blob: a150ff5d838cf2858fa4a25c7a83839ac135e083 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * IP Virtual Server
3 * data structure and functionality definitions
4 */
5
Julius Volzbc4768e2008-07-31 20:45:24 -07006#ifndef _NET_IP_VS_H
7#define _NET_IP_VS_H
Linus Torvalds1da177e2005-04-16 15:20:36 -07008
Julius Volzbc4768e2008-07-31 20:45:24 -07009#include <linux/ip_vs.h> /* definitions shared with userland */
Linus Torvalds1da177e2005-04-16 15:20:36 -070010
Julius Volzbc4768e2008-07-31 20:45:24 -070011#include <asm/types.h> /* for __uXX types */
12
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/list.h> /* for struct list_head */
14#include <linux/spinlock.h> /* for struct rwlock_t */
Arun Sharma600634972011-07-26 16:09:06 -070015#include <linux/atomic.h> /* for struct atomic_t */
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/compiler.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020017#include <linux/timer.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050018#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020020#include <net/checksum.h>
Julius Volze7ade462008-09-02 15:55:33 +020021#include <linux/netfilter.h> /* for union nf_inet_addr */
KOVACS Krisztian1668e012008-10-01 07:33:10 -070022#include <linux/ip.h>
Julius Volze7ade462008-09-02 15:55:33 +020023#include <linux/ipv6.h> /* for struct ipv6hdr */
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +000024#include <net/ipv6.h>
Jesper Dangaard Brouer54d83ef2012-10-22 21:22:45 +020025#if IS_ENABLED(CONFIG_IP_VS_IPV6)
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +020026#include <linux/netfilter_ipv6/ip6_tables.h>
27#endif
Jesper Dangaard Brouera638e5142012-09-26 14:06:20 +020028#if IS_ENABLED(CONFIG_NF_CONNTRACK)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +020029#include <net/netfilter/nf_conntrack.h>
30#endif
Hans Schillstrom61b1ab42011-01-03 14:44:42 +010031#include <net/net_namespace.h> /* Netw namespace */
32
33/*
34 * Generic access of ipvs struct
35 */
36static inline struct netns_ipvs *net_ipvs(struct net* net)
37{
38 return net->ipvs;
39}
Hans Schillstromfc723252011-01-03 14:44:43 +010040/*
41 * Get net ptr from skb in traffic cases
42 * use skb_sknet when call is from userland (ioctl or netlink)
43 */
Hans Schillstroma0840e22011-01-03 14:44:58 +010044static inline struct net *skb_net(const struct sk_buff *skb)
Hans Schillstromfc723252011-01-03 14:44:43 +010045{
46#ifdef CONFIG_NET_NS
47#ifdef CONFIG_IP_VS_DEBUG
48 /*
49 * This is used for debug only.
50 * Start with the most likely hit
51 * End with BUG
52 */
53 if (likely(skb->dev && skb->dev->nd_net))
54 return dev_net(skb->dev);
Hans Schillstroma09d1972011-04-04 15:25:18 +020055 if (skb_dst(skb) && skb_dst(skb)->dev)
Hans Schillstromfc723252011-01-03 14:44:43 +010056 return dev_net(skb_dst(skb)->dev);
57 WARN(skb->sk, "Maybe skb_sknet should be used in %s() at line:%d\n",
58 __func__, __LINE__);
59 if (likely(skb->sk && skb->sk->sk_net))
60 return sock_net(skb->sk);
61 pr_err("There is no net ptr to find in the skb in %s() line:%d\n",
62 __func__, __LINE__);
63 BUG();
64#else
65 return dev_net(skb->dev ? : skb_dst(skb)->dev);
66#endif
67#else
68 return &init_net;
69#endif
70}
71
Hans Schillstroma0840e22011-01-03 14:44:58 +010072static inline struct net *skb_sknet(const struct sk_buff *skb)
Hans Schillstromfc723252011-01-03 14:44:43 +010073{
74#ifdef CONFIG_NET_NS
75#ifdef CONFIG_IP_VS_DEBUG
76 /* Start with the most likely hit */
77 if (likely(skb->sk && skb->sk->sk_net))
78 return sock_net(skb->sk);
79 WARN(skb->dev, "Maybe skb_net should be used instead in %s() line:%d\n",
80 __func__, __LINE__);
81 if (likely(skb->dev && skb->dev->nd_net))
82 return dev_net(skb->dev);
83 pr_err("There is no net ptr to find in the skb in %s() line:%d\n",
84 __func__, __LINE__);
85 BUG();
86#else
87 return sock_net(skb->sk);
88#endif
89#else
90 return &init_net;
91#endif
92}
Hans Schillstromb17fc992011-01-03 14:44:56 +010093/*
94 * This one needed for single_open_net since net is stored directly in
Lucas De Marchi25985ed2011-03-30 22:57:33 -030095 * private not as a struct i.e. seq_file_net can't be used.
Hans Schillstromb17fc992011-01-03 14:44:56 +010096 */
97static inline struct net *seq_file_single_net(struct seq_file *seq)
98{
99#ifdef CONFIG_NET_NS
100 return (struct net *)seq->private;
101#else
102 return &init_net;
103#endif
104}
Catalin(ux) M. BOIE6f7edb42010-01-05 05:50:24 +0100105
106/* Connections' size value needed by ip_vs_ctl.c */
107extern int ip_vs_conn_tab_size;
108
Julius Volz64aae3c2008-09-02 15:55:34 +0200109struct ip_vs_iphdr {
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200110 __u32 len; /* IPv4 simply where L4 starts
111 IPv6 where L4 Transport Header starts */
Jesper Dangaard Brouer2f747132012-09-26 14:06:59 +0200112 __u32 thoff_reasm; /* Transport Header Offset in nfct_reasm skb */
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200113 __u16 fragoffs; /* IPv6 fragment offset, 0 if first frag (or not frag)*/
114 __s16 protocol;
115 __s32 flags;
Julius Volz64aae3c2008-09-02 15:55:34 +0200116 union nf_inet_addr saddr;
117 union nf_inet_addr daddr;
118};
119
Jesper Dangaard Brouer2f747132012-09-26 14:06:59 +0200120/* Dependency to module: nf_defrag_ipv6 */
121#if defined(CONFIG_NF_DEFRAG_IPV6) || defined(CONFIG_NF_DEFRAG_IPV6_MODULE)
122static inline struct sk_buff *skb_nfct_reasm(const struct sk_buff *skb)
123{
124 return skb->nfct_reasm;
125}
126static inline void *frag_safe_skb_hp(const struct sk_buff *skb, int offset,
127 int len, void *buffer,
128 const struct ip_vs_iphdr *ipvsh)
129{
130 if (unlikely(ipvsh->fragoffs && skb_nfct_reasm(skb)))
131 return skb_header_pointer(skb_nfct_reasm(skb),
132 ipvsh->thoff_reasm, len, buffer);
133
134 return skb_header_pointer(skb, offset, len, buffer);
135}
136#else
137static inline struct sk_buff *skb_nfct_reasm(const struct sk_buff *skb)
138{
139 return NULL;
140}
141static inline void *frag_safe_skb_hp(const struct sk_buff *skb, int offset,
142 int len, void *buffer,
143 const struct ip_vs_iphdr *ipvsh)
144{
145 return skb_header_pointer(skb, offset, len, buffer);
146}
147#endif
148
Julius Volz64aae3c2008-09-02 15:55:34 +0200149static inline void
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200150ip_vs_fill_ip4hdr(const void *nh, struct ip_vs_iphdr *iphdr)
151{
152 const struct iphdr *iph = nh;
153
154 iphdr->len = iph->ihl * 4;
155 iphdr->fragoffs = 0;
156 iphdr->protocol = iph->protocol;
157 iphdr->saddr.ip = iph->saddr;
158 iphdr->daddr.ip = iph->daddr;
159}
160
161/* This function handles filling *ip_vs_iphdr, both for IPv4 and IPv6.
162 * IPv6 requires some extra work, as finding proper header position,
163 * depend on the IPv6 extension headers.
164 */
165static inline void
166ip_vs_fill_iph_skb(int af, const struct sk_buff *skb, struct ip_vs_iphdr *iphdr)
Julius Volz64aae3c2008-09-02 15:55:34 +0200167{
168#ifdef CONFIG_IP_VS_IPV6
169 if (af == AF_INET6) {
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200170 const struct ipv6hdr *iph =
171 (struct ipv6hdr *)skb_network_header(skb);
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000172 iphdr->saddr.in6 = iph->saddr;
173 iphdr->daddr.in6 = iph->daddr;
Jesper Dangaard Brouer2f747132012-09-26 14:06:59 +0200174 /* ipv6_find_hdr() updates len, flags, thoff_reasm */
175 iphdr->thoff_reasm = 0;
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200176 iphdr->len = 0;
177 iphdr->flags = 0;
178 iphdr->protocol = ipv6_find_hdr(skb, &iphdr->len, -1,
179 &iphdr->fragoffs,
180 &iphdr->flags);
Jesper Dangaard Brouer2f747132012-09-26 14:06:59 +0200181 /* get proto from re-assembled packet and it's offset */
182 if (skb_nfct_reasm(skb))
183 iphdr->protocol = ipv6_find_hdr(skb_nfct_reasm(skb),
184 &iphdr->thoff_reasm,
185 -1, NULL, NULL);
186
Julius Volz64aae3c2008-09-02 15:55:34 +0200187 } else
188#endif
189 {
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200190 const struct iphdr *iph =
191 (struct iphdr *)skb_network_header(skb);
192 iphdr->len = iph->ihl * 4;
193 iphdr->fragoffs = 0;
194 iphdr->protocol = iph->protocol;
195 iphdr->saddr.ip = iph->saddr;
196 iphdr->daddr.ip = iph->daddr;
197 }
198}
199
200/* This function is a faster version of ip_vs_fill_iph_skb().
