2 * ip_vs_app.c: Application module support for IPVS
4 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
11 * Most code here is taken from ip_masq_app.c in kernel 2.2. The difference
12 * is that ip_vs_app module handles the reverse direction (incoming requests
13 * and outgoing responses).
15 * IP_MASQ_APP application masquerading module
17 * Author: Juan Jose Ciarlante, <jjciarla@raiz.uncu.edu.ar>
21 #define KMSG_COMPONENT "IPVS"
22 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/skbuff.h>
29 #include <linux/netfilter.h>
30 #include <linux/slab.h>
31 #include <net/net_namespace.h>
32 #include <net/protocol.h>
34 #include <linux/stat.h>
35 #include <linux/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/mutex.h>
39 #include <net/ip_vs.h>
41 EXPORT_SYMBOL(register_ip_vs_app);
42 EXPORT_SYMBOL(unregister_ip_vs_app);
43 EXPORT_SYMBOL(register_ip_vs_app_inc);
45 static DEFINE_MUTEX(__ip_vs_app_mutex);
48 * Get an ip_vs_app object
50 static inline int ip_vs_app_get(struct ip_vs_app *app)
52 return try_module_get(app->module);
56 static inline void ip_vs_app_put(struct ip_vs_app *app)
58 module_put(app->module);
61 static void ip_vs_app_inc_destroy(struct ip_vs_app *inc)
63 kfree(inc->timeout_table);
67 static void ip_vs_app_inc_rcu_free(struct rcu_head *head)
69 struct ip_vs_app *inc = container_of(head, struct ip_vs_app, rcu_head);
71 ip_vs_app_inc_destroy(inc);
75 * Allocate/initialize app incarnation and register it in proto apps.
78 ip_vs_app_inc_new(struct netns_ipvs *ipvs, struct ip_vs_app *app, __u16 proto,
81 struct ip_vs_protocol *pp;
82 struct ip_vs_app *inc;
85 if (!(pp = ip_vs_proto_get(proto)))
86 return -EPROTONOSUPPORT;
88 if (!pp->unregister_app)
91 inc = kmemdup(app, sizeof(*inc), GFP_KERNEL);
94 INIT_LIST_HEAD(&inc->p_list);
95 INIT_LIST_HEAD(&inc->incs_list);
97 inc->port = htons(port);
98 atomic_set(&inc->usecnt, 0);
102 ip_vs_create_timeout_table(app->timeouts,
104 if (!inc->timeout_table) {
110 ret = pp->register_app(ipvs, inc);
114 list_add(&inc->a_list, &app->incs_list);
115 IP_VS_DBG(9, "%s App %s:%u registered\n",
116 pp->name, inc->name, ntohs(inc->port));
121 ip_vs_app_inc_destroy(inc);
127 * Release app incarnation
130 ip_vs_app_inc_release(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
132 struct ip_vs_protocol *pp;
134 if (!(pp = ip_vs_proto_get(inc->protocol)))
137 if (pp->unregister_app)
138 pp->unregister_app(ipvs, inc);
140 IP_VS_DBG(9, "%s App %s:%u unregistered\n",
141 pp->name, inc->name, ntohs(inc->port));
143 list_del(&inc->a_list);
145 call_rcu(&inc->rcu_head, ip_vs_app_inc_rcu_free);
150 * Get reference to app inc (only called from softirq)
153 int ip_vs_app_inc_get(struct ip_vs_app *inc)
157 result = ip_vs_app_get(inc->app);
159 atomic_inc(&inc->usecnt);
165 * Put the app inc (only called from timer or net softirq)
167 void ip_vs_app_inc_put(struct ip_vs_app *inc)
169 atomic_dec(&inc->usecnt);
170 ip_vs_app_put(inc->app);
175 * Register an application incarnation in protocol applications
178 register_ip_vs_app_inc(struct netns_ipvs *ipvs, struct ip_vs_app *app, __u16 proto,
183 mutex_lock(&__ip_vs_app_mutex);
185 result = ip_vs_app_inc_new(ipvs, app, proto, port);
187 mutex_unlock(&__ip_vs_app_mutex);
193 /* Register application for netns */
194 struct ip_vs_app *register_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app)
199 mutex_lock(&__ip_vs_app_mutex);
201 list_for_each_entry(a, &ipvs->app_list, a_list) {
202 if (!strcmp(app->name, a->name)) {
207 a = kmemdup(app, sizeof(*app), GFP_KERNEL);
212 INIT_LIST_HEAD(&a->incs_list);
213 list_add(&a->a_list, &ipvs->app_list);
214 /* increase the module use count */
215 ip_vs_use_count_inc();
218 mutex_unlock(&__ip_vs_app_mutex);
220 return err ? ERR_PTR(err) : a;
225 * ip_vs_app unregistration routine
226 * We are sure there are no app incarnations attached to services
227 * Caller should use synchronize_rcu() or rcu_barrier()
229 void unregister_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app)
231 struct ip_vs_app *a, *anxt, *inc, *nxt;
233 mutex_lock(&__ip_vs_app_mutex);
235 list_for_each_entry_safe(a, anxt, &ipvs->app_list, a_list) {
236 if (app && strcmp(app->name, a->name))
238 list_for_each_entry_safe(inc, nxt, &a->incs_list, a_list) {
239 ip_vs_app_inc_release(ipvs, inc);
242 list_del(&a->a_list);
245 /* decrease the module use count */
246 ip_vs_use_count_dec();
249 mutex_unlock(&__ip_vs_app_mutex);
254 * Bind ip_vs_conn to its ip_vs_app (called by cp constructor)
256 int ip_vs_bind_app(struct ip_vs_conn *cp,
257 struct ip_vs_protocol *pp)
259 return pp->app_conn_bind(cp);
264 * Unbind cp from application incarnation (called by cp destructor)
266 void ip_vs_unbind_app(struct ip_vs_conn *cp)
268 struct ip_vs_app *inc = cp->app;
273 if (inc->unbind_conn)
274 inc->unbind_conn(inc, cp);
276 inc->done_conn(inc, cp);
277 ip_vs_app_inc_put(inc);
283 * Fixes th->seq based on ip_vs_seq info.
