret = nf_ct_netns_get(par->net, par->family);
515 if (ret < 0) {
516 pr_info("cannot load conntrack support for proto=%u\n",
517 par->family);
518 clusterip_config_entry_put(config);
519 clusterip_config_put(config);
520 return ret;
521 }
522
523 if (!par->net->xt.clusterip_deprecated_warning) {
524 pr_info("ipt_CLUSTERIP is deprecated and it will removed soon, "
525 "use xt_cluster instead\n");
526 par->net->xt.clusterip_deprecated_warning = true;
527 }
528
529 cipinfo->config = config;
530 return ret;
531 }
532
533 /* drop reference count of cluster config when rule is deleted */
clusterip_tg_destroy(const struct xt_tgdtor_param * par)534 static void clusterip_tg_destroy(const struct xt_tgdtor_param *par)
535 {
536 const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo;
537
538 /* if no more entries are referencing the config, remove it
539 * from the list and destroy the proc entry */
540 clusterip_config_entry_put(cipinfo->config);
541
542 clusterip_config_put(cipinfo->config);
543
544 nf_ct_netns_put(par->net, par->family);
545 }
546
547 #ifdef CONFIG_COMPAT
548 struct compat_ipt_clusterip_tgt_info
549 {
550 u_int32_t flags;
551 u_int8_t clustermac[6];
552 u_int16_t num_total_nodes;
553 u_int16_t num_local_nodes;
554 u_int16_t local_nodes[CLUSTERIP_MAX_NODES];
555 u_int32_t hash_mode;
556 u_int32_t hash_initval;
557 compat_uptr_t config;
558 };
559 #endif /* CONFIG_COMPAT */
560
561 static struct xt_target clusterip_tg_reg __read_mostly = {
562 .name = "CLUSTERIP",
563 .family = NFPROTO_IPV4,
564 .target = clusterip_tg,
565 .checkentry = clusterip_tg_check,
566 .destroy = clusterip_tg_destroy,
567 .targetsize = sizeof(struct ipt_clusterip_tgt_info),
568 .usersize = offsetof(struct ipt_clusterip_tgt_info, config),
569 #ifdef CONFIG_COMPAT
570 .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info),
571 #endif /* CONFIG_COMPAT */
572 .me = THIS_MODULE
573 };
574
575
576 /***********************************************************************
577 * ARP MANGLING CODE
578 ***********************************************************************/
579
580 /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */
581 struct arp_payload {
582 u_int8_t src_hw[ETH_ALEN];
583 __be32 src_ip;
584 u_int8_t dst_hw[ETH_ALEN];
585 __be32 dst_ip;
586 } __packed;
587
588 #ifdef DEBUG
arp_print(struct arp_payload * payload)589 static void arp_print(struct arp_payload *payload)
590 {
591 #define HBUFFERLEN 30
592 char hbuffer[HBUFFERLEN];
593 int j, k;
594
595 for (k = 0, j = 0; k < HBUFFERLEN - 3 && j < ETH_ALEN; j++) {
596 hbuffer[k++] = hex_asc_hi(payload->src_hw[j]);
597 hbuffer[k++] = hex_asc_lo(payload->src_hw[j]);
598 hbuffer[k++] = ':';
599 }
600 hbuffer[--k] = '\0';
601
602 pr_debug("src %pI4@%s, dst %pI4\n",
603 &payload->src_ip, hbuffer, &payload->dst_ip);
604 }
605 #endif
606
607 static unsigned int
arp_mangle(void * priv,struct sk_buff * skb,const struct nf_hook_state * state)608 arp_mangle(void *priv,
609 struct sk_buff *skb,
610 const struct nf_hook_state *state)
611 {
612 struct arphdr *arp = arp_hdr(skb);
613 struct arp_payload *payload;
614 struct clusterip_config *c;
615 struct net *net = state->net;
616
617 /* we don't care about non-ethernet and non-ipv4 ARP */
618 if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
619 arp->ar_pro != htons(ETH_P_IP) ||
620 arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN)
621 return NF_ACCEPT;
622
623 /* we only want to mangle arp requests and replies */
624 if (arp->ar_op != htons(ARPOP_REPLY) &&
625 arp->ar_op != htons(ARPOP_REQUEST))
626 return NF_ACCEPT;
627
628 payload = (void *)(arp+1);
629
630 /* if there is no clusterip configuration for the arp reply's
631 * source ip, we don't want to mangle it */