201 * Where we only populate {s,d}addr (and avoid calling ipv6_find_hdr()).
202 * This is used by the some of the ip_vs_*_schedule() functions.
203 * (Mostly done to avoid ABI breakage of external schedulers)
204 */
205static inline void
206ip_vs_fill_iph_addr_only(int af, const struct sk_buff *skb,
207 struct ip_vs_iphdr *iphdr)
208{
209#ifdef CONFIG_IP_VS_IPV6
210 if (af == AF_INET6) {
211 const struct ipv6hdr *iph =
212 (struct ipv6hdr *)skb_network_header(skb);
213 iphdr->saddr.in6 = iph->saddr;
214 iphdr->daddr.in6 = iph->daddr;
Jesper Dangaard Brouer54d83ef2012-10-22 21:22:45 +0200215 } else
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200216#endif
Jesper Dangaard Brouer54d83ef2012-10-22 21:22:45 +0200217 {
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200218 const struct iphdr *iph =
219 (struct iphdr *)skb_network_header(skb);
Julius Volz64aae3c2008-09-02 15:55:34 +0200220 iphdr->saddr.ip = iph->saddr;
221 iphdr->daddr.ip = iph->daddr;
222 }
223}
224
225static inline void ip_vs_addr_copy(int af, union nf_inet_addr *dst,
226 const union nf_inet_addr *src)
227{
228#ifdef CONFIG_IP_VS_IPV6
229 if (af == AF_INET6)
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000230 dst->in6 = src->in6;
Julius Volz64aae3c2008-09-02 15:55:34 +0200231 else
232#endif
233 dst->ip = src->ip;
234}
235
236static inline int ip_vs_addr_equal(int af, const union nf_inet_addr *a,
237 const union nf_inet_addr *b)
238{
239#ifdef CONFIG_IP_VS_IPV6
240 if (af == AF_INET6)
241 return ipv6_addr_equal(&a->in6, &b->in6);
242#endif
243 return a->ip == b->ip;
244}
245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246#ifdef CONFIG_IP_VS_DEBUG
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200247#include <linux/net.h>
248
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249extern int ip_vs_get_debug_level(void);
Julius Volzc842a3a2008-09-02 15:55:35 +0200250
251static inline const char *ip_vs_dbg_addr(int af, char *buf, size_t buf_len,
252 const union nf_inet_addr *addr,
253 int *idx)
254{
255 int len;
256#ifdef CONFIG_IP_VS_IPV6
257 if (af == AF_INET6)
Jesper Dangaard Brouer120b9c12012-09-26 14:05:53 +0200258 len = snprintf(&buf[*idx], buf_len - *idx, "[%pI6c]",
Harvey Harrison0c6ce782008-10-28 16:09:23 -0700259 &addr->in6) + 1;
Julius Volzc842a3a2008-09-02 15:55:35 +0200260 else
261#endif
Harvey Harrison3685f25d2008-10-31 00:56:49 -0700262 len = snprintf(&buf[*idx], buf_len - *idx, "%pI4",
263 &addr->ip) + 1;
Julius Volzc842a3a2008-09-02 15:55:35 +0200264
265 *idx += len;
266 BUG_ON(*idx > buf_len + 1);
267 return &buf[*idx - len];
268}
269
Hannes Eder9aada7a2009-07-30 14:29:44 -0700270#define IP_VS_DBG_BUF(level, msg, ...) \
271 do { \
272 char ip_vs_dbg_buf[160]; \
273 int ip_vs_dbg_idx = 0; \
274 if (level <= ip_vs_get_debug_level()) \
275 printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \
276 } while (0)
277#define IP_VS_ERR_BUF(msg...) \
278 do { \
279 char ip_vs_dbg_buf[160]; \
280 int ip_vs_dbg_idx = 0; \
281 pr_err(msg); \
282 } while (0)
Julius Volzc842a3a2008-09-02 15:55:35 +0200283
284/* Only use from within IP_VS_DBG_BUF() or IP_VS_ERR_BUF macros */
Hannes Eder9aada7a2009-07-30 14:29:44 -0700285#define IP_VS_DBG_ADDR(af, addr) \
286 ip_vs_dbg_addr(af, ip_vs_dbg_buf, \
287 sizeof(ip_vs_dbg_buf), addr, \
288 &ip_vs_dbg_idx)
Julius Volzc842a3a2008-09-02 15:55:35 +0200289
Hannes Eder9aada7a2009-07-30 14:29:44 -0700290#define IP_VS_DBG(level, msg, ...) \
291 do { \
292 if (level <= ip_vs_get_debug_level()) \
293 printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \
294 } while (0)
295#define IP_VS_DBG_RL(msg, ...) \
296 do { \
297 if (net_ratelimit()) \
298 printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \
299 } while (0)
Julian Anastasov0d796412010-10-17 16:46:17 +0300300#define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700301 do { \
302 if (level <= ip_vs_get_debug_level()) \
Julian Anastasov0d796412010-10-17 16:46:17 +0300303 pp->debug_packet(af, pp, skb, ofs, msg); \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700304 } while (0)
Julian Anastasov0d796412010-10-17 16:46:17 +0300305#define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700306 do { \
307 if (level <= ip_vs_get_debug_level() && \
308 net_ratelimit()) \
Julian Anastasov0d796412010-10-17 16:46:17 +0300309 pp->debug_packet(af, pp, skb, ofs, msg); \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700310 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311#else /* NO DEBUGGING at ALL */
Julius Volzc842a3a2008-09-02 15:55:35 +0200312#define IP_VS_DBG_BUF(level, msg...) do {} while (0)
313#define IP_VS_ERR_BUF(msg...) do {} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314#define IP_VS_DBG(level, msg...) do {} while (0)
315#define IP_VS_DBG_RL(msg...) do {} while (0)
Julian Anastasov0d796412010-10-17 16:46:17 +0300316#define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) do {} while (0)
317#define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) do {} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318#endif
319
320#define IP_VS_BUG() BUG()
Hannes Eder1e3e2382009-08-02 11:05:41 +0000321#define IP_VS_ERR_RL(msg, ...) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700322 do { \
323 if (net_ratelimit()) \
Hannes Eder1e3e2382009-08-02 11:05:41 +0000324 pr_err(msg, ##__VA_ARGS__); \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700325 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327#ifdef CONFIG_IP_VS_DEBUG
328#define EnterFunction(level) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700329 do { \
330 if (level <= ip_vs_get_debug_level()) \
331 printk(KERN_DEBUG \
332 pr_fmt("Enter: %s, %s line %i\n"), \
333 __func__, __FILE__, __LINE__); \
334 } while (0)
335#define LeaveFunction(level) \
336 do { \
337 if (level <= ip_vs_get_debug_level()) \
338 printk(KERN_DEBUG \
339 pr_fmt("Leave: %s, %s line %i\n"), \
340 __func__, __FILE__, __LINE__); \
341 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342#else
343#define EnterFunction(level) do {} while (0)
344#define LeaveFunction(level) do {} while (0)
345#endif
346
347#define IP_VS_WAIT_WHILE(expr) while (expr) { cpu_relax(); }
348
349
350/*
351 * The port number of FTP service (in network order).
352 */
Harvey Harrisonf3a7c662009-02-14 22:58:35 -0800353#define FTPPORT cpu_to_be16(21)
354#define FTPDATA cpu_to_be16(20)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 * TCP State Values
358 */
359enum {
360 IP_VS_TCP_S_NONE = 0,
361 IP_VS_TCP_S_ESTABLISHED,
362 IP_VS_TCP_S_SYN_SENT,
363 IP_VS_TCP_S_SYN_RECV,
364 IP_VS_TCP_S_FIN_WAIT,
365 IP_VS_TCP_S_TIME_WAIT,
366 IP_VS_TCP_S_CLOSE,
367 IP_VS_TCP_S_CLOSE_WAIT,
368 IP_VS_TCP_S_LAST_ACK,
369 IP_VS_TCP_S_LISTEN,
370 IP_VS_TCP_S_SYNACK,
371 IP_VS_TCP_S_LAST
372};
373
374/*
375 * UDP State Values
376 */
377enum {
378 IP_VS_UDP_S_NORMAL,
379 IP_VS_UDP_S_LAST,
380};
381
382/*
383 * ICMP State Values
384 */
385enum {
386 IP_VS_ICMP_S_NORMAL,
387 IP_VS_ICMP_S_LAST,
388};
389
390/*
Venkata Mohan Reddy2906f662010-02-18 12:31:05 +0100391 * SCTP State Values
392 */
393enum ip_vs_sctp_states {
394 IP_VS_SCTP_S_NONE,
395 IP_VS_SCTP_S_INIT_CLI,
396 IP_VS_SCTP_S_INIT_SER,
397 IP_VS_SCTP_S_INIT_ACK_CLI,
398 IP_VS_SCTP_S_INIT_ACK_SER,
399 IP_VS_SCTP_S_ECHO_CLI,
400 IP_VS_SCTP_S_ECHO_SER,
401 IP_VS_SCTP_S_ESTABLISHED,
402 IP_VS_SCTP_S_SHUT_CLI,
403 IP_VS_SCTP_S_SHUT_SER,
404 IP_VS_SCTP_S_SHUT_ACK_CLI,
405 IP_VS_SCTP_S_SHUT_ACK_SER,
406 IP_VS_SCTP_S_CLOSED,
407 IP_VS_SCTP_S_LAST
408};
409
410/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 * Delta sequence info structure
412 * Each ip_vs_conn has 2 (output AND input seq. changes).
