blob: cd6018a9ee2467cec1fbe4d621fc294bc843a05e [file] [log] [blame]
Simon Horman07dcc682014-09-30 10:50:06 +09001/* IP Virtual Server
2 * data structure and functionality definitions
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 */
4
Julius Volzbc4768e2008-07-31 20:45:24 -07005#ifndef _NET_IP_VS_H
6#define _NET_IP_VS_H
Linus Torvalds1da177e2005-04-16 15:20:36 -07007
Julius Volzbc4768e2008-07-31 20:45:24 -07008#include <linux/ip_vs.h> /* definitions shared with userland */
Linus Torvalds1da177e2005-04-16 15:20:36 -07009
Julius Volzbc4768e2008-07-31 20:45:24 -070010#include <asm/types.h> /* for __uXX types */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/list.h> /* for struct list_head */
13#include <linux/spinlock.h> /* for struct rwlock_t */
Simon Horman07dcc682014-09-30 10:50:06 +090014#include <linux/atomic.h> /* for struct atomic_t */
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/compiler.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020016#include <linux/timer.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050017#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020019#include <net/checksum.h>
Julius Volze7ade462008-09-02 15:55:33 +020020#include <linux/netfilter.h> /* for union nf_inet_addr */
KOVACS Krisztian1668e012008-10-01 07:33:10 -070021#include <linux/ip.h>
Julius Volze7ade462008-09-02 15:55:33 +020022#include <linux/ipv6.h> /* for struct ipv6hdr */
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +000023#include <net/ipv6.h>
Jesper Dangaard Brouer54d83ef2012-10-22 21:22:45 +020024#if IS_ENABLED(CONFIG_IP_VS_IPV6)
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +020025#include <linux/netfilter_ipv6/ip6_tables.h>
26#endif
Jesper Dangaard Brouera638e512012-09-26 14:06:20 +020027#if IS_ENABLED(CONFIG_NF_CONNTRACK)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +020028#include <net/netfilter/nf_conntrack.h>
29#endif
Hans Schillstrom61b1ab42011-01-03 14:44:42 +010030#include <net/net_namespace.h> /* Netw namespace */
31
Alex Gartrell4fd9bee2015-08-26 09:40:29 -070032#define IP_VS_HDR_INVERSE 1
33#define IP_VS_HDR_ICMP 2
34
Simon Horman07dcc682014-09-30 10:50:06 +090035/* Generic access of ipvs struct */
Hans Schillstrom61b1ab42011-01-03 14:44:42 +010036static inline struct netns_ipvs *net_ipvs(struct net* net)
37{
38 return net->ipvs;
39}
Simon Horman07dcc682014-09-30 10:50:06 +090040
Simon Horman07dcc682014-09-30 10:50:06 +090041/* This one needed for single_open_net since net is stored directly in
Lucas De Marchi25985ed2011-03-30 22:57:33 -030042 * private not as a struct i.e. seq_file_net can't be used.
Hans Schillstromb17fc992011-01-03 14:44:56 +010043 */
44static inline struct net *seq_file_single_net(struct seq_file *seq)
45{
46#ifdef CONFIG_NET_NS
47 return (struct net *)seq->private;
48#else
49 return &init_net;
50#endif
51}
Catalin(ux) M. BOIE6f7edb42010-01-05 05:50:24 +010052
53/* Connections' size value needed by ip_vs_ctl.c */
54extern int ip_vs_conn_tab_size;
55
Julius Volz64aae3c2008-09-02 15:55:34 +020056struct ip_vs_iphdr {
Alex Gartrell4fd9bee2015-08-26 09:40:29 -070057 int hdr_flags; /* ipvs flags */
Alex Gartrellb0e010c2015-08-26 09:40:28 -070058 __u32 off; /* Where IP or IPv4 header starts */
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +020059 __u32 len; /* IPv4 simply where L4 starts
Simon Horman07dcc682014-09-30 10:50:06 +090060 * IPv6 where L4 Transport Header starts */
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +020061 __u16 fragoffs; /* IPv6 fragment offset, 0 if first frag (or not frag)*/
62 __s16 protocol;
63 __s32 flags;
Julius Volz64aae3c2008-09-02 15:55:34 +020064 union nf_inet_addr saddr;
65 union nf_inet_addr daddr;
66};
67
Jesper Dangaard Brouer2f747132012-09-26 14:06:59 +020068static inline void *frag_safe_skb_hp(const struct sk_buff *skb, int offset,
69 int len, void *buffer,
70 const struct ip_vs_iphdr *ipvsh)
71{
72 return skb_header_pointer(skb, offset, len, buffer);
73}
Jesper Dangaard Brouer2f747132012-09-26 14:06:59 +020074
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +020075/* This function handles filling *ip_vs_iphdr, both for IPv4 and IPv6.