285 static inline void vs_fix_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
287 __u32 seq = ntohl(th->seq);
290 * Adjust seq with delta-offset for all packets after
291 * the most recent resized pkt seq and with previous_delta offset
292 * for all packets before most recent resized pkt seq.
294 if (vseq->delta || vseq->previous_delta) {
295 if(after(seq, vseq->init_seq)) {
296 th->seq = htonl(seq + vseq->delta);
297 IP_VS_DBG(9, "%s(): added delta (%d) to seq\n",
298 __func__, vseq->delta);
300 th->seq = htonl(seq + vseq->previous_delta);
301 IP_VS_DBG(9, "%s(): added previous_delta (%d) to seq\n",
302 __func__, vseq->previous_delta);
309 * Fixes th->ack_seq based on ip_vs_seq info.
312 vs_fix_ack_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
314 __u32 ack_seq = ntohl(th->ack_seq);
317 * Adjust ack_seq with delta-offset for
318 * the packets AFTER most recent resized pkt has caused a shift
319 * for packets before most recent resized pkt, use previous_delta
321 if (vseq->delta || vseq->previous_delta) {
322 /* since ack_seq is the number of octet that is expected
323 to receive next, so compare it with init_seq+delta */
324 if(after(ack_seq, vseq->init_seq+vseq->delta)) {
325 th->ack_seq = htonl(ack_seq - vseq->delta);
326 IP_VS_DBG(9, "%s(): subtracted delta "
327 "(%d) from ack_seq\n", __func__, vseq->delta);
330 th->ack_seq = htonl(ack_seq - vseq->previous_delta);
331 IP_VS_DBG(9, "%s(): subtracted "
332 "previous_delta (%d) from ack_seq\n",
333 __func__, vseq->previous_delta);
340 * Updates ip_vs_seq if pkt has been resized
341 * Assumes already checked proto==IPPROTO_TCP and diff!=0.
343 static inline void vs_seq_update(struct ip_vs_conn *cp, struct ip_vs_seq *vseq,
344 unsigned int flag, __u32 seq, int diff)
346 /* spinlock is to keep updating cp->flags atomic */
347 spin_lock_bh(&cp->lock);
348 if (!(cp->flags & flag) || after(seq, vseq->init_seq)) {
349 vseq->previous_delta = vseq->delta;
351 vseq->init_seq = seq;
354 spin_unlock_bh(&cp->lock);
357 static inline int app_tcp_pkt_out(struct ip_vs_conn *cp, struct sk_buff *skb,
358 struct ip_vs_app *app)
361 const unsigned int tcp_offset = ip_hdrlen(skb);
365 if (!skb_make_writable(skb, tcp_offset + sizeof(*th)))
368 th = (struct tcphdr *)(skb_network_header(skb) + tcp_offset);
371 * Remember seq number in case this pkt gets resized
373 seq = ntohl(th->seq);
376 * Fix seq stuff if flagged as so.
378 if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
379 vs_fix_seq(&cp->out_seq, th);
380 if (cp->flags & IP_VS_CONN_F_IN_SEQ)
381 vs_fix_ack_seq(&cp->in_seq, th);
384 * Call private output hook function
386 if (app->pkt_out == NULL)
389 if (!app->pkt_out(app, cp, skb, &diff))
393 * Update ip_vs seq stuff if len has changed.
396 vs_seq_update(cp, &cp->out_seq,
397 IP_VS_CONN_F_OUT_SEQ, seq, diff);
403 * Output pkt hook. Will call bound ip_vs_app specific function
404 * called by ipvs packet handler, assumes previously checked cp!=NULL
405 * returns false if it can't handle packet (oom)
407 int ip_vs_app_pkt_out(struct ip_vs_conn *cp, struct sk_buff *skb)
409 struct ip_vs_app *app;
412 * check if application module is bound to
415 if ((app = cp->app) == NULL)
418 /* TCP is complicated */
419 if (cp->protocol == IPPROTO_TCP)
420 return app_tcp_pkt_out(cp, skb, app);
423 * Call private output hook function
425 if (app->pkt_out == NULL)
428 return app->pkt_out(app, cp, skb, NULL);
432 static inline int app_tcp_pkt_in(struct ip_vs_conn *cp, struct sk_buff *skb,
433 struct ip_vs_app *app)
436 const unsigned int tcp_offset = ip_hdrlen(skb);
440 if (!skb_make_writable(skb, tcp_offset + sizeof(*th)))
443 th = (struct tcphdr *)(skb_network_header(skb) + tcp_offset);
446 * Remember seq number in case this pkt gets resized
448 seq = ntohl(th->seq);
451 * Fix seq stuff if flagged as so.