413 * Only used in the VS/NAT.
414 */
415struct ip_vs_seq {
416 __u32 init_seq; /* Add delta from this seq */
417 __u32 delta; /* Delta in sequence numbers */
418 __u32 previous_delta; /* Delta in sequence numbers
419 before last resized pkt */
420};
421
Hans Schillstromb17fc992011-01-03 14:44:56 +0100422/*
423 * counters per cpu
424 */
425struct ip_vs_counters {
426 __u32 conns; /* connections scheduled */
427 __u32 inpkts; /* incoming packets */
428 __u32 outpkts; /* outgoing packets */
429 __u64 inbytes; /* incoming bytes */
430 __u64 outbytes; /* outgoing bytes */
431};
432/*
433 * Stats per cpu
434 */
435struct ip_vs_cpu_stats {
436 struct ip_vs_counters ustats;
437 struct u64_stats_sync syncp;
438};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440/*
Sven Wegener3a14a312008-08-10 18:24:41 +0000441 * IPVS statistics objects
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 */
Sven Wegener3a14a312008-08-10 18:24:41 +0000443struct ip_vs_estimator {
444 struct list_head list;
445
446 u64 last_inbytes;
447 u64 last_outbytes;
448 u32 last_conns;
449 u32 last_inpkts;
450 u32 last_outpkts;
451
452 u32 cps;
453 u32 inpps;
454 u32 outpps;
455 u32 inbps;
456 u32 outbps;
457};
458
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000459struct ip_vs_stats {
Hans Schillstromb17fc992011-01-03 14:44:56 +0100460 struct ip_vs_stats_user ustats; /* statistics */
Sven Wegenere9c0ce22008-09-08 13:39:04 +0200461 struct ip_vs_estimator est; /* estimator */
Julian Anastasovb962abd2013-03-09 23:25:08 +0200462 struct ip_vs_cpu_stats __percpu *cpustats; /* per cpu counters */
Hans Schillstromb17fc992011-01-03 14:44:56 +0100463 spinlock_t lock; /* spin lock */
Julian Anastasov55a3d4e2011-03-14 01:37:49 +0200464 struct ip_vs_stats_user ustats0; /* reset values */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465};
466
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200467struct dst_entry;
468struct iphdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469struct ip_vs_conn;
470struct ip_vs_app;
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200471struct sk_buff;
Hans Schillstrom252c6412011-01-03 14:44:46 +0100472struct ip_vs_proto_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474struct ip_vs_protocol {
475 struct ip_vs_protocol *next;
476 char *name;
Julian Anastasov2ad17de2008-04-29 03:21:23 -0700477 u16 protocol;
478 u16 num_states;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 int dont_defrag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481 void (*init)(struct ip_vs_protocol *pp);
482
483 void (*exit)(struct ip_vs_protocol *pp);
484
Hans Schillstrom582b8e32012-04-26 09:45:35 +0200485 int (*init_netns)(struct net *net, struct ip_vs_proto_data *pd);
Hans Schillstrom252c6412011-01-03 14:44:46 +0100486
487 void (*exit_netns)(struct net *net, struct ip_vs_proto_data *pd);
488
Julius Volz51ef3482008-09-02 15:55:40 +0200489 int (*conn_schedule)(int af, struct sk_buff *skb,
Hans Schillstrom93304192011-01-03 14:44:51 +0100490 struct ip_vs_proto_data *pd,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200491 int *verdict, struct ip_vs_conn **cpp,
492 struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 struct ip_vs_conn *
Julius Volz51ef3482008-09-02 15:55:40 +0200495 (*conn_in_get)(int af,
496 const struct sk_buff *skb,
Julius Volz51ef3482008-09-02 15:55:40 +0200497 const struct ip_vs_iphdr *iph,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 int inverse);
499
500 struct ip_vs_conn *
Julius Volz51ef3482008-09-02 15:55:40 +0200501 (*conn_out_get)(int af,
502 const struct sk_buff *skb,
Julius Volz51ef3482008-09-02 15:55:40 +0200503 const struct ip_vs_iphdr *iph,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 int inverse);
505
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200506 int (*snat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp,
507 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200509 int (*dnat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp,
510 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Julius Volz51ef3482008-09-02 15:55:40 +0200512 int (*csum_check)(int af, struct sk_buff *skb,
513 struct ip_vs_protocol *pp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
515 const char *(*state_name)(int state);
516
Simon Horman4a516f12011-09-16 14:11:49 +0900517 void (*state_transition)(struct ip_vs_conn *cp, int direction,
518 const struct sk_buff *skb,
519 struct ip_vs_proto_data *pd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Hans Schillstromab8a5e82011-01-03 14:44:53 +0100521 int (*register_app)(struct net *net, struct ip_vs_app *inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
Hans Schillstromab8a5e82011-01-03 14:44:53 +0100523 void (*unregister_app)(struct net *net, struct ip_vs_app *inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525 int (*app_conn_bind)(struct ip_vs_conn *cp);
526
Julian Anastasov0d796412010-10-17 16:46:17 +0300527 void (*debug_packet)(int af, struct ip_vs_protocol *pp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 const struct sk_buff *skb,
529 int offset,
530 const char *msg);
531
Hans Schillstrom93304192011-01-03 14:44:51 +0100532 void (*timeout_change)(struct ip_vs_proto_data *pd, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533};
534
Hans Schillstrom252c6412011-01-03 14:44:46 +0100535/*
536 * protocol data per netns
537 */
538struct ip_vs_proto_data {
539 struct ip_vs_proto_data *next;
540 struct ip_vs_protocol *pp;
541 int *timeout_table; /* protocol timeout table */
542 atomic_t appcnt; /* counter of proto app incs. */
543 struct tcp_states_t *tcp_state_table;
544};
545
546extern struct ip_vs_protocol *ip_vs_proto_get(unsigned short proto);
547extern struct ip_vs_proto_data *ip_vs_proto_data_get(struct net *net,
548 unsigned short proto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Simon Hormanf11017e2010-08-22 21:37:52 +0900550struct ip_vs_conn_param {
Hans Schillstrom6e67e582011-01-03 14:44:57 +0100551 struct net *net;
Simon Hormanf11017e2010-08-22 21:37:52 +0900552 const union nf_inet_addr *caddr;
553 const union nf_inet_addr *vaddr;
554 __be16 cport;
555 __be16 vport;
556 __u16 protocol;
557 u16 af;
Simon Horman85999282010-08-22 21:37:53 +0900558
559 const struct ip_vs_pe *pe;
560 char *pe_data;
561 __u8 pe_data_len;
Simon Hormanf11017e2010-08-22 21:37:52 +0900562};
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564/*
565 * IP_VS structure allocated for each dynamically scheduled connection
566 */
567struct ip_vs_conn {
Changli Gao731109e2011-02-19 18:05:08 +0800568 struct hlist_node c_list; /* hashed list heads */
Hans Schillstrom6e67e582011-01-03 14:44:57 +0100569#ifdef CONFIG_NET_NS
570 struct net *net; /* Name space */
571#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 /* Protocol, addresses and port numbers */
Hans Schillstrom6e67e582011-01-03 14:44:57 +0100573 u16 af; /* address family */
574 __be16 cport;
575 __be16 vport;
576 __be16 dport;
577 __u32 fwmark; /* Fire wall mark from skb */
578 union nf_inet_addr caddr; /* client address */
579 union nf_inet_addr vaddr; /* virtual address */
580 union nf_inet_addr daddr; /* destination address */
581 volatile __u32 flags; /* status flags */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 __u16 protocol; /* Which protocol (TCP/UDP) */
583
584 /* counter and timer */
585 atomic_t refcnt; /* reference count */
586 struct timer_list timer; /* Expiration timer */
587 volatile unsigned long timeout; /* timeout */
588
589 /* Flags and state transition */
590 spinlock_t lock; /* lock for state transition */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 volatile __u16 state; /* state info */
Rumen G. Bogdanovskiefac5272007-11-07 02:36:55 -0800592 volatile __u16 old_state; /* old state, to be used for
593 * state transition triggerd
594 * synchronization
595 */
Julian Anastasov749c42b2012-04-24 23:46:40 +0300596 unsigned long sync_endtime; /* jiffies + sent_retries */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598 /* Control members */
599 struct ip_vs_conn *control; /* Master control connection */
600 atomic_t n_control; /* Number of controlled ones */
601 struct ip_vs_dest *dest; /* real server */
602 atomic_t in_pkts; /* incoming packet counter */
603
604 /* packet transmitter for different forwarding methods. If it
605 mangles the packet, it must return NF_DROP or better NF_STOLEN,
606 otherwise this must be changed to a sk_buff **.
Julian Anastasovfc604762010-10-17 16:38:15 +0300607 NF_ACCEPT can be returned when destination is local.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 */
609 int (*packet_xmit)(struct sk_buff *skb, struct ip_vs_conn *cp,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200610 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 /* Note: we can group the following members into a structure,
613 in order to save more space, and the following members are
614 only used in VS/NAT anyway */
615 struct ip_vs_app *app; /* bound ip_vs_app object */
616 void *app_data; /* Application private data */
617 struct ip_vs_seq in_seq; /* incoming seq. struct */
618 struct ip_vs_seq out_seq; /* outgoing seq. struct */
Simon Horman85999282010-08-22 21:37:53 +0900619
Simon Hormane9e5eee2010-11-08 20:05:57 +0900620 const struct ip_vs_pe *pe;
Simon Horman85999282010-08-22 21:37:53 +0900621 char *pe_data;
622 __u8 pe_data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623};
624
Hans Schillstrom6e67e582011-01-03 14:44:57 +0100625/*
626 * To save some memory in conn table when name space is disabled.