76 * IPv6 requires some extra work, as finding proper header position,
77 * depend on the IPv6 extension headers.
78 */
Alex Gartrellb0e010c2015-08-26 09:40:28 -070079static inline int
80ip_vs_fill_iph_skb_off(int af, const struct sk_buff *skb, int offset,
Alex Gartrell4fd9bee2015-08-26 09:40:29 -070081 int hdr_flags, struct ip_vs_iphdr *iphdr)
Julius Volz64aae3c2008-09-02 15:55:34 +020082{
Alex Gartrell4fd9bee2015-08-26 09:40:29 -070083 iphdr->hdr_flags = hdr_flags;
Alex Gartrellb0e010c2015-08-26 09:40:28 -070084 iphdr->off = offset;
Alex Gartrell4fd9bee2015-08-26 09:40:29 -070085
Julius Volz64aae3c2008-09-02 15:55:34 +020086#ifdef CONFIG_IP_VS_IPV6
87 if (af == AF_INET6) {
Alex Gartrellb0e010c2015-08-26 09:40:28 -070088 struct ipv6hdr _iph;
89 const struct ipv6hdr *iph = skb_header_pointer(
90 skb, offset, sizeof(_iph), &_iph);
91 if (!iph)
92 return 0;
93
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +000094 iphdr->saddr.in6 = iph->saddr;
95 iphdr->daddr.in6 = iph->daddr;
Jiri Pirko6aafeef2013-11-06 17:52:20 +010096 /* ipv6_find_hdr() updates len, flags */
Alex Gartrellb0e010c2015-08-26 09:40:28 -070097 iphdr->len = offset;
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +020098 iphdr->flags = 0;
99 iphdr->protocol = ipv6_find_hdr(skb, &iphdr->len, -1,
100 &iphdr->fragoffs,
101 &iphdr->flags);
Alex Gartrellb0e010c2015-08-26 09:40:28 -0700102 if (iphdr->protocol < 0)
103 return 0;
Julius Volz64aae3c2008-09-02 15:55:34 +0200104 } else
105#endif
106 {
Alex Gartrellb0e010c2015-08-26 09:40:28 -0700107 struct iphdr _iph;
108 const struct iphdr *iph = skb_header_pointer(
109 skb, offset, sizeof(_iph), &_iph);
110 if (!iph)
111 return 0;
112
113 iphdr->len = offset + iph->ihl * 4;
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200114 iphdr->fragoffs = 0;
115 iphdr->protocol = iph->protocol;
116 iphdr->saddr.ip = iph->saddr;
117 iphdr->daddr.ip = iph->daddr;
118 }
Alex Gartrellb0e010c2015-08-26 09:40:28 -0700119
120 return 1;
121}
122
123static inline int
Alex Gartrell4fd9bee2015-08-26 09:40:29 -0700124ip_vs_fill_iph_skb_icmp(int af, const struct sk_buff *skb, int offset,
125 bool inverse, struct ip_vs_iphdr *iphdr)
Alex Gartrellb0e010c2015-08-26 09:40:28 -0700126{
Alex Gartrell4fd9bee2015-08-26 09:40:29 -0700127 int hdr_flags = IP_VS_HDR_ICMP;
128
129 if (inverse)
130 hdr_flags |= IP_VS_HDR_INVERSE;
131
132 return ip_vs_fill_iph_skb_off(af, skb, offset, hdr_flags, iphdr);
133}
134
135static inline int
136ip_vs_fill_iph_skb(int af, const struct sk_buff *skb, bool inverse,
137 struct ip_vs_iphdr *iphdr)
138{
139 int hdr_flags = 0;
140
141 if (inverse)
142 hdr_flags |= IP_VS_HDR_INVERSE;
143
144 return ip_vs_fill_iph_skb_off(af, skb, skb_network_offset(skb),
145 hdr_flags, iphdr);
146}
147
148static inline bool
149ip_vs_iph_inverse(const struct ip_vs_iphdr *iph)
150{
151 return !!(iph->hdr_flags & IP_VS_HDR_INVERSE);
152}
153
154static inline bool
155ip_vs_iph_icmp(const struct ip_vs_iphdr *iph)
156{
157 return !!(iph->hdr_flags & IP_VS_HDR_ICMP);
Jesper Dangaard Brouer63dca2c2012-09-26 14:06:41 +0200158}
159
Julius Volz64aae3c2008-09-02 15:55:34 +0200160static inline void ip_vs_addr_copy(int af, union nf_inet_addr *dst,
161 const union nf_inet_addr *src)
162{
163#ifdef CONFIG_IP_VS_IPV6
164 if (af == AF_INET6)
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000165 dst->in6 = src->in6;
Julius Volz64aae3c2008-09-02 15:55:34 +0200166 else
167#endif
168 dst->ip = src->ip;
169}
170