453 if (cp->flags & IP_VS_CONN_F_IN_SEQ)
454 vs_fix_seq(&cp->in_seq, th);
455 if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
456 vs_fix_ack_seq(&cp->out_seq, th);
459 * Call private input hook function
461 if (app->pkt_in == NULL)
464 if (!app->pkt_in(app, cp, skb, &diff))
468 * Update ip_vs seq stuff if len has changed.
471 vs_seq_update(cp, &cp->in_seq,
472 IP_VS_CONN_F_IN_SEQ, seq, diff);
478 * Input pkt hook. Will call bound ip_vs_app specific function
479 * called by ipvs packet handler, assumes previously checked cp!=NULL.
480 * returns false if can't handle packet (oom).
482 int ip_vs_app_pkt_in(struct ip_vs_conn *cp, struct sk_buff *skb)
484 struct ip_vs_app *app;
487 * check if application module is bound to
490 if ((app = cp->app) == NULL)
493 /* TCP is complicated */
494 if (cp->protocol == IPPROTO_TCP)
495 return app_tcp_pkt_in(cp, skb, app);
498 * Call private input hook function
500 if (app->pkt_in == NULL)
503 return app->pkt_in(app, cp, skb, NULL);
507 #ifdef CONFIG_PROC_FS
509 * /proc/net/ip_vs_app entry function
512 static struct ip_vs_app *ip_vs_app_idx(struct netns_ipvs *ipvs, loff_t pos)
514 struct ip_vs_app *app, *inc;
516 list_for_each_entry(app, &ipvs->app_list, a_list) {
517 list_for_each_entry(inc, &app->incs_list, a_list) {
526 static void *ip_vs_app_seq_start(struct seq_file *seq, loff_t *pos)
528 struct net *net = seq_file_net(seq);
529 struct netns_ipvs *ipvs = net_ipvs(net);
531 mutex_lock(&__ip_vs_app_mutex);
533 return *pos ? ip_vs_app_idx(ipvs, *pos - 1) : SEQ_START_TOKEN;
536 static void *ip_vs_app_seq_next(struct seq_file *seq, void *v, loff_t *pos)
538 struct ip_vs_app *inc, *app;
540 struct net *net = seq_file_net(seq);
541 struct netns_ipvs *ipvs = net_ipvs(net);
544 if (v == SEQ_START_TOKEN)
545 return ip_vs_app_idx(ipvs, 0);
550 if ((e = inc->a_list.next) != &app->incs_list)
551 return list_entry(e, struct ip_vs_app, a_list);
553 /* go on to next application */
554 for (e = app->a_list.next; e != &ipvs->app_list; e = e->next) {
555 app = list_entry(e, struct ip_vs_app, a_list);
556 list_for_each_entry(inc, &app->incs_list, a_list) {
563 static void ip_vs_app_seq_stop(struct seq_file *seq, void *v)
565 mutex_unlock(&__ip_vs_app_mutex);
568 static int ip_vs_app_seq_show(struct seq_file *seq, void *v)
570 if (v == SEQ_START_TOKEN)
571 seq_puts(seq, "prot port usecnt name\n");
573 const struct ip_vs_app *inc = v;
575 seq_printf(seq, "%-3s %-7u %-6d %-17s\n",
576 ip_vs_proto_name(inc->protocol),
578 atomic_read(&inc->usecnt),
584 static const struct seq_operations ip_vs_app_seq_ops = {
585 .start = ip_vs_app_seq_start,
586 .next = ip_vs_app_seq_next,
587 .stop = ip_vs_app_seq_stop,
588 .show = ip_vs_app_seq_show,
591 static int ip_vs_app_open(struct inode *inode, struct file *file)
593 return seq_open_net(inode, file, &ip_vs_app_seq_ops,
594 sizeof(struct seq_net_private));
597 static const struct file_operations ip_vs_app_fops = {
598 .owner = THIS_MODULE,
599 .open = ip_vs_app_open,
602 .release = seq_release_net,
606 int __net_init ip_vs_app_net_init(struct netns_ipvs *ipvs)
608 struct net *net = ipvs->net;
610 INIT_LIST_HEAD(&ipvs->app_list);
611 proc_create("ip_vs_app", 0, net->proc_net, &ip_vs_app_fops);
615 void __net_exit ip_vs_app_net_cleanup(struct netns_ipvs *ipvs)
617 struct net *net = ipvs->net;
619 unregister_ip_vs_app(ipvs, NULL /* all */);
620 remove_proc_entry("ip_vs_app", net->proc_net);