632 c = clusterip_config_find_get(net, payload->src_ip, 0);
633 if (!c)
634 return NF_ACCEPT;
635
636 /* normally the linux kernel always replies to arp queries of
637 * addresses on different interfacs. However, in the CLUSTERIP case
638 * this wouldn't work, since we didn't subscribe the mcast group on
639 * other interfaces */
640 if (c->ifindex != state->out->ifindex) {
641 pr_debug("not mangling arp reply on different interface: cip'%d'-skb'%d'\n",
642 c->ifindex, state->out->ifindex);
643 clusterip_config_put(c);
644 return NF_ACCEPT;
645 }
646
647 /* mangle reply hardware address */
648 memcpy(payload->src_hw, c->clustermac, arp->ar_hln);
649
650 #ifdef DEBUG
651 pr_debug("mangled arp reply: ");
652 arp_print(payload);
653 #endif
654
655 clusterip_config_put(c);
656
657 return NF_ACCEPT;
658 }
659
660 static const struct nf_hook_ops cip_arp_ops = {
661 .hook = arp_mangle,
662 .pf = NFPROTO_ARP,
663 .hooknum = NF_ARP_OUT,
664 .priority = -1
665 };
666
667 /***********************************************************************
668 * PROC DIR HANDLING
669 ***********************************************************************/
670
671 #ifdef CONFIG_PROC_FS
672
673 struct clusterip_seq_position {
674 unsigned int pos; /* position */
675 unsigned int weight; /* number of bits set == size */
676 unsigned int bit; /* current bit */
677 unsigned long val; /* current value */
678 };
679
clusterip_seq_start(struct seq_file * s,loff_t * pos)680 static void *clusterip_seq_start(struct seq_file *s, loff_t *pos)
681 {
682 struct clusterip_config *c = s->private;
683 unsigned int weight;
684 u_int32_t local_nodes;
685 struct clusterip_seq_position *idx;
686
687 /* FIXME: possible race */
688 local_nodes = c->local_nodes;
689 weight = hweight32(local_nodes);
690 if (*pos >= weight)
691 return NULL;
692
693 idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL);
694 if (!idx)
695 return ERR_PTR(-ENOMEM);
696
697 idx->pos = *pos;
698 idx->weight = weight;
699 idx->bit = ffs(local_nodes);
700 idx->val = local_nodes;
701 clear_bit(idx->bit - 1, &idx->val);
702
703 return idx;
704 }
705
clusterip_seq_next(struct seq_file * s,void * v,loff_t * pos)706 static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos)
707 {
708 struct clusterip_seq_position *idx = v;
709
710 *pos = ++idx->pos;
711 if (*pos >= idx->weight) {
712 kfree(v);
713 return NULL;
714 }
715 idx->bit = ffs(idx->val);
716 clear_bit(idx->bit - 1, &idx->val);
717 return idx;
718 }
719
clusterip_seq_stop(struct seq_file * s,void * v)720 static void clusterip_seq_stop(struct seq_file *s, void *v)
721 {
722 if (!IS_ERR(v))
723 kfree(v);
724 }
725
clusterip_seq_show(struct seq_file * s,void * v)726 static int clusterip_seq_show(struct seq_file *s, void *v)
727 {
728 struct clusterip_seq_position *idx = v;
729
730 if (idx->pos != 0)
731 seq_putc(s, ',');
732
733 seq_printf(s, "%u", idx->bit);
734
735 if (idx->pos == idx->weight - 1)
736 seq_putc(s, '\n');
737
738 return 0;
739 }
740
741 static const struct seq_operations clusterip_seq_ops = {
742 .start = clusterip_seq_start,
743 .next = clusterip_seq_next,
744 .stop = clusterip_seq_stop,
745 .show = clusterip_seq_show,
746 };
747
clusterip_proc_open(struct inode * inode,struct file * file)748 static int clusterip_proc_open(struct inode *inode, struct file *file)
749 {
750 int ret = seq_open(file, &clusterip_seq_ops);
751
752 if (!ret) {
753 struct seq_file *sf = file->private_data;
754 struct clusterip_config *c = PDE_DATA(inode);
755
756 sf->private = c;
757
758 clusterip_config_get(c);
759 }
760
761 return ret;
762 }
763
clusterip_proc_release(struct inode * inode,struct file * file)764 static int clusterip_proc_release(struct inode *inode, struct file *file)
765 {
766 struct clusterip_config *c = PDE_DATA(inode);