Julian Anastasov9a05475c2013-03-21 11:58:12 +0200171static inline void ip_vs_addr_set(int af, union nf_inet_addr *dst,
172 const union nf_inet_addr *src)
173{
174#ifdef CONFIG_IP_VS_IPV6
175 if (af == AF_INET6) {
176 dst->in6 = src->in6;
177 return;
178 }
179#endif
180 dst->ip = src->ip;
181 dst->all[1] = 0;
182 dst->all[2] = 0;
183 dst->all[3] = 0;
184}
185
Julius Volz64aae3c2008-09-02 15:55:34 +0200186static inline int ip_vs_addr_equal(int af, const union nf_inet_addr *a,
187 const union nf_inet_addr *b)
188{
189#ifdef CONFIG_IP_VS_IPV6
190 if (af == AF_INET6)
191 return ipv6_addr_equal(&a->in6, &b->in6);
192#endif
193 return a->ip == b->ip;
194}
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196#ifdef CONFIG_IP_VS_DEBUG
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200197#include <linux/net.h>
198
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700199int ip_vs_get_debug_level(void);
Julius Volzc842a3a2008-09-02 15:55:35 +0200200
201static inline const char *ip_vs_dbg_addr(int af, char *buf, size_t buf_len,
202 const union nf_inet_addr *addr,
203 int *idx)
204{
205 int len;
206#ifdef CONFIG_IP_VS_IPV6
207 if (af == AF_INET6)
Jesper Dangaard Brouer120b9c12012-09-26 14:05:53 +0200208 len = snprintf(&buf[*idx], buf_len - *idx, "[%pI6c]",
Harvey Harrison0c6ce782008-10-28 16:09:23 -0700209 &addr->in6) + 1;
Julius Volzc842a3a2008-09-02 15:55:35 +0200210 else
211#endif
Harvey Harrison3685f25d2008-10-31 00:56:49 -0700212 len = snprintf(&buf[*idx], buf_len - *idx, "%pI4",
213 &addr->ip) + 1;
Julius Volzc842a3a2008-09-02 15:55:35 +0200214
215 *idx += len;
216 BUG_ON(*idx > buf_len + 1);
217 return &buf[*idx - len];
218}
219
Hannes Eder9aada7a2009-07-30 14:29:44 -0700220#define IP_VS_DBG_BUF(level, msg, ...) \
221 do { \
222 char ip_vs_dbg_buf[160]; \
223 int ip_vs_dbg_idx = 0; \
224 if (level <= ip_vs_get_debug_level()) \
225 printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \
226 } while (0)
227#define IP_VS_ERR_BUF(msg...) \
228 do { \
229 char ip_vs_dbg_buf[160]; \
230 int ip_vs_dbg_idx = 0; \
231 pr_err(msg); \
232 } while (0)
Julius Volzc842a3a2008-09-02 15:55:35 +0200233
234/* Only use from within IP_VS_DBG_BUF() or IP_VS_ERR_BUF macros */
Hannes Eder9aada7a2009-07-30 14:29:44 -0700235#define IP_VS_DBG_ADDR(af, addr) \
236 ip_vs_dbg_addr(af, ip_vs_dbg_buf, \
237 sizeof(ip_vs_dbg_buf), addr, \
238 &ip_vs_dbg_idx)
Julius Volzc842a3a2008-09-02 15:55:35 +0200239
Hannes Eder9aada7a2009-07-30 14:29:44 -0700240#define IP_VS_DBG(level, msg, ...) \
241 do { \
242 if (level <= ip_vs_get_debug_level()) \
243 printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \
244 } while (0)
245#define IP_VS_DBG_RL(msg, ...) \
246 do { \
247 if (net_ratelimit()) \
248 printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \
249 } while (0)
Julian Anastasov0d796412010-10-17 16:46:17 +0300250#define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700251 do { \
252 if (level <= ip_vs_get_debug_level()) \
Julian Anastasov0d796412010-10-17 16:46:17 +0300253 pp->debug_packet(af, pp, skb, ofs, msg); \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700254 } while (0)
Julian Anastasov0d796412010-10-17 16:46:17 +0300255#define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700256 do { \
257 if (level <= ip_vs_get_debug_level() && \
258 net_ratelimit()) \
Julian Anastasov0d796412010-10-17 16:46:17 +0300259 pp->debug_packet(af, pp, skb, ofs, msg); \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700260 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261#else /* NO DEBUGGING at ALL */