627 */
628static inline struct net *ip_vs_conn_net(const struct ip_vs_conn *cp)
629{
630#ifdef CONFIG_NET_NS
631 return cp->net;
632#else
633 return &init_net;
634#endif
635}
636static inline void ip_vs_conn_net_set(struct ip_vs_conn *cp, struct net *net)
637{
638#ifdef CONFIG_NET_NS
639 cp->net = net;
640#endif
641}
642
643static inline int ip_vs_conn_net_eq(const struct ip_vs_conn *cp,
644 struct net *net)
645{
646#ifdef CONFIG_NET_NS
647 return cp->net == net;
648#else
649 return 1;
650#endif
651}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653/*
Julius Volzc860c6b2008-09-02 15:55:36 +0200654 * Extended internal versions of struct ip_vs_service_user and
655 * ip_vs_dest_user for IPv6 support.
656 *
657 * We need these to conveniently pass around service and destination
658 * options, but unfortunately, we also need to keep the old definitions to
659 * maintain userspace backwards compatibility for the setsockopt interface.
660 */
661struct ip_vs_service_user_kern {
662 /* virtual service addresses */
663 u16 af;
664 u16 protocol;
665 union nf_inet_addr addr; /* virtual ip address */
666 u16 port;
667 u32 fwmark; /* firwall mark of service */
668
669 /* virtual service options */
670 char *sched_name;
Simon Horman0d1e71b2010-08-22 21:37:54 +0900671 char *pe_name;
Eric Dumazet95c96172012-04-15 05:58:06 +0000672 unsigned int flags; /* virtual service flags */
673 unsigned int timeout; /* persistent timeout in sec */
Julius Volzc860c6b2008-09-02 15:55:36 +0200674 u32 netmask; /* persistent netmask */
675};
676
677
678struct ip_vs_dest_user_kern {
679 /* destination server address */
680 union nf_inet_addr addr;
681 u16 port;
682
683 /* real server options */
Eric Dumazet95c96172012-04-15 05:58:06 +0000684 unsigned int conn_flags; /* connection flags */
Julius Volzc860c6b2008-09-02 15:55:36 +0200685 int weight; /* destination weight */
686
687 /* thresholds for active connections */
688 u32 u_threshold; /* upper threshold */
689 u32 l_threshold; /* lower threshold */
690};
691
692
693/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 * The information about the virtual service offered to the net
695 * and the forwarding entries
696 */
697struct ip_vs_service {
698 struct list_head s_list; /* for normal service table */
699 struct list_head f_list; /* for fwmark-based service table */
700 atomic_t refcnt; /* reference counter */
701 atomic_t usecnt; /* use counter */
702
Julius Volze7ade462008-09-02 15:55:33 +0200703 u16 af; /* address family */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 __u16 protocol; /* which protocol (TCP/UDP) */
Julius Volze7ade462008-09-02 15:55:33 +0200705 union nf_inet_addr addr; /* IP address for virtual service */
Al Viro014d7302006-09-28 14:29:52 -0700706 __be16 port; /* port number for the service */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 __u32 fwmark; /* firewall mark of the service */
Eric Dumazet95c96172012-04-15 05:58:06 +0000708 unsigned int flags; /* service status flags */
709 unsigned int timeout; /* persistent timeout in ticks */
Al Viro014d7302006-09-28 14:29:52 -0700710 __be32 netmask; /* grouping granularity */
Hans Schillstromfc723252011-01-03 14:44:43 +0100711 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
713 struct list_head destinations; /* real server d-linked list */
714 __u32 num_dests; /* number of servers */
715 struct ip_vs_stats stats; /* statistics for the service */
716 struct ip_vs_app *inc; /* bind conns to this app inc */
717
718 /* for scheduling */
719 struct ip_vs_scheduler *scheduler; /* bound scheduler object */
720 rwlock_t sched_lock; /* lock sched_data */
721 void *sched_data; /* scheduler application data */
Simon Horman85999282010-08-22 21:37:53 +0900722
723 /* alternate persistence engine */
724 struct ip_vs_pe *pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725};
726
Julian Anastasov026ace02013-03-21 11:58:06 +0200727/* Information for cached dst */
728struct ip_vs_dest_dst {
729 struct dst_entry *dst_cache; /* destination cache entry */
730 u32 dst_cookie;
731 union nf_inet_addr dst_saddr;
732 struct rcu_head rcu_head;
733};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
735/*
736 * The real server destination forwarding entry
737 * with ip address, port number, and so on.
738 */
739struct ip_vs_dest {
740 struct list_head n_list; /* for the dests in the service */
741 struct list_head d_list; /* for table with all the dests */
742
Julius Volze7ade462008-09-02 15:55:33 +0200743 u16 af; /* address family */
Al Viro014d7302006-09-28 14:29:52 -0700744 __be16 port; /* port number of the server */
Hans Schillstromf2431e62011-01-03 14:45:00 +0100745 union nf_inet_addr addr; /* IP address of the server */
Eric Dumazet95c96172012-04-15 05:58:06 +0000746 volatile unsigned int flags; /* dest status flags */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 atomic_t conn_flags; /* flags to copy to conn */
748 atomic_t weight; /* server weight */
749
750 atomic_t refcnt; /* reference counter */
751 struct ip_vs_stats stats; /* statistics */
752
753 /* connection counters and thresholds */
754 atomic_t activeconns; /* active connections */
755 atomic_t inactconns; /* inactive connections */
756 atomic_t persistconns; /* persistent connections */
757 __u32 u_threshold; /* upper threshold */
758 __u32 l_threshold; /* lower threshold */
759
760 /* for destination cache */
761 spinlock_t dst_lock; /* lock of dst_cache */
Julian Anastasov026ace02013-03-21 11:58:06 +0200762 struct ip_vs_dest_dst __rcu *dest_dst; /* cached dst info */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
764 /* for virtual service */
765 struct ip_vs_service *svc; /* service it belongs to */
766 __u16 protocol; /* which protocol (TCP/UDP) */
Al Viro014d7302006-09-28 14:29:52 -0700767 __be16 vport; /* virtual port number */
Hans Schillstromf2431e62011-01-03 14:45:00 +0100768 union nf_inet_addr vaddr; /* virtual IP address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 __u32 vfwmark; /* firewall mark of service */
770};
771
772
773/*
774 * The scheduler object
775 */
776struct ip_vs_scheduler {
777 struct list_head n_list; /* d-linked list head */
778 char *name; /* scheduler name */
779 atomic_t refcnt; /* reference counter */
780 struct module *module; /* THIS_MODULE/NULL */
781
782 /* scheduler initializing service */
783 int (*init_service)(struct ip_vs_service *svc);
784 /* scheduling service finish */
785 int (*done_service)(struct ip_vs_service *svc);
786 /* scheduler updating service */
787 int (*update_service)(struct ip_vs_service *svc);
788
789 /* selecting a server from the given service */
790 struct ip_vs_dest* (*schedule)(struct ip_vs_service *svc,
791 const struct sk_buff *skb);
792};
793
Simon Horman85999282010-08-22 21:37:53 +0900794/* The persistence engine object */
795struct ip_vs_pe {
796 struct list_head n_list; /* d-linked list head */
797 char *name; /* scheduler name */
798 atomic_t refcnt; /* reference counter */
799 struct module *module; /* THIS_MODULE/NULL */
800
801 /* get the connection template, if any */
802 int (*fill_param)(struct ip_vs_conn_param *p, struct sk_buff *skb);
803 bool (*ct_match)(const struct ip_vs_conn_param *p,
804 struct ip_vs_conn *ct);
805 u32 (*hashkey_raw)(const struct ip_vs_conn_param *p, u32 initval,
806 bool inverse);
Simon Hormana3c918a2010-08-22 21:37:53 +0900807 int (*show_pe_data)(const struct ip_vs_conn *cp, char *buf);
Simon Horman85999282010-08-22 21:37:53 +0900808};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
810/*
811 * The application module object (a.k.a. app incarnation)
812 */
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000813struct ip_vs_app {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 struct list_head a_list; /* member in app list */
815 int type; /* IP_VS_APP_TYPE_xxx */
816 char *name; /* application module name */
817 __u16 protocol;
818 struct module *module; /* THIS_MODULE/NULL */
819 struct list_head incs_list; /* list of incarnations */
820
821 /* members for application incarnations */
822 struct list_head p_list; /* member in proto app list */
823 struct ip_vs_app *app; /* its real application */
Al Viro014d7302006-09-28 14:29:52 -0700824 __be16 port; /* port number in net order */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 atomic_t usecnt; /* usage counter */
826
Julian Anastasov8b27b102010-10-17 16:17:20 +0300827 /*
828 * output hook: Process packet in inout direction, diff set for TCP.
829 * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok,
830 * 2=Mangled but checksum was not updated
831 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 int (*pkt_out)(struct ip_vs_app *, struct ip_vs_conn *,
Herbert Xu3db05fe2007-10-15 00:53:15 -0700833 struct sk_buff *, int *diff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Julian Anastasov8b27b102010-10-17 16:17:20 +0300835 /*
836 * input hook: Process packet in outin direction, diff set for TCP.