767 int ret;
768
769 ret = seq_release(inode, file);
770
771 if (!ret)
772 clusterip_config_put(c);
773
774 return ret;
775 }
776
clusterip_proc_write(struct file * file,const char __user * input,size_t size,loff_t * ofs)777 static ssize_t clusterip_proc_write(struct file *file, const char __user *input,
778 size_t size, loff_t *ofs)
779 {
780 struct clusterip_config *c = PDE_DATA(file_inode(file));
781 #define PROC_WRITELEN 10
782 char buffer[PROC_WRITELEN+1];
783 unsigned long nodenum;
784 int rc;
785
786 if (size > PROC_WRITELEN)
787 return -EIO;
788 if (copy_from_user(buffer, input, size))
789 return -EFAULT;
790 buffer[size] = 0;
791
792 if (*buffer == '+') {
793 rc = kstrtoul(buffer+1, 10, &nodenum);
794 if (rc)
795 return rc;
796 if (clusterip_add_node(c, nodenum))
797 return -ENOMEM;
798 } else if (*buffer == '-') {
799 rc = kstrtoul(buffer+1, 10, &nodenum);
800 if (rc)
801 return rc;
802 if (clusterip_del_node(c, nodenum))
803 return -ENOENT;
804 } else
805 return -EIO;
806
807 return size;
808 }
809
810 static const struct proc_ops clusterip_proc_ops = {
811 .proc_open = clusterip_proc_open,
812 .proc_read = seq_read,
813 .proc_write = clusterip_proc_write,
814 .proc_lseek = seq_lseek,
815 .proc_release = clusterip_proc_release,
816 };
817
818 #endif /* CONFIG_PROC_FS */
819
clusterip_net_init(struct net * net)820 static int clusterip_net_init(struct net *net)
821 {
822 struct clusterip_net *cn = clusterip_pernet(net);
823 int ret;
824
825 INIT_LIST_HEAD(&cn->configs);
826
827 spin_lock_init(&cn->lock);
828
829 ret = nf_register_net_hook(net, &cip_arp_ops);
830 if (ret < 0)
831 return ret;
832
833 #ifdef CONFIG_PROC_FS
834 cn->procdir = proc_mkdir("ipt_CLUSTERIP", net->proc_net);
835 if (!cn->procdir) {
836 nf_unregister_net_hook(net, &cip_arp_ops);
837 pr_err("Unable to proc dir entry\n");
838 return -ENOMEM;
839 }
840 mutex_init(&cn->mutex);
841 #endif /* CONFIG_PROC_FS */
842
843 return 0;
844 }
845
clusterip_net_exit(struct net * net)846 static void clusterip_net_exit(struct net *net)
847 {
848 #ifdef CONFIG_PROC_FS
849 struct clusterip_net *cn = clusterip_pernet(net);
850
851 mutex_lock(&cn->mutex);
852 proc_remove(cn->procdir);
853 cn->procdir = NULL;
854 mutex_unlock(&cn->mutex);
855 #endif
856 nf_unregister_net_hook(net, &cip_arp_ops);
857 }
858
859 static struct pernet_operations clusterip_net_ops = {
860 .init = clusterip_net_init,
861 .exit = clusterip_net_exit,
862 .id = &clusterip_net_id,
863 .size = sizeof(struct clusterip_net),
864 };
865
866 static struct notifier_block cip_netdev_notifier = {
867 .notifier_call = clusterip_netdev_event
868 };
869
clusterip_tg_init(void)870 static int __init clusterip_tg_init(void)
871 {
872 int ret;
873
874 ret = register_pernet_subsys(&clusterip_net_ops);
875 if (ret < 0)
876 return ret;
877
878 ret = xt_register_target(&clusterip_tg_reg);
879 if (ret < 0)
880 goto cleanup_subsys;
881
882 ret = register_netdevice_notifier(&cip_netdev_notifier);
883 if (ret < 0)
884 goto unregister_target;
885
886 pr_info("ClusterIP Version %s loaded successfully\n",
887 CLUSTERIP_VERSION);
888
889 return 0;
890
891 unregister_target:
892 xt_unregister_target(&clusterip_tg_reg);
893 cleanup_subsys:
894 unregister_pernet_subsys(&clusterip_net_ops);
895 return ret;
896 }
897
clusterip_tg_exit(void)898 static void __exit clusterip_tg_exit(void)
899 {
900 pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION);
901
902 unregister_netdevice_notifier(&cip_netdev_notifier);
903 xt_unregister_target(&clusterip_tg_reg);
904 unregister_pernet_subsys(&clusterip_net_ops);
905
906 /* Wait for completion of call_rcu()'s (clusterip_config_rcu_free) */
907 rcu_barrier();
908 }
909
910 module_init(clusterip_tg_init);
911 module_exit(clusterip_tg_exit);
912