Julius Volzc842a3a2008-09-02 15:55:35 +0200262#define IP_VS_DBG_BUF(level, msg...) do {} while (0)
263#define IP_VS_ERR_BUF(msg...) do {} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264#define IP_VS_DBG(level, msg...) do {} while (0)
265#define IP_VS_DBG_RL(msg...) do {} while (0)
Julian Anastasov0d796412010-10-17 16:46:17 +0300266#define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) do {} while (0)
267#define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) do {} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268#endif
269
270#define IP_VS_BUG() BUG()
Hannes Eder1e3e2382009-08-02 11:05:41 +0000271#define IP_VS_ERR_RL(msg, ...) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700272 do { \
273 if (net_ratelimit()) \
Hannes Eder1e3e2382009-08-02 11:05:41 +0000274 pr_err(msg, ##__VA_ARGS__); \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700275 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277#ifdef CONFIG_IP_VS_DEBUG
278#define EnterFunction(level) \
Hannes Eder9aada7a2009-07-30 14:29:44 -0700279 do { \
280 if (level <= ip_vs_get_debug_level()) \
281 printk(KERN_DEBUG \
282 pr_fmt("Enter: %s, %s line %i\n"), \
283 __func__, __FILE__, __LINE__); \
284 } while (0)
285#define LeaveFunction(level) \
286 do { \
287 if (level <= ip_vs_get_debug_level()) \
288 printk(KERN_DEBUG \
289 pr_fmt("Leave: %s, %s line %i\n"), \
290 __func__, __FILE__, __LINE__); \
291 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292#else
293#define EnterFunction(level) do {} while (0)
294#define LeaveFunction(level) do {} while (0)
295#endif
296
Simon Horman07dcc682014-09-30 10:50:06 +0900297/* The port number of FTP service (in network order). */
Harvey Harrisonf3a7c662009-02-14 22:58:35 -0800298#define FTPPORT cpu_to_be16(21)
299#define FTPDATA cpu_to_be16(20)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Simon Horman07dcc682014-09-30 10:50:06 +0900301/* TCP State Values */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302enum {
303 IP_VS_TCP_S_NONE = 0,
304 IP_VS_TCP_S_ESTABLISHED,
305 IP_VS_TCP_S_SYN_SENT,
306 IP_VS_TCP_S_SYN_RECV,
307 IP_VS_TCP_S_FIN_WAIT,
308 IP_VS_TCP_S_TIME_WAIT,
309 IP_VS_TCP_S_CLOSE,
310 IP_VS_TCP_S_CLOSE_WAIT,
311 IP_VS_TCP_S_LAST_ACK,
312 IP_VS_TCP_S_LISTEN,
313 IP_VS_TCP_S_SYNACK,
314 IP_VS_TCP_S_LAST
315};
316
Simon Horman07dcc682014-09-30 10:50:06 +0900317/* UDP State Values */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318enum {
319 IP_VS_UDP_S_NORMAL,
320 IP_VS_UDP_S_LAST,
321};
322
Simon Horman07dcc682014-09-30 10:50:06 +0900323/* ICMP State Values */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324enum {
325 IP_VS_ICMP_S_NORMAL,
326 IP_VS_ICMP_S_LAST,
327};
328
Simon Horman07dcc682014-09-30 10:50:06 +0900329/* SCTP State Values */
Venkata Mohan Reddy2906f662010-02-18 12:31:05 +0100330enum ip_vs_sctp_states {
331 IP_VS_SCTP_S_NONE,
Julian Anastasov61e7c422013-06-18 10:08:07 +0300332 IP_VS_SCTP_S_INIT1,
333 IP_VS_SCTP_S_INIT,
334 IP_VS_SCTP_S_COOKIE_SENT,
335 IP_VS_SCTP_S_COOKIE_REPLIED,
336 IP_VS_SCTP_S_COOKIE_WAIT,
337 IP_VS_SCTP_S_COOKIE,
338 IP_VS_SCTP_S_COOKIE_ECHOED,
Venkata Mohan Reddy2906f662010-02-18 12:31:05 +0100339 IP_VS_SCTP_S_ESTABLISHED,
Julian Anastasov61e7c422013-06-18 10:08:07 +0300340 IP_VS_SCTP_S_SHUTDOWN_SENT,
341 IP_VS_SCTP_S_SHUTDOWN_RECEIVED,
342 IP_VS_SCTP_S_SHUTDOWN_ACK_SENT,
343 IP_VS_SCTP_S_REJECTED,
Venkata Mohan Reddy2906f662010-02-18 12:31:05 +0100344 IP_VS_SCTP_S_CLOSED,
345 IP_VS_SCTP_S_LAST
