2 * ip_vs_app.c: Application module support for IPVS
4 * Version: $Id: ip_vs_app.c,v 1.17 2003/03/22 06:31:21 wensong Exp $
6 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Most code here is taken from ip_masq_app.c in kernel 2.2. The difference
14 * is that ip_vs_app module handles the reverse direction (incoming requests
15 * and outgoing responses).
17 * IP_MASQ_APP application masquerading module
19 * Author: Juan Jose Ciarlante, <jjciarla@raiz.uncu.edu.ar>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/skbuff.h>
28 #include <net/net_namespace.h>
29 #include <net/protocol.h>
31 #include <asm/system.h>
32 #include <linux/stat.h>
33 #include <linux/proc_fs.h>
34 #include <linux/seq_file.h>
35 #include <linux/mutex.h>
37 #include <net/ip_vs.h>
39 EXPORT_SYMBOL(register_ip_vs_app);
40 EXPORT_SYMBOL(unregister_ip_vs_app);
41 EXPORT_SYMBOL(register_ip_vs_app_inc);
43 /* ipvs application list head */
44 static LIST_HEAD(ip_vs_app_list);
45 static DEFINE_MUTEX(__ip_vs_app_mutex);
49 * Get an ip_vs_app object
51 static inline int ip_vs_app_get(struct ip_vs_app *app)
53 /* test and get the module atomically */
55 return try_module_get(app->module);
61 static inline void ip_vs_app_put(struct ip_vs_app *app)
64 module_put(app->module);
69 * Allocate/initialize app incarnation and register it in proto apps.
72 ip_vs_app_inc_new(struct ip_vs_app *app, __u16 proto, __u16 port)
74 struct ip_vs_protocol *pp;
75 struct ip_vs_app *inc;
78 if (!(pp = ip_vs_proto_get(proto)))
79 return -EPROTONOSUPPORT;
81 if (!pp->unregister_app)
84 inc = kmemdup(app, sizeof(*inc), GFP_KERNEL);
87 INIT_LIST_HEAD(&inc->p_list);
88 INIT_LIST_HEAD(&inc->incs_list);
90 inc->port = htons(port);
91 atomic_set(&inc->usecnt, 0);
95 ip_vs_create_timeout_table(app->timeouts,
97 if (!inc->timeout_table) {
103 ret = pp->register_app(inc);
107 list_add(&inc->a_list, &app->incs_list);
108 IP_VS_DBG(9, "%s application %s:%u registered\n",
109 pp->name, inc->name, inc->port);
114 kfree(inc->timeout_table);
121 * Release app incarnation
124 ip_vs_app_inc_release(struct ip_vs_app *inc)
126 struct ip_vs_protocol *pp;
128 if (!(pp = ip_vs_proto_get(inc->protocol)))
131 if (pp->unregister_app)
132 pp->unregister_app(inc);
134 IP_VS_DBG(9, "%s App %s:%u unregistered\n",
135 pp->name, inc->name, inc->port);
137 list_del(&inc->a_list);
139 kfree(inc->timeout_table);
145 * Get reference to app inc (only called from softirq)
148 int ip_vs_app_inc_get(struct ip_vs_app *inc)
152 atomic_inc(&inc->usecnt);
153 if (unlikely((result = ip_vs_app_get(inc->app)) != 1))
154 atomic_dec(&inc->usecnt);
160 * Put the app inc (only called from timer or net softirq)
162 void ip_vs_app_inc_put(struct ip_vs_app *inc)
164 ip_vs_app_put(inc->app);
165 atomic_dec(&inc->usecnt);
170 * Register an application incarnation in protocol applications
173 register_ip_vs_app_inc(struct ip_vs_app *app, __u16 proto, __u16 port)
177 mutex_lock(&__ip_vs_app_mutex);
179 result = ip_vs_app_inc_new(app, proto, port);
181 mutex_unlock(&__ip_vs_app_mutex);
188 * ip_vs_app registration routine
190 int register_ip_vs_app(struct ip_vs_app *app)
192 /* increase the module use count */
193 ip_vs_use_count_inc();
195 mutex_lock(&__ip_vs_app_mutex);
197 list_add(&app->a_list, &ip_vs_app_list);
199 mutex_unlock(&__ip_vs_app_mutex);
206 * ip_vs_app unregistration routine
207 * We are sure there are no app incarnations attached to services
209 void unregister_ip_vs_app(struct ip_vs_app *app)
211 struct ip_vs_app *inc, *nxt;
213 mutex_lock(&__ip_vs_app_mutex);
215 list_for_each_entry_safe(inc, nxt, &app->incs_list, a_list) {
216 ip_vs_app_inc_release(inc);
219 list_del(&app->a_list);
221 mutex_unlock(&__ip_vs_app_mutex);
223 /* decrease the module use count */
224 ip_vs_use_count_dec();
229 * Bind ip_vs_conn to its ip_vs_app (called by cp constructor)
231 int ip_vs_bind_app(struct ip_vs_conn *cp, struct ip_vs_protocol *pp)
233 return pp->app_conn_bind(cp);
238 * Unbind cp from application incarnation (called by cp destructor)
240 void ip_vs_unbind_app(struct ip_vs_conn *cp)
242 struct ip_vs_app *inc = cp->app;
247 if (inc->unbind_conn)
248 inc->unbind_conn(inc, cp);
250 inc->done_conn(inc, cp);
251 ip_vs_app_inc_put(inc);
257 * Fixes th->seq based on ip_vs_seq info.