837 * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok,
838 * 2=Mangled but checksum was not updated
839 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 int (*pkt_in)(struct ip_vs_app *, struct ip_vs_conn *,
Herbert Xu3db05fe2007-10-15 00:53:15 -0700841 struct sk_buff *, int *diff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 /* ip_vs_app initializer */
844 int (*init_conn)(struct ip_vs_app *, struct ip_vs_conn *);
845
846 /* ip_vs_app finish */
847 int (*done_conn)(struct ip_vs_app *, struct ip_vs_conn *);
848
849
850 /* not used now */
851 int (*bind_conn)(struct ip_vs_app *, struct ip_vs_conn *,
852 struct ip_vs_protocol *);
853
854 void (*unbind_conn)(struct ip_vs_app *, struct ip_vs_conn *);
855
856 int * timeout_table;
857 int * timeouts;
858 int timeouts_size;
859
860 int (*conn_schedule)(struct sk_buff *skb, struct ip_vs_app *app,
861 int *verdict, struct ip_vs_conn **cpp);
862
863 struct ip_vs_conn *
864 (*conn_in_get)(const struct sk_buff *skb, struct ip_vs_app *app,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200865 const struct iphdr *iph, int inverse);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
867 struct ip_vs_conn *
868 (*conn_out_get)(const struct sk_buff *skb, struct ip_vs_app *app,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200869 const struct iphdr *iph, int inverse);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 int (*state_transition)(struct ip_vs_conn *cp, int direction,
872 const struct sk_buff *skb,
873 struct ip_vs_app *app);
874
875 void (*timeout_change)(struct ip_vs_app *app, int flags);
876};
877
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200878struct ipvs_master_sync_state {
879 struct list_head sync_queue;
880 struct ip_vs_sync_buff *sync_buff;
881 int sync_queue_len;
882 unsigned int sync_queue_delay;
883 struct task_struct *master_thread;
884 struct delayed_work master_wakeup_work;
885 struct netns_ipvs *ipvs;
886};
887
Julian Anastasov2553d062011-03-04 12:18:07 +0200888/* IPVS in network namespace */
889struct netns_ipvs {
890 int gen; /* Generation */
Hans Schillstrom7a4f0762011-05-03 22:09:31 +0200891 int enable; /* enable like nf_hooks do */
Julian Anastasov2553d062011-03-04 12:18:07 +0200892 /*
893 * Hash table: for real service lookups
894 */
895 #define IP_VS_RTAB_BITS 4
896 #define IP_VS_RTAB_SIZE (1 << IP_VS_RTAB_BITS)
897 #define IP_VS_RTAB_MASK (IP_VS_RTAB_SIZE - 1)
898
899 struct list_head rs_table[IP_VS_RTAB_SIZE];
900 /* ip_vs_app */
901 struct list_head app_list;
Julian Anastasov2553d062011-03-04 12:18:07 +0200902 /* ip_vs_proto */
903 #define IP_VS_PROTO_TAB_SIZE 32 /* must be power of 2 */
904 struct ip_vs_proto_data *proto_data_table[IP_VS_PROTO_TAB_SIZE];
905 /* ip_vs_proto_tcp */
906#ifdef CONFIG_IP_VS_PROTO_TCP
907 #define TCP_APP_TAB_BITS 4
908 #define TCP_APP_TAB_SIZE (1 << TCP_APP_TAB_BITS)
909 #define TCP_APP_TAB_MASK (TCP_APP_TAB_SIZE - 1)
910 struct list_head tcp_apps[TCP_APP_TAB_SIZE];
911 spinlock_t tcp_app_lock;
912#endif
913 /* ip_vs_proto_udp */
914#ifdef CONFIG_IP_VS_PROTO_UDP
915 #define UDP_APP_TAB_BITS 4
916 #define UDP_APP_TAB_SIZE (1 << UDP_APP_TAB_BITS)
917 #define UDP_APP_TAB_MASK (UDP_APP_TAB_SIZE - 1)
918 struct list_head udp_apps[UDP_APP_TAB_SIZE];
919 spinlock_t udp_app_lock;
920#endif
921 /* ip_vs_proto_sctp */
922#ifdef CONFIG_IP_VS_PROTO_SCTP
923 #define SCTP_APP_TAB_BITS 4
924 #define SCTP_APP_TAB_SIZE (1 << SCTP_APP_TAB_BITS)
925 #define SCTP_APP_TAB_MASK (SCTP_APP_TAB_SIZE - 1)
926 /* Hash table for SCTP application incarnations */
927 struct list_head sctp_apps[SCTP_APP_TAB_SIZE];
928 spinlock_t sctp_app_lock;
929#endif
930 /* ip_vs_conn */
931 atomic_t conn_count; /* connection counter */
932
933 /* ip_vs_ctl */
Julian Anastasov2a0751a2011-03-04 12:20:35 +0200934 struct ip_vs_stats tot_stats; /* Statistics & est. */
Julian Anastasov2553d062011-03-04 12:18:07 +0200935
936 int num_services; /* no of virtual services */
Simon Hormanf2247fb2011-02-04 18:33:02 +0900937
Julian Anastasov2553d062011-03-04 12:18:07 +0200938 rwlock_t rs_lock; /* real services table */
Julian Anastasov2553d062011-03-04 12:18:07 +0200939 /* Trash for destinations */
940 struct list_head dest_trash;
941 /* Service counters */
942 atomic_t ftpsvc_counter;
943 atomic_t nullsvc_counter;
944
Simon Hormanf2247fb2011-02-04 18:33:02 +0900945#ifdef CONFIG_SYSCTL
946 /* 1/rate drop and drop-entry variables */
947 struct delayed_work defense_work; /* Work handler */
948 int drop_rate;
949 int drop_counter;
950 atomic_t dropentry;
951 /* locks in ctl.c */
952 spinlock_t dropentry_lock; /* drop entry handling */
953 spinlock_t droppacket_lock; /* drop packet handling */
954 spinlock_t securetcp_lock; /* state and timeout tables */
955
Julian Anastasov2553d062011-03-04 12:18:07 +0200956 /* sys-ctl struct */
957 struct ctl_table_header *sysctl_hdr;
958 struct ctl_table *sysctl_tbl;
Simon Hormanf2247fb2011-02-04 18:33:02 +0900959#endif
960
Julian Anastasov2553d062011-03-04 12:18:07 +0200961 /* sysctl variables */
962 int sysctl_amemthresh;
963 int sysctl_am_droprate;
964 int sysctl_drop_entry;
965 int sysctl_drop_packet;
966 int sysctl_secure_tcp;
967#ifdef CONFIG_IP_VS_NFCT
968 int sysctl_conntrack;
969#endif
970 int sysctl_snat_reroute;
971 int sysctl_sync_ver;
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200972 int sysctl_sync_ports;
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +0200973 int sysctl_sync_qlen_max;
974 int sysctl_sync_sock_size;
Julian Anastasov2553d062011-03-04 12:18:07 +0200975 int sysctl_cache_bypass;
976 int sysctl_expire_nodest_conn;
977 int sysctl_expire_quiescent_template;
978 int sysctl_sync_threshold[2];
Julian Anastasov749c42b2012-04-24 23:46:40 +0300979 unsigned int sysctl_sync_refresh_period;
980 int sysctl_sync_retries;
Julian Anastasov2553d062011-03-04 12:18:07 +0200981 int sysctl_nat_icmp_send;
Julian Anastasov3654e612012-07-20 11:59:53 +0300982 int sysctl_pmtu_disc;
Julian Anastasov0c125822013-03-09 23:25:04 +0200983 int sysctl_backup_only;
Julian Anastasov2553d062011-03-04 12:18:07 +0200984
985 /* ip_vs_lblc */
986 int sysctl_lblc_expiration;
987 struct ctl_table_header *lblc_ctl_header;
988 struct ctl_table *lblc_ctl_table;
989 /* ip_vs_lblcr */
990 int sysctl_lblcr_expiration;
991 struct ctl_table_header *lblcr_ctl_header;
992 struct ctl_table *lblcr_ctl_table;
993 /* ip_vs_est */
994 struct list_head est_list; /* estimator list */
995 spinlock_t est_lock;
996 struct timer_list est_timer; /* Estimation timer */
997 /* ip_vs_sync */
Julian Anastasov2553d062011-03-04 12:18:07 +0200998 spinlock_t sync_lock;
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200999 struct ipvs_master_sync_state *ms;
Julian Anastasov2553d062011-03-04 12:18:07 +02001000 spinlock_t sync_buff_lock;
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +02001001 struct task_struct **backup_threads;
1002 int threads_mask;
Julian Anastasov2553d062011-03-04 12:18:07 +02001003 int send_mesg_maxlen;
1004 int recv_mesg_maxlen;
1005 volatile int sync_state;
1006 volatile int master_syncid;
1007 volatile int backup_syncid;
Hans Schillstromae1d48b2011-10-11 10:54:35 +09001008 struct mutex sync_mutex;
Julian Anastasov2553d062011-03-04 12:18:07 +02001009 /* multicast interface name */
1010 char master_mcast_ifn[IP_VS_IFNAME_MAXLEN];
1011 char backup_mcast_ifn[IP_VS_IFNAME_MAXLEN];
1012 /* net name space ptr */
1013 struct net *net; /* Needed by timer routines */
1014};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Simon Horman59e03502011-02-04 18:33:01 +09001016#define DEFAULT_SYNC_THRESHOLD 3
1017#define DEFAULT_SYNC_PERIOD 50
Simon Horman7532e8d2011-02-04 18:33:01 +09001018#define DEFAULT_SYNC_VER 1
Julian Anastasov749c42b2012-04-24 23:46:40 +03001019#define DEFAULT_SYNC_REFRESH_PERIOD (0U * HZ)
1020#define DEFAULT_SYNC_RETRIES 0
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +02001021#define IPVS_SYNC_WAKEUP_RATE 8
1022#define IPVS_SYNC_QLEN_MAX (IPVS_SYNC_WAKEUP_RATE * 4)
1023#define IPVS_SYNC_SEND_DELAY (HZ / 50)
1024#define IPVS_SYNC_CHECK_PERIOD HZ
Julian Anastasov749c42b2012-04-24 23:46:40 +03001025#define IPVS_SYNC_FLUSH_TIME (HZ * 2)
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +02001026#define IPVS_SYNC_PORTS_MAX (1 << 6)
Simon Horman59e03502011-02-04 18:33:01 +09001027
1028#ifdef CONFIG_SYSCTL
1029
1030static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs)
1031{
1032 return ipvs->sysctl_sync_threshold[0];
1033}
1034
1035static inline int sysctl_sync_period(struct netns_ipvs *ipvs)
1036{
Julian Anastasov749c42b2012-04-24 23:46:40 +03001037 return ACCESS_ONCE(ipvs->sysctl_sync_threshold[1]);
1038}
1039
1040static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs)