346};
347
Simon Horman07dcc682014-09-30 10:50:06 +0900348/* Delta sequence info structure
349 * Each ip_vs_conn has 2 (output AND input seq. changes).
350 * Only used in the VS/NAT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 */
352struct ip_vs_seq {
353 __u32 init_seq; /* Add delta from this seq */
354 __u32 delta; /* Delta in sequence numbers */
355 __u32 previous_delta; /* Delta in sequence numbers
Simon Horman07dcc682014-09-30 10:50:06 +0900356 * before last resized pkt */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357};
358
Simon Horman07dcc682014-09-30 10:50:06 +0900359/* counters per cpu */
Hans Schillstromb17fc992011-01-03 14:44:56 +0100360struct ip_vs_counters {
Julian Anastasovcd67cd52015-02-06 09:44:44 +0200361 __u64 conns; /* connections scheduled */
362 __u64 inpkts; /* incoming packets */
363 __u64 outpkts; /* outgoing packets */
Hans Schillstromb17fc992011-01-03 14:44:56 +0100364 __u64 inbytes; /* incoming bytes */
365 __u64 outbytes; /* outgoing bytes */
366};
Simon Horman07dcc682014-09-30 10:50:06 +0900367/* Stats per cpu */
Hans Schillstromb17fc992011-01-03 14:44:56 +0100368struct ip_vs_cpu_stats {
Julian Anastasovcd67cd52015-02-06 09:44:44 +0200369 struct ip_vs_counters cnt;
Hans Schillstromb17fc992011-01-03 14:44:56 +0100370 struct u64_stats_sync syncp;
371};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Simon Horman07dcc682014-09-30 10:50:06 +0900373/* IPVS statistics objects */
Sven Wegener3a14a3132008-08-10 18:24:41 +0000374struct ip_vs_estimator {
375 struct list_head list;
376
377 u64 last_inbytes;
378 u64 last_outbytes;
Julian Anastasovcd67cd52015-02-06 09:44:44 +0200379 u64 last_conns;
380 u64 last_inpkts;
381 u64 last_outpkts;
Sven Wegener3a14a3132008-08-10 18:24:41 +0000382
Julian Anastasovcd67cd52015-02-06 09:44:44 +0200383 u64 cps;
384 u64 inpps;
385 u64 outpps;
386 u64 inbps;
387 u64 outbps;
388};
389
390/*
391 * IPVS statistics object, 64-bit kernel version of struct ip_vs_stats_user
392 */
393struct ip_vs_kstats {
394 u64 conns; /* connections scheduled */
395 u64 inpkts; /* incoming packets */
396 u64 outpkts; /* outgoing packets */
397 u64 inbytes; /* incoming bytes */
398 u64 outbytes; /* outgoing bytes */
399
400 u64 cps; /* current connection rate */
401 u64 inpps; /* current in packet rate */
402 u64 outpps; /* current out packet rate */
403 u64 inbps; /* current in byte rate */
404 u64 outbps; /* current out byte rate */
Sven Wegener3a14a3132008-08-10 18:24:41 +0000405};
406
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000407struct ip_vs_stats {
Julian Anastasovcd67cd52015-02-06 09:44:44 +0200408 struct ip_vs_kstats kstats; /* kernel statistics */
Sven Wegenere9c0ce22008-09-08 13:39:04 +0200409 struct ip_vs_estimator est; /* estimator */
Julian Anastasovb962abd2013-03-09 23:25:08 +0200410 struct ip_vs_cpu_stats __percpu *cpustats; /* per cpu counters */
Hans Schillstromb17fc992011-01-03 14:44:56 +0100411 spinlock_t lock; /* spin lock */
Julian Anastasovcd67cd52015-02-06 09:44:44 +0200412 struct ip_vs_kstats kstats0; /* reset values */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413};
414
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200415struct dst_entry;
416struct iphdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417struct ip_vs_conn;
418struct ip_vs_app;
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -0200419struct sk_buff;
Hans Schillstrom252c6412011-01-03 14:44:46 +0100420struct ip_vs_proto_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422struct ip_vs_protocol {