259 static inline void vs_fix_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
261 __u32 seq = ntohl(th->seq);
264 * Adjust seq with delta-offset for all packets after
265 * the most recent resized pkt seq and with previous_delta offset
266 * for all packets before most recent resized pkt seq.
268 if (vseq->delta || vseq->previous_delta) {
269 if(after(seq, vseq->init_seq)) {
270 th->seq = htonl(seq + vseq->delta);
271 IP_VS_DBG(9, "vs_fix_seq(): added delta (%d) to seq\n",
274 th->seq = htonl(seq + vseq->previous_delta);
275 IP_VS_DBG(9, "vs_fix_seq(): added previous_delta "
276 "(%d) to seq\n", vseq->previous_delta);
283 * Fixes th->ack_seq based on ip_vs_seq info.
286 vs_fix_ack_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
288 __u32 ack_seq = ntohl(th->ack_seq);
291 * Adjust ack_seq with delta-offset for
292 * the packets AFTER most recent resized pkt has caused a shift
293 * for packets before most recent resized pkt, use previous_delta
295 if (vseq->delta || vseq->previous_delta) {
296 /* since ack_seq is the number of octet that is expected
297 to receive next, so compare it with init_seq+delta */
298 if(after(ack_seq, vseq->init_seq+vseq->delta)) {
299 th->ack_seq = htonl(ack_seq - vseq->delta);
300 IP_VS_DBG(9, "vs_fix_ack_seq(): subtracted delta "
301 "(%d) from ack_seq\n", vseq->delta);
304 th->ack_seq = htonl(ack_seq - vseq->previous_delta);
305 IP_VS_DBG(9, "vs_fix_ack_seq(): subtracted "
306 "previous_delta (%d) from ack_seq\n",
307 vseq->previous_delta);
314 * Updates ip_vs_seq if pkt has been resized
315 * Assumes already checked proto==IPPROTO_TCP and diff!=0.
317 static inline void vs_seq_update(struct ip_vs_conn *cp, struct ip_vs_seq *vseq,
318 unsigned flag, __u32 seq, int diff)
320 /* spinlock is to keep updating cp->flags atomic */
321 spin_lock(&cp->lock);
322 if (!(cp->flags & flag) || after(seq, vseq->init_seq)) {
323 vseq->previous_delta = vseq->delta;
325 vseq->init_seq = seq;
328 spin_unlock(&cp->lock);
331 static inline int app_tcp_pkt_out(struct ip_vs_conn *cp, struct sk_buff **pskb,
332 struct ip_vs_app *app)
335 const unsigned int tcp_offset = ip_hdrlen(*pskb);
339 if (!ip_vs_make_skb_writable(pskb, tcp_offset + sizeof(*th)))
342 th = (struct tcphdr *)(skb_network_header(*pskb) + tcp_offset);
345 * Remember seq number in case this pkt gets resized
347 seq = ntohl(th->seq);
350 * Fix seq stuff if flagged as so.
352 if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
353 vs_fix_seq(&cp->out_seq, th);
354 if (cp->flags & IP_VS_CONN_F_IN_SEQ)
355 vs_fix_ack_seq(&cp->in_seq, th);
358 * Call private output hook function
360 if (app->pkt_out == NULL)
363 if (!app->pkt_out(app, cp, pskb, &diff))
367 * Update ip_vs seq stuff if len has changed.
370 vs_seq_update(cp, &cp->out_seq,
371 IP_VS_CONN_F_OUT_SEQ, seq, diff);
377 * Output pkt hook. Will call bound ip_vs_app specific function
378 * called by ipvs packet handler, assumes previously checked cp!=NULL
379 * returns false if it can't handle packet (oom)
381 int ip_vs_app_pkt_out(struct ip_vs_conn *cp, struct sk_buff **pskb)
383 struct ip_vs_app *app;
386 * check if application module is bound to
389 if ((app = cp->app) == NULL)
392 /* TCP is complicated */
393 if (cp->protocol == IPPROTO_TCP)
394 return app_tcp_pkt_out(cp, pskb, app);
397 * Call private output hook function
399 if (app->pkt_out == NULL)
402 return app->pkt_out(app, cp, pskb, NULL);
406 static inline int app_tcp_pkt_in(struct ip_vs_conn *cp, struct sk_buff **pskb,
407 struct ip_vs_app *app)
410 const unsigned int tcp_offset = ip_hdrlen(*pskb);
414 if (!ip_vs_make_skb_writable(pskb, tcp_offset + sizeof(*th)))
417 th = (struct tcphdr *)(skb_network_header(*pskb) + tcp_offset);
420 * Remember seq number in case this pkt gets resized
422 seq = ntohl(th->seq);
425 * Fix seq stuff if flagged as so.