1041{
1042 return ACCESS_ONCE(ipvs->sysctl_sync_refresh_period);
1043}
1044
1045static inline int sysctl_sync_retries(struct netns_ipvs *ipvs)
1046{
1047 return ipvs->sysctl_sync_retries;
Simon Horman59e03502011-02-04 18:33:01 +09001048}
1049
Simon Horman7532e8d2011-02-04 18:33:01 +09001050static inline int sysctl_sync_ver(struct netns_ipvs *ipvs)
1051{
1052 return ipvs->sysctl_sync_ver;
1053}
1054
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +02001055static inline int sysctl_sync_ports(struct netns_ipvs *ipvs)
1056{
1057 return ACCESS_ONCE(ipvs->sysctl_sync_ports);
1058}
1059
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +02001060static inline int sysctl_sync_qlen_max(struct netns_ipvs *ipvs)
1061{
1062 return ipvs->sysctl_sync_qlen_max;
1063}
1064
1065static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs)
1066{
1067 return ipvs->sysctl_sync_sock_size;
1068}
1069
Julian Anastasov3654e612012-07-20 11:59:53 +03001070static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs)
1071{
1072 return ipvs->sysctl_pmtu_disc;
1073}
1074
Julian Anastasov0c125822013-03-09 23:25:04 +02001075static inline int sysctl_backup_only(struct netns_ipvs *ipvs)
1076{
1077 return ipvs->sync_state & IP_VS_STATE_BACKUP &&
1078 ipvs->sysctl_backup_only;
1079}
1080
Simon Horman59e03502011-02-04 18:33:01 +09001081#else
1082
1083static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs)
1084{
1085 return DEFAULT_SYNC_THRESHOLD;
1086}
1087
1088static inline int sysctl_sync_period(struct netns_ipvs *ipvs)
1089{
1090 return DEFAULT_SYNC_PERIOD;
1091}
1092
Julian Anastasov749c42b2012-04-24 23:46:40 +03001093static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs)
1094{
1095 return DEFAULT_SYNC_REFRESH_PERIOD;
1096}
1097
1098static inline int sysctl_sync_retries(struct netns_ipvs *ipvs)
1099{
1100 return DEFAULT_SYNC_RETRIES & 3;
1101}
1102
Simon Horman7532e8d2011-02-04 18:33:01 +09001103static inline int sysctl_sync_ver(struct netns_ipvs *ipvs)
1104{
1105 return DEFAULT_SYNC_VER;
1106}
1107
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +02001108static inline int sysctl_sync_ports(struct netns_ipvs *ipvs)
1109{
1110 return 1;
1111}
1112
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +02001113static inline int sysctl_sync_qlen_max(struct netns_ipvs *ipvs)
1114{
1115 return IPVS_SYNC_QLEN_MAX;
1116}
1117
1118static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs)
1119{
1120 return 0;
1121}
1122
Julian Anastasov3654e612012-07-20 11:59:53 +03001123static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs)
1124{
1125 return 1;
1126}
1127
Julian Anastasov0c125822013-03-09 23:25:04 +02001128static inline int sysctl_backup_only(struct netns_ipvs *ipvs)
1129{
1130 return 0;
1131}
1132
Simon Horman59e03502011-02-04 18:33:01 +09001133#endif
1134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135/*
1136 * IPVS core functions
1137 * (from ip_vs_core.c)
1138 */
Eric Dumazet95c96172012-04-15 05:58:06 +00001139extern const char *ip_vs_proto_name(unsigned int proto);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140extern void ip_vs_init_hash_table(struct list_head *table, int rows);
Sven Wegenerafdd6142008-08-10 09:18:01 +00001141#define IP_VS_INIT_HASH_TABLE(t) ip_vs_init_hash_table((t), ARRAY_SIZE((t)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143#define IP_VS_APP_TYPE_FTP 1
1144
1145/*
1146 * ip_vs_conn handling functions
1147 * (from ip_vs_conn.c)
1148 */
1149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150enum {
1151 IP_VS_DIR_INPUT = 0,
1152 IP_VS_DIR_OUTPUT,
1153 IP_VS_DIR_INPUT_ONLY,
1154 IP_VS_DIR_LAST,
1155};
1156
Hans Schillstrom6e67e582011-01-03 14:44:57 +01001157static inline void ip_vs_conn_fill_param(struct net *net, int af, int protocol,
Simon Hormanf11017e2010-08-22 21:37:52 +09001158 const union nf_inet_addr *caddr,
1159 __be16 cport,
1160 const union nf_inet_addr *vaddr,
1161 __be16 vport,
1162 struct ip_vs_conn_param *p)
1163{
Hans Schillstrom6e67e582011-01-03 14:44:57 +01001164 p->net = net;
Simon Hormanf11017e2010-08-22 21:37:52 +09001165 p->af = af;
1166 p->protocol = protocol;
1167 p->caddr = caddr;
1168 p->cport = cport;
1169 p->vaddr = vaddr;
1170 p->vport = vport;
Simon Horman85999282010-08-22 21:37:53 +09001171 p->pe = NULL;
1172 p->pe_data = NULL;
Simon Hormanf11017e2010-08-22 21:37:52 +09001173}
Julius Volz28364a52008-09-02 15:55:43 +02001174
Simon Hormanf11017e2010-08-22 21:37:52 +09001175struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p);
1176struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p);
Julius Volz28364a52008-09-02 15:55:43 +02001177
Simon Horman5c0d2372010-08-02 17:12:44 +02001178struct ip_vs_conn * ip_vs_conn_in_get_proto(int af, const struct sk_buff *skb,
Simon Horman5c0d2372010-08-02 17:12:44 +02001179 const struct ip_vs_iphdr *iph,
Simon Horman5c0d2372010-08-02 17:12:44 +02001180 int inverse);
1181
Simon Hormanf11017e2010-08-22 21:37:52 +09001182struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Simon Horman5c0d2372010-08-02 17:12:44 +02001184struct ip_vs_conn * ip_vs_conn_out_get_proto(int af, const struct sk_buff *skb,
Simon Horman5c0d2372010-08-02 17:12:44 +02001185 const struct ip_vs_iphdr *iph,
Simon Horman5c0d2372010-08-02 17:12:44 +02001186 int inverse);
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188/* put back the conn without restarting its timer */
1189static inline void __ip_vs_conn_put(struct ip_vs_conn *cp)
1190{
1191 atomic_dec(&cp->refcnt);
1192}
1193extern void ip_vs_conn_put(struct ip_vs_conn *cp);
Al Viro014d7302006-09-28 14:29:52 -07001194extern void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Simon Hormanf11017e2010-08-22 21:37:52 +09001196struct ip_vs_conn *ip_vs_conn_new(const struct ip_vs_conn_param *p,
1197 const union nf_inet_addr *daddr,
Eric Dumazet95c96172012-04-15 05:58:06 +00001198 __be16 dport, unsigned int flags,
Hans Schillstrom0e051e62010-11-19 14:25:07 +01001199 struct ip_vs_dest *dest, __u32 fwmark);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200extern void ip_vs_conn_expire_now(struct ip_vs_conn *cp);
1201
1202extern const char * ip_vs_state_name(__u16 proto, int state);
1203
Hans Schillstrom4a85b962011-01-03 14:44:47 +01001204extern void ip_vs_tcp_conn_listen(struct net *net, struct ip_vs_conn *cp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205extern int ip_vs_check_template(struct ip_vs_conn *ct);
Hans Schillstromf6340ee2011-01-03 14:44:59 +01001206extern void ip_vs_random_dropentry(struct net *net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207extern int ip_vs_conn_init(void);
1208extern void ip_vs_conn_cleanup(void);
1209
1210static inline void ip_vs_control_del(struct ip_vs_conn *cp)
1211{
1212 struct ip_vs_conn *ctl_cp = cp->control;
1213 if (!ctl_cp) {
Julius Volzcfc78c52008-09-02 15:55:53 +02001214 IP_VS_ERR_BUF("request control DEL for uncontrolled: "
1215 "%s:%d to %s:%d\n",
1216 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1217 ntohs(cp->cport),
1218 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
1219 ntohs(cp->vport));
1220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 return;
1222 }
1223
Julius Volzcfc78c52008-09-02 15:55:53 +02001224 IP_VS_DBG_BUF(7, "DELeting control for: "
1225 "cp.dst=%s:%d ctl_cp.dst=%s:%d\n",
1226 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1227 ntohs(cp->cport),
1228 IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr),
1229 ntohs(ctl_cp->cport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
1231 cp->control = NULL;
1232 if (atomic_read(&ctl_cp->n_control) == 0) {
Julius Volzcfc78c52008-09-02 15:55:53 +02001233 IP_VS_ERR_BUF("BUG control DEL with n=0 : "
1234 "%s:%d to %s:%d\n",
1235 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1236 ntohs(cp->cport),
1237 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
1238 ntohs(cp->vport));
1239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 return;
1241 }
1242 atomic_dec(&ctl_cp->n_control);
1243}
1244
1245static inline void
1246ip_vs_control_add(struct ip_vs_conn *cp, struct ip_vs_conn *ctl_cp)
1247{
1248 if (cp->control) {
Julius Volzcfc78c52008-09-02 15:55:53 +02001249 IP_VS_ERR_BUF("request control ADD for already controlled: "
1250 "%s:%d to %s:%d\n",
1251 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1252 ntohs(cp->cport),
1253 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
1254 ntohs(cp->vport));
1255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 ip_vs_control_del(cp);
1257 }
1258
Julius Volzcfc78c52008-09-02 15:55:53 +02001259 IP_VS_DBG_BUF(7, "ADDing control for: "
1260 "cp.dst=%s:%d ctl_cp.dst=%s:%d\n",