423 struct ip_vs_protocol *next;
424 char *name;
Julian Anastasov2ad17de2008-04-29 03:21:23 -0700425 u16 protocol;
426 u16 num_states;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 int dont_defrag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 void (*init)(struct ip_vs_protocol *pp);
430
431 void (*exit)(struct ip_vs_protocol *pp);
432
Eric W. Biederman1281a9c22015-09-21 13:02:36 -0500433 int (*init_netns)(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd);
Hans Schillstrom252c6412011-01-03 14:44:46 +0100434
Eric W. Biederman1281a9c22015-09-21 13:02:36 -0500435 void (*exit_netns)(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd);
Hans Schillstrom252c6412011-01-03 14:44:46 +0100436
Eric W. Biedermand8f44c32015-09-21 13:02:43 -0500437 int (*conn_schedule)(struct netns_ipvs *ipvs,
438 int af, struct sk_buff *skb,
Hans Schillstrom93304192011-01-03 14:44:51 +0100439 struct ip_vs_proto_data *pd,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200440 int *verdict, struct ip_vs_conn **cpp,
441 struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
443 struct ip_vs_conn *
Eric W. Biedermanab161972015-09-21 13:02:38 -0500444 (*conn_in_get)(struct netns_ipvs *ipvs,
445 int af,
Julius Volz51ef3482008-09-02 15:55:40 +0200446 const struct sk_buff *skb,
Alex Gartrell802c41a2015-08-26 09:40:32 -0700447 const struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449 struct ip_vs_conn *
Eric W. Biederman0cf705c8c2015-09-21 13:02:39 -0500450 (*conn_out_get)(struct netns_ipvs *ipvs,
451 int af,
Julius Volz51ef3482008-09-02 15:55:40 +0200452 const struct sk_buff *skb,
Alex Gartrell802c41a2015-08-26 09:40:32 -0700453 const struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200455 int (*snat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp,
456 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200458 int (*dnat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp,
459 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Julius Volz51ef3482008-09-02 15:55:40 +0200461 int (*csum_check)(int af, struct sk_buff *skb,
462 struct ip_vs_protocol *pp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 const char *(*state_name)(int state);
465
Simon Horman4a516f12011-09-16 14:11:49 +0900466 void (*state_transition)(struct ip_vs_conn *cp, int direction,
467 const struct sk_buff *skb,
468 struct ip_vs_proto_data *pd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Eric W. Biederman19648912015-09-21 13:02:29 -0500470 int (*register_app)(struct netns_ipvs *ipvs, struct ip_vs_app *inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Eric W. Biederman19648912015-09-21 13:02:29 -0500472 void (*unregister_app)(struct netns_ipvs *ipvs, struct ip_vs_app *inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 int (*app_conn_bind)(struct ip_vs_conn *cp);
475
Julian Anastasov0d796412010-10-17 16:46:17 +0300476 void (*debug_packet)(int af, struct ip_vs_protocol *pp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 const struct sk_buff *skb,
478 int offset,
479 const char *msg);
480
Hans Schillstrom93304192011-01-03 14:44:51 +0100481 void (*timeout_change)(struct ip_vs_proto_data *pd, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482};
483
Simon Horman07dcc682014-09-30 10:50:06 +0900484/* protocol data per netns */
Hans Schillstrom252c6412011-01-03 14:44:46 +0100485struct ip_vs_proto_data {
486 struct ip_vs_proto_data *next;
487 struct ip_vs_protocol *pp;
