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/protocol.h>
30 #include <asm/system.h>
31 #include <linux/stat.h>
32 #include <linux/proc_fs.h>
33 #include <linux/seq_file.h>
35 #include <net/ip_vs.h>
37 EXPORT_SYMBOL(register_ip_vs_app);
38 EXPORT_SYMBOL(unregister_ip_vs_app);
39 EXPORT_SYMBOL(register_ip_vs_app_inc);
41 /* ipvs application list head */
42 static LIST_HEAD(ip_vs_app_list);
43 static DECLARE_MUTEX(__ip_vs_app_mutex);
47 * Get an ip_vs_app object
49 static inline int ip_vs_app_get(struct ip_vs_app *app)
51 /* test and get the module atomically */
53 return try_module_get(app->module);
59 static inline void ip_vs_app_put(struct ip_vs_app *app)
62 module_put(app->module);
67 * Allocate/initialize app incarnation and register it in proto apps.
70 ip_vs_app_inc_new(struct ip_vs_app *app, __u16 proto, __u16 port)
72 struct ip_vs_protocol *pp;
73 struct ip_vs_app *inc;
76 if (!(pp = ip_vs_proto_get(proto)))
77 return -EPROTONOSUPPORT;
79 if (!pp->unregister_app)
82 inc = kmalloc(sizeof(struct ip_vs_app), GFP_KERNEL);
85 memcpy(inc, app, sizeof(*inc));
86 INIT_LIST_HEAD(&inc->p_list);
87 INIT_LIST_HEAD(&inc->incs_list);
89 inc->port = htons(port);
90 atomic_set(&inc->usecnt, 0);
94 ip_vs_create_timeout_table(app->timeouts,
96 if (!inc->timeout_table) {
102 ret = pp->register_app(inc);
106 list_add(&inc->a_list, &app->incs_list);
107 IP_VS_DBG(9, "%s application %s:%u registered\n",
108 pp->name, inc->name, inc->port);
113 if (inc->timeout_table)
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 if (inc->timeout_table != NULL)
140 kfree(inc->timeout_table);
146 * Get reference to app inc (only called from softirq)
149 int ip_vs_app_inc_get(struct ip_vs_app *inc)
153 atomic_inc(&inc->usecnt);
154 if (unlikely((result = ip_vs_app_get(inc->app)) != 1))
155 atomic_dec(&inc->usecnt);
161 * Put the app inc (only called from timer or net softirq)
163 void ip_vs_app_inc_put(struct ip_vs_app *inc)
165 ip_vs_app_put(inc->app);
166 atomic_dec(&inc->usecnt);
171 * Register an application incarnation in protocol applications
174 register_ip_vs_app_inc(struct ip_vs_app *app, __u16 proto, __u16 port)
178 down(&__ip_vs_app_mutex);
180 result = ip_vs_app_inc_new(app, proto, port);
182 up(&__ip_vs_app_mutex);
189 * ip_vs_app registration routine
191 int register_ip_vs_app(struct ip_vs_app *app)
193 /* increase the module use count */
194 ip_vs_use_count_inc();
196 down(&__ip_vs_app_mutex);
198 list_add(&app->a_list, &ip_vs_app_list);
200 up(&__ip_vs_app_mutex);
207 * ip_vs_app unregistration routine
208 * We are sure there are no app incarnations attached to services
210 void unregister_ip_vs_app(struct ip_vs_app *app)
212 struct ip_vs_app *inc, *nxt;
214 down(&__ip_vs_app_mutex);
216 list_for_each_entry_safe(inc, nxt, &app->incs_list, a_list) {
217 ip_vs_app_inc_release(inc);
220 list_del(&app->a_list);
222 up(&__ip_vs_app_mutex);
224 /* decrease the module use count */
225 ip_vs_use_count_dec();
231 * Get reference to app by name (called from user context)
233 struct ip_vs_app *ip_vs_app_get_by_name(char *appname)
235 struct ip_vs_app *app, *a = NULL;
237 down(&__ip_vs_app_mutex);
239 list_for_each_entry(ent, &ip_vs_app_list, a_list) {
240 if (strcmp(app->name, appname))
243 /* softirq may call ip_vs_app_get too, so the caller
244 must disable softirq on the current CPU */
245 if (ip_vs_app_get(app))
250 up(&__ip_vs_app_mutex);
258 * Bind ip_vs_conn to its ip_vs_app (called by cp constructor)
260 int ip_vs_bind_app(struct ip_vs_conn *cp, struct ip_vs_protocol *pp)
262 return pp->app_conn_bind(cp);
267 * Unbind cp from application incarnation (called by cp destructor)
269 void ip_vs_unbind_app(struct ip_vs_conn *cp)
271 struct ip_vs_app *inc = cp->app;
276 if (inc->unbind_conn)
277 inc->unbind_conn(inc, cp);
279 inc->done_conn(inc, cp);
280 ip_vs_app_inc_put(inc);
286 * Fixes th->seq based on ip_vs_seq info.
