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
);
63 * Allocate/initialize app incarnation and register it in proto apps.
66 ip_vs_app_inc_new(struct net
*net
, struct ip_vs_app
*app
, __u16 proto
,
69 struct ip_vs_protocol
*pp
;
70 struct ip_vs_app
*inc
;
73 if (!(pp
= ip_vs_proto_get(proto
)))
74 return -EPROTONOSUPPORT
;
76 if (!pp
->unregister_app
)
79 inc
= kmemdup(app
, sizeof(*inc
), GFP_KERNEL
);
82 INIT_LIST_HEAD(&inc
->p_list
);
83 INIT_LIST_HEAD(&inc
->incs_list
);
85 inc
->port
= htons(port
);
86 atomic_set(&inc
->usecnt
, 0);
90 ip_vs_create_timeout_table(app
->timeouts
,
92 if (!inc
->timeout_table
) {
98 ret
= pp
->register_app(net
, inc
);
102 list_add(&inc
->a_list
, &app
->incs_list
);
103 IP_VS_DBG(9, "%s App %s:%u registered\n",
104 pp
->name
, inc
->name
, ntohs(inc
->port
));
109 kfree(inc
->timeout_table
);
116 * Release app incarnation
119 ip_vs_app_inc_release(struct net
*net
, struct ip_vs_app
*inc
)
121 struct ip_vs_protocol
*pp
;
123 if (!(pp
= ip_vs_proto_get(inc
->protocol
)))
126 if (pp
->unregister_app
)
127 pp
->unregister_app(net
, inc
);
129 IP_VS_DBG(9, "%s App %s:%u unregistered\n",
130 pp
->name
, inc
->name
, ntohs(inc
->port
));
132 list_del(&inc
->a_list
);
134 kfree(inc
->timeout_table
);
140 * Get reference to app inc (only called from softirq)
143 int ip_vs_app_inc_get(struct ip_vs_app
*inc
)
147 atomic_inc(&inc
->usecnt
);
148 if (unlikely((result
= ip_vs_app_get(inc
->app
)) != 1))
149 atomic_dec(&inc
->usecnt
);
155 * Put the app inc (only called from timer or net softirq)
157 void ip_vs_app_inc_put(struct ip_vs_app
*inc
)
159 ip_vs_app_put(inc
->app
);
160 atomic_dec(&inc
->usecnt
);
165 * Register an application incarnation in protocol applications
168 register_ip_vs_app_inc(struct net
*net
, struct ip_vs_app
*app
, __u16 proto
,
173 mutex_lock(&__ip_vs_app_mutex
);
175 result
= ip_vs_app_inc_new(net
, app
, proto
, port
);
177 mutex_unlock(&__ip_vs_app_mutex
);
183 /* Register application for netns */
184 struct ip_vs_app
*register_ip_vs_app(struct net
*net
, struct ip_vs_app
*app
)
186 struct netns_ipvs
*ipvs
= net_ipvs(net
);
191 return ERR_PTR(-ENOENT
);
193 mutex_lock(&__ip_vs_app_mutex
);
195 list_for_each_entry(a
, &ipvs
->app_list
, a_list
) {
196 if (!strcmp(app
->name
, a
->name
)) {
201 a
= kmemdup(app
, sizeof(*app
), GFP_KERNEL
);
206 INIT_LIST_HEAD(&a
->incs_list
);
207 list_add(&a
->a_list
, &ipvs
->app_list
);
208 /* increase the module use count */
209 ip_vs_use_count_inc();
212 mutex_unlock(&__ip_vs_app_mutex
);
214 return err
? ERR_PTR(err
) : a
;
219 * ip_vs_app unregistration routine
220 * We are sure there are no app incarnations attached to services
222 void unregister_ip_vs_app(struct net
*net
, struct ip_vs_app
*app
)
224 struct netns_ipvs
*ipvs
= net_ipvs(net
);
225 struct ip_vs_app
*a
, *anxt
, *inc
, *nxt
;
230 mutex_lock(&__ip_vs_app_mutex
);
232 list_for_each_entry_safe(a
, anxt
, &ipvs
->app_list
, a_list
) {
233 if (app
&& strcmp(app
->name
, a
->name
))
235 list_for_each_entry_safe(inc
, nxt
, &a
->incs_list
, a_list
) {
236 ip_vs_app_inc_release(net
, inc
);
239 list_del(&a
->a_list
);
242 /* decrease the module use count */
243 ip_vs_use_count_dec();
246 mutex_unlock(&__ip_vs_app_mutex
);
251 * Bind ip_vs_conn to its ip_vs_app (called by cp constructor)
253 int ip_vs_bind_app(struct ip_vs_conn
*cp
,
254 struct ip_vs_protocol
*pp
)
256 return pp
->app_conn_bind(cp
);
261 * Unbind cp from application incarnation (called by cp destructor)
263 void ip_vs_unbind_app(struct ip_vs_conn
*cp
)
265 struct ip_vs_app
*inc
= cp
->app
;
270 if (inc
->unbind_conn
)
271 inc
->unbind_conn(inc
, cp
);
273 inc
->done_conn(inc
, cp
);
274 ip_vs_app_inc_put(inc
);
280 * Fixes th->seq based on ip_vs_seq info.