488 int *timeout_table; /* protocol timeout table */
489 atomic_t appcnt; /* counter of proto app incs. */
490 struct tcp_states_t *tcp_state_table;
491};
492
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700493struct ip_vs_protocol *ip_vs_proto_get(unsigned short proto);
Eric W. Biederman18d6ade2015-09-21 13:02:01 -0500494struct ip_vs_proto_data *ip_vs_proto_data_get(struct netns_ipvs *ipvs,
Joe Perches5c3a0fd2013-09-21 10:22:42 -0700495 unsigned short proto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Simon Hormanf11017e2010-08-22 21:37:52 +0900497struct ip_vs_conn_param {
Eric W. Biedermane64e2b42015-09-21 13:01:42 -0500498 struct netns_ipvs *ipvs;
Simon Hormanf11017e2010-08-22 21:37:52 +0900499 const union nf_inet_addr *caddr;
500 const union nf_inet_addr *vaddr;
501 __be16 cport;
502 __be16 vport;
503 __u16 protocol;
504 u16 af;
Simon Horman85999282010-08-22 21:37:53 +0900505
506 const struct ip_vs_pe *pe;
507 char *pe_data;
508 __u8 pe_data_len;
Simon Hormanf11017e2010-08-22 21:37:52 +0900509};
510
Simon Horman07dcc682014-09-30 10:50:06 +0900511/* IP_VS structure allocated for each dynamically scheduled connection */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512struct ip_vs_conn {
Changli Gao731109e2011-02-19 18:05:08 +0800513 struct hlist_node c_list; /* hashed list heads */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 /* Protocol, addresses and port numbers */
Hans Schillstrom6e67e582011-01-03 14:44:57 +0100515 __be16 cport;
Hans Schillstrom6e67e582011-01-03 14:44:57 +0100516 __be16 dport;
Julian Anastasov1845ed02013-03-21 11:58:11 +0200517 __be16 vport;
518 u16 af; /* address family */
Hans Schillstrom6e67e582011-01-03 14:44:57 +0100519 union nf_inet_addr caddr; /* client address */
520 union nf_inet_addr vaddr; /* virtual address */
521 union nf_inet_addr daddr; /* destination address */
522 volatile __u32 flags; /* status flags */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 __u16 protocol; /* Which protocol (TCP/UDP) */
Alex Gartrellba385282014-09-09 16:40:23 -0700524 __u16 daf; /* Address family of the dest */
Eric W. Biederman58dbc6f2015-09-21 13:01:41 -0500525 struct netns_ipvs *ipvs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
527 /* counter and timer */
528 atomic_t refcnt; /* reference count */
529 struct timer_list timer; /* Expiration timer */
530 volatile unsigned long timeout; /* timeout */
531
532 /* Flags and state transition */
533 spinlock_t lock; /* lock for state transition */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 volatile __u16 state; /* state info */
Rumen G. Bogdanovskiefac5272007-11-07 02:36:55 -0800535 volatile __u16 old_state; /* old state, to be used for
536 * state transition triggerd
537 * synchronization
538 */
Julian Anastasov1845ed02013-03-21 11:58:11 +0200539 __u32 fwmark; /* Fire wall mark from skb */
Julian Anastasov749c42b2012-04-24 23:46:40 +0300540 unsigned long sync_endtime; /* jiffies + sent_retries */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 /* Control members */
543 struct ip_vs_conn *control; /* Master control connection */
544 atomic_t n_control; /* Number of controlled ones */
545 struct ip_vs_dest *dest; /* real server */
546 atomic_t in_pkts; /* incoming packet counter */
547
Simon Horman07dcc682014-09-30 10:50:06 +0900548 /* Packet transmitter for different forwarding methods. If it
549 * mangles the packet, it must return NF_DROP or better NF_STOLEN,
550 * otherwise this must be changed to a sk_buff **.