427 if (cp->flags & IP_VS_CONN_F_IN_SEQ)
428 vs_fix_seq(&cp->in_seq, th);
429 if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
430 vs_fix_ack_seq(&cp->out_seq, th);
433 * Call private input hook function
435 if (app->pkt_in == NULL)
438 if (!app->pkt_in(app, cp, pskb, &diff))
442 * Update ip_vs seq stuff if len has changed.
445 vs_seq_update(cp, &cp->in_seq,
446 IP_VS_CONN_F_IN_SEQ, seq, diff);
452 * Input pkt hook. Will call bound ip_vs_app specific function
453 * called by ipvs packet handler, assumes previously checked cp!=NULL.
454 * returns false if can't handle packet (oom).
456 int ip_vs_app_pkt_in(struct ip_vs_conn *cp, struct sk_buff **pskb)
458 struct ip_vs_app *app;
461 * check if application module is bound to
464 if ((app = cp->app) == NULL)
467 /* TCP is complicated */
468 if (cp->protocol == IPPROTO_TCP)
469 return app_tcp_pkt_in(cp, pskb, app);
472 * Call private input hook function
474 if (app->pkt_in == NULL)
477 return app->pkt_in(app, cp, pskb, NULL);
481 #ifdef CONFIG_PROC_FS
483 * /proc/net/ip_vs_app entry function
486 static struct ip_vs_app *ip_vs_app_idx(loff_t pos)
488 struct ip_vs_app *app, *inc;
490 list_for_each_entry(app, &ip_vs_app_list, a_list) {
491 list_for_each_entry(inc, &app->incs_list, a_list) {
500 static void *ip_vs_app_seq_start(struct seq_file *seq, loff_t *pos)
502 mutex_lock(&__ip_vs_app_mutex);
504 return *pos ? ip_vs_app_idx(*pos - 1) : SEQ_START_TOKEN;
507 static void *ip_vs_app_seq_next(struct seq_file *seq, void *v, loff_t *pos)
509 struct ip_vs_app *inc, *app;
513 if (v == SEQ_START_TOKEN)
514 return ip_vs_app_idx(0);
519 if ((e = inc->a_list.next) != &app->incs_list)
520 return list_entry(e, struct ip_vs_app, a_list);
522 /* go on to next application */
523 for (e = app->a_list.next; e != &ip_vs_app_list; e = e->next) {
524 app = list_entry(e, struct ip_vs_app, a_list);
525 list_for_each_entry(inc, &app->incs_list, a_list) {
532 static void ip_vs_app_seq_stop(struct seq_file *seq, void *v)
534 mutex_unlock(&__ip_vs_app_mutex);
537 static int ip_vs_app_seq_show(struct seq_file *seq, void *v)
539 if (v == SEQ_START_TOKEN)
540 seq_puts(seq, "prot port usecnt name\n");
542 const struct ip_vs_app *inc = v;
544 seq_printf(seq, "%-3s %-7u %-6d %-17s\n",
545 ip_vs_proto_name(inc->protocol),
547 atomic_read(&inc->usecnt),
553 static const struct seq_operations ip_vs_app_seq_ops = {
554 .start = ip_vs_app_seq_start,
555 .next = ip_vs_app_seq_next,
556 .stop = ip_vs_app_seq_stop,
557 .show = ip_vs_app_seq_show,
560 static int ip_vs_app_open(struct inode *inode, struct file *file)
562 return seq_open(file, &ip_vs_app_seq_ops);
565 static const struct file_operations ip_vs_app_fops = {
566 .owner = THIS_MODULE,
567 .open = ip_vs_app_open,
570 .release = seq_release,
576 * Replace a segment of data with a new segment
578 int ip_vs_skb_replace(struct sk_buff *skb, gfp_t pri,
579 char *o_buf, int o_len, char *n_buf, int n_len)
587 diff = n_len - o_len;
588 o_offset = o_buf - (char *)skb->data;
589 /* The length of left data after o_buf+o_len in the skb data */
590 o_left = skb->len - (o_offset + o_len);
593 memmove(o_buf + n_len, o_buf + o_len, o_left);
594 memcpy(o_buf, n_buf, n_len);
595 skb_trim(skb, skb->len + diff);
596 } else if (diff <= skb_tailroom(skb)) {
598 memmove(o_buf + n_len, o_buf + o_len, o_left);
599 memcpy(o_buf, n_buf, n_len);
601 if (pskb_expand_head(skb, skb_headroom(skb), diff, pri))
604 memmove(skb->data + o_offset + n_len,
605 skb->data + o_offset + o_len, o_left);
606 skb_copy_to_linear_data_offset(skb, o_offset, n_buf, n_len);
609 /* must update the iph total length here */
610 ip_hdr(skb)->tot_len = htons(skb->len);
617 int ip_vs_app_init(void)
619 /* we will replace it with proc_net_ipvs_create() soon */
620 proc_net_fops_create(&init_net, "ip_vs_app", 0, &ip_vs_app_fops);
625 void ip_vs_app_cleanup(void)
627 proc_net_remove(&init_net, "ip_vs_app");