1261 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1262 ntohs(cp->cport),
1263 IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr),
1264 ntohs(ctl_cp->cport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266 cp->control = ctl_cp;
1267 atomic_inc(&ctl_cp->n_control);
1268}
1269
Hans Schillstrom7a4f0762011-05-03 22:09:31 +02001270/*
1271 * IPVS netns init & cleanup functions
1272 */
Hans Schillstrom503cf152011-05-01 18:50:16 +02001273extern int ip_vs_estimator_net_init(struct net *net);
1274extern int ip_vs_control_net_init(struct net *net);
1275extern int ip_vs_protocol_net_init(struct net *net);
1276extern int ip_vs_app_net_init(struct net *net);
1277extern int ip_vs_conn_net_init(struct net *net);
1278extern int ip_vs_sync_net_init(struct net *net);
1279extern void ip_vs_conn_net_cleanup(struct net *net);
1280extern void ip_vs_app_net_cleanup(struct net *net);
1281extern void ip_vs_protocol_net_cleanup(struct net *net);
1282extern void ip_vs_control_net_cleanup(struct net *net);
1283extern void ip_vs_estimator_net_cleanup(struct net *net);
1284extern void ip_vs_sync_net_cleanup(struct net *net);
1285extern void ip_vs_service_net_cleanup(struct net *net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
1287/*
1288 * IPVS application functions
1289 * (from ip_vs_app.c)
1290 */
1291#define IP_VS_APP_MAX_PORTS 8
Julian Anastasovbe97fdb2012-07-12 23:06:20 +03001292extern struct ip_vs_app *register_ip_vs_app(struct net *net,
1293 struct ip_vs_app *app);
Hans Schillstromab8a5e82011-01-03 14:44:53 +01001294extern void unregister_ip_vs_app(struct net *net, struct ip_vs_app *app);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295extern int ip_vs_bind_app(struct ip_vs_conn *cp, struct ip_vs_protocol *pp);
1296extern void ip_vs_unbind_app(struct ip_vs_conn *cp);
Hans Schillstromab8a5e82011-01-03 14:44:53 +01001297extern int register_ip_vs_app_inc(struct net *net, struct ip_vs_app *app,
1298 __u16 proto, __u16 port);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299extern int ip_vs_app_inc_get(struct ip_vs_app *inc);
1300extern void ip_vs_app_inc_put(struct ip_vs_app *inc);
1301
Herbert Xu3db05fe2007-10-15 00:53:15 -07001302extern int ip_vs_app_pkt_out(struct ip_vs_conn *, struct sk_buff *skb);
1303extern int ip_vs_app_pkt_in(struct ip_vs_conn *, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Simon Horman8be67a62010-08-22 21:37:54 +09001305void ip_vs_bind_pe(struct ip_vs_service *svc, struct ip_vs_pe *pe);
1306void ip_vs_unbind_pe(struct ip_vs_service *svc);
1307int register_ip_vs_pe(struct ip_vs_pe *pe);
1308int unregister_ip_vs_pe(struct ip_vs_pe *pe);
Simon Hormane9e5eee2010-11-08 20:05:57 +09001309struct ip_vs_pe *ip_vs_pe_getbyname(const char *name);
Hans Schillstromfe5e7a12010-11-19 14:25:12 +01001310struct ip_vs_pe *__ip_vs_pe_getbyname(const char *pe_name);
Simon Hormane9e5eee2010-11-08 20:05:57 +09001311
Paul Gortmaker69e7dae2011-08-29 12:40:15 -04001312/*
1313 * Use a #define to avoid all of module.h just for these trivial ops
1314 */
1315#define ip_vs_pe_get(pe) \
1316 if (pe && pe->module) \
Simon Hormane9e5eee2010-11-08 20:05:57 +09001317 __module_get(pe->module);
Simon Hormane9e5eee2010-11-08 20:05:57 +09001318
Paul Gortmaker69e7dae2011-08-29 12:40:15 -04001319#define ip_vs_pe_put(pe) \
1320 if (pe && pe->module) \
Simon Hormane9e5eee2010-11-08 20:05:57 +09001321 module_put(pe->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
1323/*
1324 * IPVS protocol functions (from ip_vs_proto.c)
1325 */
1326extern int ip_vs_protocol_init(void);
1327extern void ip_vs_protocol_cleanup(void);
Hans Schillstrom93304192011-01-03 14:44:51 +01001328extern void ip_vs_protocol_timeout_change(struct netns_ipvs *ipvs, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329extern int *ip_vs_create_timeout_table(int *table, int size);
1330extern int
Jan Engelhardt36cbd3d2009-08-05 10:42:58 -07001331ip_vs_set_state_timeout(int *table, int num, const char *const *names,
1332 const char *name, int to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333extern void
Julian Anastasov0d796412010-10-17 16:46:17 +03001334ip_vs_tcpudp_debug_packet(int af, struct ip_vs_protocol *pp,
1335 const struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 int offset, const char *msg);
1337
1338extern struct ip_vs_protocol ip_vs_protocol_tcp;
1339extern struct ip_vs_protocol ip_vs_protocol_udp;
1340extern struct ip_vs_protocol ip_vs_protocol_icmp;
1341extern struct ip_vs_protocol ip_vs_protocol_esp;
1342extern struct ip_vs_protocol ip_vs_protocol_ah;
Venkata Mohan Reddy2906f662010-02-18 12:31:05 +01001343extern struct ip_vs_protocol ip_vs_protocol_sctp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
1345/*
1346 * Registering/unregistering scheduler functions
1347 * (from ip_vs_sched.c)
1348 */
1349extern int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
1350extern int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
1351extern int ip_vs_bind_scheduler(struct ip_vs_service *svc,
1352 struct ip_vs_scheduler *scheduler);
1353extern int ip_vs_unbind_scheduler(struct ip_vs_service *svc);
1354extern struct ip_vs_scheduler *ip_vs_scheduler_get(const char *sched_name);
1355extern void ip_vs_scheduler_put(struct ip_vs_scheduler *scheduler);
1356extern struct ip_vs_conn *
Julian Anastasov190ecd22010-10-17 16:24:37 +03001357ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +02001358 struct ip_vs_proto_data *pd, int *ignored,
1359 struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360extern int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +02001361 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
Patrick Schaaf41ac51e2011-02-11 14:01:12 +01001363extern void ip_vs_scheduler_err(struct ip_vs_service *svc, const char *msg);
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
1366/*
1367 * IPVS control data and functions (from ip_vs_ctl.c)
1368 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369extern struct ip_vs_stats ip_vs_stats;
Hans Schillstromb880c1f2010-11-19 14:25:14 +01001370extern int sysctl_ip_vs_sync_ver;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372extern struct ip_vs_service *
Hans Schillstromfc723252011-01-03 14:44:43 +01001373ip_vs_service_get(struct net *net, int af, __u32 fwmark, __u16 protocol,
Julius Volz3c2e0502008-09-02 15:55:38 +02001374 const union nf_inet_addr *vaddr, __be16 vport);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
1376static inline void ip_vs_service_put(struct ip_vs_service *svc)
1377{
1378 atomic_dec(&svc->usecnt);
1379}
1380
1381extern struct ip_vs_dest *
Hans Schillstromfc723252011-01-03 14:44:43 +01001382ip_vs_lookup_real_service(struct net *net, int af, __u16 protocol,
Julius Volz7937df12008-09-02 15:55:48 +02001383 const union nf_inet_addr *daddr, __be16 dport);
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385extern int ip_vs_use_count_inc(void);
1386extern void ip_vs_use_count_dec(void);
Hans Schillstrom8537de82012-04-26 07:47:44 +02001387extern int ip_vs_register_nl_ioctl(void);
1388extern void ip_vs_unregister_nl_ioctl(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389extern int ip_vs_control_init(void);
1390extern void ip_vs_control_cleanup(void);
Rumen G. Bogdanovski1e356f92007-11-07 02:35:54 -08001391extern struct ip_vs_dest *
Hans Schillstromfc723252011-01-03 14:44:43 +01001392ip_vs_find_dest(struct net *net, int af, const union nf_inet_addr *daddr,
1393 __be16 dport, const union nf_inet_addr *vaddr, __be16 vport,
Julian Anastasov52793db2011-12-30 14:19:02 +09001394 __u16 protocol, __u32 fwmark, __u32 flags);
Rumen G. Bogdanovski1e356f92007-11-07 02:35:54 -08001395extern struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
1397
1398/*
1399 * IPVS sync daemon data and function prototypes
1400 * (from ip_vs_sync.c)
1401 */
Hans Schillstromf1313152011-01-03 14:44:55 +01001402extern int start_sync_thread(struct net *net, int state, char *mcast_ifn,
1403 __u8 syncid);
1404extern int stop_sync_thread(struct net *net, int state);
Julian Anastasov749c42b2012-04-24 23:46:40 +03001405extern void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
1407
1408/*
1409 * IPVS rate estimator prototypes (from ip_vs_est.c)
1410 */
Julian Anastasov6ef757f2011-03-14 01:44:28 +02001411extern void ip_vs_start_estimator(struct net *net, struct ip_vs_stats *stats);
1412extern void ip_vs_stop_estimator(struct net *net, struct ip_vs_stats *stats);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413extern void ip_vs_zero_estimator(struct ip_vs_stats *stats);
Julian Anastasovea9f22c2011-03-14 01:41:54 +02001414extern void ip_vs_read_estimator(struct ip_vs_stats_user *dst,