282 static inline void vs_fix_seq(const struct ip_vs_seq
*vseq
, struct tcphdr
*th
)
284 __u32 seq
= ntohl(th
->seq
);
287 * Adjust seq with delta-offset for all packets after
288 * the most recent resized pkt seq and with previous_delta offset
289 * for all packets before most recent resized pkt seq.
291 if (vseq
->delta
|| vseq
->previous_delta
) {
292 if(after(seq
, vseq
->init_seq
)) {
293 th
->seq
= htonl(seq
+ vseq
->delta
);
294 IP_VS_DBG(9, "%s(): added delta (%d) to seq\n",
295 __func__
, vseq
->delta
);
297 th
->seq
= htonl(seq
+ vseq
->previous_delta
);
298 IP_VS_DBG(9, "%s(): added previous_delta (%d) to seq\n",
299 __func__
, vseq
->previous_delta
);
306 * Fixes th->ack_seq based on ip_vs_seq info.
309 vs_fix_ack_seq(const struct ip_vs_seq
*vseq
, struct tcphdr
*th
)
311 __u32 ack_seq
= ntohl(th
->ack_seq
);
314 * Adjust ack_seq with delta-offset for
315 * the packets AFTER most recent resized pkt has caused a shift
316 * for packets before most recent resized pkt, use previous_delta
318 if (vseq
->delta
|| vseq
->previous_delta
) {
319 /* since ack_seq is the number of octet that is expected
320 to receive next, so compare it with init_seq+delta */
321 if(after(ack_seq
, vseq
->init_seq
+vseq
->delta
)) {
322 th
->ack_seq
= htonl(ack_seq
- vseq
->delta
);
323 IP_VS_DBG(9, "%s(): subtracted delta "
324 "(%d) from ack_seq\n", __func__
, vseq
->delta
);
327 th
->ack_seq
= htonl(ack_seq
- vseq
->previous_delta
);
328 IP_VS_DBG(9, "%s(): subtracted "
329 "previous_delta (%d) from ack_seq\n",
330 __func__
, vseq
->previous_delta
);
337 * Updates ip_vs_seq if pkt has been resized
338 * Assumes already checked proto==IPPROTO_TCP and diff!=0.
340 static inline void vs_seq_update(struct ip_vs_conn
*cp
, struct ip_vs_seq
*vseq
,
341 unsigned int flag
, __u32 seq
, int diff
)
343 /* spinlock is to keep updating cp->flags atomic */
344 spin_lock(&cp
->lock
);
345 if (!(cp
->flags
& flag
) || after(seq
, vseq
->init_seq
)) {
346 vseq
->previous_delta
= vseq
->delta
;
348 vseq
->init_seq
= seq
;
351 spin_unlock(&cp
->lock
);
354 static inline int app_tcp_pkt_out(struct ip_vs_conn
*cp
, struct sk_buff
*skb
,
355 struct ip_vs_app
*app
)
358 const unsigned int tcp_offset
= ip_hdrlen(skb
);
362 if (!skb_make_writable(skb
, tcp_offset
+ sizeof(*th
)))
365 th
= (struct tcphdr
*)(skb_network_header(skb
) + tcp_offset
);
368 * Remember seq number in case this pkt gets resized
370 seq
= ntohl(th
->seq
);
373 * Fix seq stuff if flagged as so.
375 if (cp
->flags
& IP_VS_CONN_F_OUT_SEQ
)
376 vs_fix_seq(&cp
->out_seq
, th
);
377 if (cp
->flags
& IP_VS_CONN_F_IN_SEQ
)
378 vs_fix_ack_seq(&cp
->in_seq
, th
);
381 * Call private output hook function
383 if (app
->pkt_out
== NULL
)
386 if (!app
->pkt_out(app
, cp
, skb
, &diff
))
390 * Update ip_vs seq stuff if len has changed.
393 vs_seq_update(cp
, &cp
->out_seq
,
394 IP_VS_CONN_F_OUT_SEQ
, seq
, diff
);
400 * Output pkt hook. Will call bound ip_vs_app specific function
401 * called by ipvs packet handler, assumes previously checked cp!=NULL
402 * returns false if it can't handle packet (oom)
404 int ip_vs_app_pkt_out(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
406 struct ip_vs_app
*app
;
409 * check if application module is bound to
412 if ((app
= cp
->app
) == NULL
)
415 /* TCP is complicated */
416 if (cp
->protocol
== IPPROTO_TCP
)
417 return app_tcp_pkt_out(cp
, skb
, app
);
420 * Call private output hook function
422 if (app
->pkt_out
== NULL
)
425 return app
->pkt_out(app
, cp
, skb
, NULL
);
429 static inline int app_tcp_pkt_in(struct ip_vs_conn
*cp
, struct sk_buff
*skb
,
430 struct ip_vs_app
*app
)
433 const unsigned int tcp_offset
= ip_hdrlen(skb
);
437 if (!skb_make_writable(skb
, tcp_offset
+ sizeof(*th
)))
440 th
= (struct tcphdr
*)(skb_network_header(skb
) + tcp_offset
);
443 * Remember seq number in case this pkt gets resized
445 seq
= ntohl(th
->seq
);
448 * Fix seq stuff if flagged as so.