288 static inline void vs_fix_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
290 __u32 seq = ntohl(th->seq);
293 * Adjust seq with delta-offset for all packets after
294 * the most recent resized pkt seq and with previous_delta offset
295 * for all packets before most recent resized pkt seq.
297 if (vseq->delta || vseq->previous_delta) {
298 if(after(seq, vseq->init_seq)) {
299 th->seq = htonl(seq + vseq->delta);
300 IP_VS_DBG(9, "vs_fix_seq(): added delta (%d) to seq\n",
303 th->seq = htonl(seq + vseq->previous_delta);
304 IP_VS_DBG(9, "vs_fix_seq(): added previous_delta "
305 "(%d) to seq\n", vseq->previous_delta);
312 * Fixes th->ack_seq based on ip_vs_seq info.
315 vs_fix_ack_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
317 __u32 ack_seq = ntohl(th->ack_seq);
320 * Adjust ack_seq with delta-offset for
321 * the packets AFTER most recent resized pkt has caused a shift
322 * for packets before most recent resized pkt, use previous_delta
324 if (vseq->delta || vseq->previous_delta) {
325 /* since ack_seq is the number of octet that is expected
326 to receive next, so compare it with init_seq+delta */
327 if(after(ack_seq, vseq->init_seq+vseq->delta)) {
328 th->ack_seq = htonl(ack_seq - vseq->delta);
329 IP_VS_DBG(9, "vs_fix_ack_seq(): subtracted delta "
330 "(%d) from ack_seq\n", vseq->delta);
333 th->ack_seq = htonl(ack_seq - vseq->previous_delta);
334 IP_VS_DBG(9, "vs_fix_ack_seq(): subtracted "
335 "previous_delta (%d) from ack_seq\n",
336 vseq->previous_delta);
343 * Updates ip_vs_seq if pkt has been resized
344 * Assumes already checked proto==IPPROTO_TCP and diff!=0.
346 static inline void vs_seq_update(struct ip_vs_conn *cp, struct ip_vs_seq *vseq,
347 unsigned flag, __u32 seq, int diff)
349 /* spinlock is to keep updating cp->flags atomic */
350 spin_lock(&cp->lock);
351 if (!(cp->flags & flag) || after(seq, vseq->init_seq)) {
352 vseq->previous_delta = vseq->delta;
354 vseq->init_seq = seq;
357 spin_unlock(&cp->lock);
360 static inline int app_tcp_pkt_out(struct ip_vs_conn *cp, struct sk_buff **pskb,
361 struct ip_vs_app *app)
364 unsigned int tcp_offset = (*pskb)->nh.iph->ihl*4;
368 if (!ip_vs_make_skb_writable(pskb, tcp_offset + sizeof(*th)))
371 th = (struct tcphdr *)((*pskb)->nh.raw + tcp_offset);
374 * Remember seq number in case this pkt gets resized
376 seq = ntohl(th->seq);
379 * Fix seq stuff if flagged as so.
381 if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
382 vs_fix_seq(&cp->out_seq, th);
383 if (cp->flags & IP_VS_CONN_F_IN_SEQ)
384 vs_fix_ack_seq(&cp->in_seq, th);
387 * Call private output hook function
389 if (app->pkt_out == NULL)
392 if (!app->pkt_out(app, cp, pskb, &diff))
396 * Update ip_vs seq stuff if len has changed.
399 vs_seq_update(cp, &cp->out_seq,
400 IP_VS_CONN_F_OUT_SEQ, seq, diff);
406 * Output pkt hook. Will call bound ip_vs_app specific function
407 * called by ipvs packet handler, assumes previously checked cp!=NULL
408 * returns false if it can't handle packet (oom)
410 int ip_vs_app_pkt_out(struct ip_vs_conn *cp, struct sk_buff **pskb)
412 struct ip_vs_app *app;
415 * check if application module is bound to
418 if ((app = cp->app) == NULL)
421 /* TCP is complicated */
422 if (cp->protocol == IPPROTO_TCP)
423 return app_tcp_pkt_out(cp, pskb, app);
426 * Call private output hook function
428 if (app->pkt_out == NULL)
431 return app->pkt_out(app, cp, pskb, NULL);
435 static inline int app_tcp_pkt_in(struct ip_vs_conn *cp, struct sk_buff **pskb,
436 struct ip_vs_app *app)
439 unsigned int tcp_offset = (*pskb)->nh.iph->ihl*4;
443 if (!ip_vs_make_skb_writable(pskb, tcp_offset + sizeof(*th)))
446 th = (struct tcphdr *)((*pskb)->nh.raw + tcp_offset);
449 * Remember seq number in case this pkt gets resized
451 seq = ntohl(th->seq);
454 * Fix seq stuff if flagged as so.