1415 struct ip_vs_stats *stats);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
1417/*
1418 * Various IPVS packet transmitters (from ip_vs_xmit.c)
1419 */
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +02001420extern int ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1421 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1422extern int ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1423 struct ip_vs_protocol *pp,
1424 struct ip_vs_iphdr *iph);
1425extern int ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1426 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1427extern int ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1428 struct ip_vs_protocol *pp,
1429 struct ip_vs_iphdr *iph);
1430extern int ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1431 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1432extern int ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1433 struct ip_vs_protocol *pp, int offset,
1434 unsigned int hooknum, struct ip_vs_iphdr *iph);
Julian Anastasov026ace02013-03-21 11:58:06 +02001435extern void ip_vs_dest_dst_rcu_free(struct rcu_head *head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
Julius Volzb3cdd2a2008-09-02 15:55:45 +02001437#ifdef CONFIG_IP_VS_IPV6
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +02001438extern int ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1439 struct ip_vs_protocol *pp,
1440 struct ip_vs_iphdr *iph);
1441extern int ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1442 struct ip_vs_protocol *pp,
1443 struct ip_vs_iphdr *iph);
1444extern int ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1445 struct ip_vs_protocol *pp,
1446 struct ip_vs_iphdr *iph);
1447extern int ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1448 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1449extern int ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1450 struct ip_vs_protocol *pp, int offset,
1451 unsigned int hooknum, struct ip_vs_iphdr *iph);
Julius Volzb3cdd2a2008-09-02 15:55:45 +02001452#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
Simon Horman3a1bbf12011-02-04 18:33:02 +09001454#ifdef CONFIG_SYSCTL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455/*
1456 * This is a simple mechanism to ignore packets when
1457 * we are loaded. Just set ip_vs_drop_rate to 'n' and
1458 * we start to drop 1/rate of the packets
1459 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Hans Schillstroma0840e22011-01-03 14:44:58 +01001461static inline int ip_vs_todrop(struct netns_ipvs *ipvs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462{
Hans Schillstroma0840e22011-01-03 14:44:58 +01001463 if (!ipvs->drop_rate)
1464 return 0;
1465 if (--ipvs->drop_counter > 0)
1466 return 0;
1467 ipvs->drop_counter = ipvs->drop_rate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return 1;
1469}
Simon Horman3a1bbf12011-02-04 18:33:02 +09001470#else
1471static inline int ip_vs_todrop(struct netns_ipvs *ipvs) { return 0; }
1472#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
1474/*
1475 * ip_vs_fwd_tag returns the forwarding tag of the connection
1476 */
1477#define IP_VS_FWD_METHOD(cp) (cp->flags & IP_VS_CONN_F_FWD_MASK)
1478
Adrian Bunk732db652005-09-01 17:40:26 -07001479static inline char ip_vs_fwd_tag(struct ip_vs_conn *cp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
1481 char fwd;
1482
1483 switch (IP_VS_FWD_METHOD(cp)) {
1484 case IP_VS_CONN_F_MASQ:
1485 fwd = 'M'; break;
1486 case IP_VS_CONN_F_LOCALNODE:
1487 fwd = 'L'; break;
1488 case IP_VS_CONN_F_TUNNEL:
1489 fwd = 'T'; break;
1490 case IP_VS_CONN_F_DROUTE:
1491 fwd = 'R'; break;
1492 case IP_VS_CONN_F_BYPASS:
1493 fwd = 'B'; break;
1494 default:
1495 fwd = '?'; break;
1496 }
1497 return fwd;
1498}
1499
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500extern void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
Julius Volzb3cdd2a2008-09-02 15:55:45 +02001501 struct ip_vs_conn *cp, int dir);
1502
1503#ifdef CONFIG_IP_VS_IPV6
1504extern void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
1505 struct ip_vs_conn *cp, int dir);
1506#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Al Virob1550f22006-11-14 21:37:50 -08001508extern __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
Al Virof9214b22006-11-16 02:41:18 -08001510static inline __wsum ip_vs_check_diff4(__be32 old, __be32 new, __wsum oldsum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511{
Al Virof9214b22006-11-16 02:41:18 -08001512 __be32 diff[2] = { ~old, new };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Joe Perches07f07572008-11-19 15:44:53 -08001514 return csum_partial(diff, sizeof(diff), oldsum);
Al Virof9214b22006-11-16 02:41:18 -08001515}
1516
Julius Volz0bbdd422008-09-02 15:55:42 +02001517#ifdef CONFIG_IP_VS_IPV6
1518static inline __wsum ip_vs_check_diff16(const __be32 *old, const __be32 *new,
1519 __wsum oldsum)
1520{
1521 __be32 diff[8] = { ~old[3], ~old[2], ~old[1], ~old[0],
1522 new[3], new[2], new[1], new[0] };
1523
Joe Perches07f07572008-11-19 15:44:53 -08001524 return csum_partial(diff, sizeof(diff), oldsum);
Julius Volz0bbdd422008-09-02 15:55:42 +02001525}
1526#endif
1527
Al Virof9214b22006-11-16 02:41:18 -08001528static inline __wsum ip_vs_check_diff2(__be16 old, __be16 new, __wsum oldsum)
1529{
1530 __be16 diff[2] = { ~old, new };
1531
Joe Perches07f07572008-11-19 15:44:53 -08001532 return csum_partial(diff, sizeof(diff), oldsum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533}
1534
Julian Anastasovcf356d62010-10-17 16:21:07 +03001535/*
1536 * Forget current conntrack (unconfirmed) and attach notrack entry
1537 */
1538static inline void ip_vs_notrack(struct sk_buff *skb)
1539{
1540#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1541 enum ip_conntrack_info ctinfo;
Jesper Juhl06b69392011-03-09 22:55:05 +01001542 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
Julian Anastasovcf356d62010-10-17 16:21:07 +03001543
1544 if (!ct || !nf_ct_is_untracked(ct)) {
Lin Ming9e33ce42012-07-07 18:26:10 +08001545 nf_conntrack_put(skb->nfct);
Julian Anastasovcf356d62010-10-17 16:21:07 +03001546 skb->nfct = &nf_ct_untracked_get()->ct_general;
1547 skb->nfctinfo = IP_CT_NEW;
1548 nf_conntrack_get(skb->nfct);
1549 }
1550#endif
1551}
1552
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001553#ifdef CONFIG_IP_VS_NFCT
1554/*
1555 * Netfilter connection tracking
1556 * (from ip_vs_nfct.c)
1557 */
Hans Schillstroma0840e22011-01-03 14:44:58 +01001558static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001559{
Simon Hormana4e2f5a2011-02-04 18:33:02 +09001560#ifdef CONFIG_SYSCTL
Hans Schillstroma0840e22011-01-03 14:44:58 +01001561 return ipvs->sysctl_conntrack;
Simon Hormana4e2f5a2011-02-04 18:33:02 +09001562#else
1563 return 0;
1564#endif
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001565}
1566
Julian Anastasov6523ce12010-09-05 18:02:29 +00001567extern void ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp,
1568 int outin);
Simon Horman3c2de2a2011-09-16 14:02:19 +09001569extern int ip_vs_confirm_conntrack(struct sk_buff *skb);
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001570extern void ip_vs_nfct_expect_related(struct sk_buff *skb, struct nf_conn *ct,
1571 struct ip_vs_conn *cp, u_int8_t proto,
1572 const __be16 port, int from_rs);
1573extern void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp);
1574
1575#else
1576
Hans Schillstroma0840e22011-01-03 14:44:58 +01001577static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001578{
1579 return 0;
1580}
1581
1582static inline void ip_vs_update_conntrack(struct sk_buff *skb,
1583 struct ip_vs_conn *cp, int outin)
1584{
1585}
1586
Krzysztof Wilczynskie23ebf02011-10-20 13:18:04 +01001587static inline int ip_vs_confirm_conntrack(struct sk_buff *skb)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001588{
1589 return NF_ACCEPT;
1590}
1591
1592static inline void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp)
1593{
1594}
1595/* CONFIG_IP_VS_NFCT */
1596#endif
Julian Anastasov6523ce12010-09-05 18:02:29 +00001597
Changli Gaob552f7e2011-02-19 17:32:28 +08001598static inline unsigned int
1599ip_vs_dest_conn_overhead(struct ip_vs_dest *dest)
1600{
1601 /*
1602 * We think the overhead of processing active connections is 256
1603 * times higher than that of inactive connections in average. (This
1604 * 256 times might not be accurate, we will change it later) We
1605 * use the following formula to estimate the overhead now:
1606 * dest->activeconns*256 + dest->inactconns
1607 */
1608 return (atomic_read(&dest->activeconns) << 8) +
1609 atomic_read(&dest->inactconns);
1610}
1611
Julius Volzbc4768e2008-07-31 20:45:24 -07001612#endif /* _NET_IP_VS_H */