450 if (cp
->flags
& IP_VS_CONN_F_IN_SEQ
)
451 vs_fix_seq(&cp
->in_seq
, th
);
452 if (cp
->flags
& IP_VS_CONN_F_OUT_SEQ
)
453 vs_fix_ack_seq(&cp
->out_seq
, th
);
456 * Call private input hook function
458 if (app
->pkt_in
== NULL
)
461 if (!app
->pkt_in(app
, cp
, skb
, &diff
))
465 * Update ip_vs seq stuff if len has changed.
468 vs_seq_update(cp
, &cp
->in_seq
,
469 IP_VS_CONN_F_IN_SEQ
, seq
, diff
);
475 * Input pkt hook. Will call bound ip_vs_app specific function
476 * called by ipvs packet handler, assumes previously checked cp!=NULL.
477 * returns false if can't handle packet (oom).
479 int ip_vs_app_pkt_in(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
481 struct ip_vs_app
*app
;
484 * check if application module is bound to
487 if ((app
= cp
->app
) == NULL
)
490 /* TCP is complicated */
491 if (cp
->protocol
== IPPROTO_TCP
)
492 return app_tcp_pkt_in(cp
, skb
, app
);
495 * Call private input hook function
497 if (app
->pkt_in
== NULL
)
500 return app
->pkt_in(app
, cp
, skb
, NULL
);
504 #ifdef CONFIG_PROC_FS
506 * /proc/net/ip_vs_app entry function
509 static struct ip_vs_app
*ip_vs_app_idx(struct netns_ipvs
*ipvs
, loff_t pos
)
511 struct ip_vs_app
*app
, *inc
;
513 list_for_each_entry(app
, &ipvs
->app_list
, a_list
) {
514 list_for_each_entry(inc
, &app
->incs_list
, a_list
) {
523 static void *ip_vs_app_seq_start(struct seq_file
*seq
, loff_t
*pos
)
525 struct net
*net
= seq_file_net(seq
);
526 struct netns_ipvs
*ipvs
= net_ipvs(net
);
528 mutex_lock(&__ip_vs_app_mutex
);
530 return *pos
? ip_vs_app_idx(ipvs
, *pos
- 1) : SEQ_START_TOKEN
;
533 static void *ip_vs_app_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
535 struct ip_vs_app
*inc
, *app
;
537 struct net
*net
= seq_file_net(seq
);
538 struct netns_ipvs
*ipvs
= net_ipvs(net
);
541 if (v
== SEQ_START_TOKEN
)
542 return ip_vs_app_idx(ipvs
, 0);
547 if ((e
= inc
->a_list
.next
) != &app
->incs_list
)
548 return list_entry(e
, struct ip_vs_app
, a_list
);
550 /* go on to next application */
551 for (e
= app
->a_list
.next
; e
!= &ipvs
->app_list
; e
= e
->next
) {
552 app
= list_entry(e
, struct ip_vs_app
, a_list
);
553 list_for_each_entry(inc
, &app
->incs_list
, a_list
) {
560 static void ip_vs_app_seq_stop(struct seq_file
*seq
, void *v
)
562 mutex_unlock(&__ip_vs_app_mutex
);
565 static int ip_vs_app_seq_show(struct seq_file
*seq
, void *v
)
567 if (v
== SEQ_START_TOKEN
)
568 seq_puts(seq
, "prot port usecnt name\n");
570 const struct ip_vs_app
*inc
= v
;
572 seq_printf(seq
, "%-3s %-7u %-6d %-17s\n",
573 ip_vs_proto_name(inc
->protocol
),
575 atomic_read(&inc
->usecnt
),
581 static const struct seq_operations ip_vs_app_seq_ops
= {
582 .start
= ip_vs_app_seq_start
,
583 .next
= ip_vs_app_seq_next
,
584 .stop
= ip_vs_app_seq_stop
,
585 .show
= ip_vs_app_seq_show
,
588 static int ip_vs_app_open(struct inode
*inode
, struct file
*file
)
590 return seq_open_net(inode
, file
, &ip_vs_app_seq_ops
,
591 sizeof(struct seq_net_private
));
594 static const struct file_operations ip_vs_app_fops
= {
595 .owner
= THIS_MODULE
,
596 .open
= ip_vs_app_open
,
599 .release
= seq_release_net
,
603 int __net_init
ip_vs_app_net_init(struct net
*net
)
605 struct netns_ipvs
*ipvs
= net_ipvs(net
);
607 INIT_LIST_HEAD(&ipvs
->app_list
);
608 proc_create("ip_vs_app", 0, net
->proc_net
, &ip_vs_app_fops
);
612 void __net_exit
ip_vs_app_net_cleanup(struct net
*net
)
614 unregister_ip_vs_app(net
, NULL
/* all */);
615 remove_proc_entry("ip_vs_app", net
->proc_net
);