456 if (cp->flags & IP_VS_CONN_F_IN_SEQ)
457 vs_fix_seq(&cp->in_seq, th);
458 if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
459 vs_fix_ack_seq(&cp->out_seq, th);
462 * Call private input hook function
464 if (app->pkt_in == NULL)
467 if (!app->pkt_in(app, cp, pskb, &diff))
471 * Update ip_vs seq stuff if len has changed.
474 vs_seq_update(cp, &cp->in_seq,
475 IP_VS_CONN_F_IN_SEQ, seq, diff);
481 * Input pkt hook. Will call bound ip_vs_app specific function
482 * called by ipvs packet handler, assumes previously checked cp!=NULL.
483 * returns false if can't handle packet (oom).
485 int ip_vs_app_pkt_in(struct ip_vs_conn *cp, struct sk_buff **pskb)
487 struct ip_vs_app *app;
490 * check if application module is bound to
493 if ((app = cp->app) == NULL)
496 /* TCP is complicated */
497 if (cp->protocol == IPPROTO_TCP)
498 return app_tcp_pkt_in(cp, pskb, app);
501 * Call private input hook function
503 if (app->pkt_in == NULL)
506 return app->pkt_in(app, cp, pskb, NULL);
510 #ifdef CONFIG_PROC_FS
512 * /proc/net/ip_vs_app entry function
515 static struct ip_vs_app *ip_vs_app_idx(loff_t pos)
517 struct ip_vs_app *app, *inc;
519 list_for_each_entry(app, &ip_vs_app_list, a_list) {
520 list_for_each_entry(inc, &app->incs_list, a_list) {
529 static void *ip_vs_app_seq_start(struct seq_file *seq, loff_t *pos)
531 down(&__ip_vs_app_mutex);
533 return *pos ? ip_vs_app_idx(*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;
542 if (v == SEQ_START_TOKEN)
543 return ip_vs_app_idx(0);
548 if ((e = inc->a_list.next) != &app->incs_list)
549 return list_entry(e, struct ip_vs_app, a_list);
551 /* go on to next application */
552 for (e = app->a_list.next; e != &ip_vs_app_list; e = e->next) {
553 app = list_entry(e, struct ip_vs_app, a_list);
554 list_for_each_entry(inc, &app->incs_list, a_list) {
561 static void ip_vs_app_seq_stop(struct seq_file *seq, void *v)
563 up(&__ip_vs_app_mutex);
566 static int ip_vs_app_seq_show(struct seq_file *seq, void *v)
568 if (v == SEQ_START_TOKEN)
569 seq_puts(seq, "prot port usecnt name\n");
571 const struct ip_vs_app *inc = v;
573 seq_printf(seq, "%-3s %-7u %-6d %-17s\n",
574 ip_vs_proto_name(inc->protocol),
576 atomic_read(&inc->usecnt),
582 static struct seq_operations ip_vs_app_seq_ops = {
583 .start = ip_vs_app_seq_start,
584 .next = ip_vs_app_seq_next,
585 .stop = ip_vs_app_seq_stop,
586 .show = ip_vs_app_seq_show,
589 static int ip_vs_app_open(struct inode *inode, struct file *file)
591 return seq_open(file, &ip_vs_app_seq_ops);
594 static struct file_operations ip_vs_app_fops = {
595 .owner = THIS_MODULE,
596 .open = ip_vs_app_open,
599 .release = seq_release,
605 * Replace a segment of data with a new segment
607 int ip_vs_skb_replace(struct sk_buff *skb, int pri,
608 char *o_buf, int o_len, char *n_buf, int n_len)
617 diff = n_len - o_len;
618 o_offset = o_buf - (char *)skb->data;
619 /* The length of left data after o_buf+o_len in the skb data */
620 o_left = skb->len - (o_offset + o_len);
623 memmove(o_buf + n_len, o_buf + o_len, o_left);
624 memcpy(o_buf, n_buf, n_len);
625 skb_trim(skb, skb->len + diff);
626 } else if (diff <= skb_tailroom(skb)) {
628 memmove(o_buf + n_len, o_buf + o_len, o_left);
629 memcpy(o_buf, n_buf, n_len);
631 if (pskb_expand_head(skb, skb_headroom(skb), diff, pri))
634 memmove(skb->data + o_offset + n_len,
635 skb->data + o_offset + o_len, o_left);
636 memcpy(skb->data + o_offset, n_buf, n_len);
639 /* must update the iph total length here */
641 iph->tot_len = htons(skb->len);
648 int ip_vs_app_init(void)
650 /* we will replace it with proc_net_ipvs_create() soon */
651 proc_net_fops_create("ip_vs_app", 0, &ip_vs_app_fops);
656 void ip_vs_app_cleanup(void)
658 proc_net_remove("ip_vs_app");