551 * NF_ACCEPT can be returned when destination is local.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 */
553 int (*packet_xmit)(struct sk_buff *skb, struct ip_vs_conn *cp,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200554 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
556 /* Note: we can group the following members into a structure,
Simon Horman07dcc682014-09-30 10:50:06 +0900557 * in order to save more space, and the following members are
558 * only used in VS/NAT anyway
559 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 struct ip_vs_app *app; /* bound ip_vs_app object */
561 void *app_data; /* Application private data */
562 struct ip_vs_seq in_seq; /* incoming seq. struct */
563 struct ip_vs_seq out_seq; /* outgoing seq. struct */
Simon Horman85999282010-08-22 21:37:53 +0900564
Simon Hormane9e5eee2010-11-08 20:05:57 +0900565 const struct ip_vs_pe *pe;
Simon Horman85999282010-08-22 21:37:53 +0900566 char *pe_data;
567 __u8 pe_data_len;
Julian Anastasov088339a2013-03-21 11:58:10 +0200568
569 struct rcu_head rcu_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570};
571
Simon Horman07dcc682014-09-30 10:50:06 +0900572/* Extended internal versions of struct ip_vs_service_user and ip_vs_dest_user
573 * for IPv6 support.
Julius Volzc860c6b2008-09-02 15:55:36 +0200574 *
Simon Horman07dcc682014-09-30 10:50:06 +0900575 * We need these to conveniently pass around service and destination
576 * options, but unfortunately, we also need to keep the old definitions to
577 * maintain userspace backwards compatibility for the setsockopt interface.
Julius Volzc860c6b2008-09-02 15:55:36 +0200578 */
579struct ip_vs_service_user_kern {
580 /* virtual service addresses */
581 u16 af;
582 u16 protocol;
583 union nf_inet_addr addr; /* virtual ip address */
Julian Anastasov0a925862013-04-17 23:50:49 +0300584 __be16 port;
Julius Volzc860c6b2008-09-02 15:55:36 +0200585 u32 fwmark; /* firwall mark of service */
586
587 /* virtual service options */
588 char *sched_name;
Simon Horman0d1e71b2010-08-22 21:37:54 +0900589 char *pe_name;
Eric Dumazet95c96172012-04-15 05:58:06 +0000590 unsigned int flags; /* virtual service flags */
591 unsigned int timeout; /* persistent timeout in sec */
Julian Anastasov0a925862013-04-17 23:50:49 +0300592 __be32 netmask; /* persistent netmask or plen */
Julius Volzc860c6b2008-09-02 15:55:36 +0200593};
594
595
596struct ip_vs_dest_user_kern {
597 /* destination server address */
598 union nf_inet_addr addr;
Julian Anastasov0a925862013-04-17 23:50:49 +0300599 __be16 port;
Julius Volzc860c6b2008-09-02 15:55:36 +0200600
601 /* real server options */
Eric Dumazet95c96172012-04-15 05:58:06 +0000602 unsigned int conn_flags; /* connection flags */
Julius Volzc860c6b2008-09-02 15:55:36 +0200603 int weight; /* destination weight */
604
605 /* thresholds for active connections */
606 u32 u_threshold; /* upper threshold */
607 u32 l_threshold; /* lower threshold */
Alex Gartrell6cff3392014-09-09 16:40:20 -0700608
609 /* Address family of addr */
610 u16 af;
Julius Volzc860c6b2008-09-02 15:55:36 +0200611};
612
613
614/*
Simon Horman07dcc682014-09-30 10:50:06 +0900615 * The information about the virtual service offered to the net and the
616 * forwarding entries.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 */
618struct ip_vs_service {
Julian Anastasovceec4c32013-03-22 11:46:53 +0200619 struct hlist_node s_list; /* for normal service table */
620 struct hlist_node f_list; /* for fwmark-based service table */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 atomic_t refcnt; /* reference counter */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Julius Volze7ade462008-09-02 15:55:33 +0200623 u16 af; /* address family */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 __u16 protocol; /* which protocol (TCP/UDP) */
Julius Volze7ade462008-09-02 15:55:33 +0200625 union nf_inet_addr addr; /* IP address for virtual service */
Al Viro014d7302006-09-28 14:29:52 -0700626 __be16 port; /* port number for the service */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 __u32 fwmark; /* firewall mark of the service */
Eric Dumazet95c96172012-04-15 05:58:06 +0000628 unsigned int flags; /* service status flags */
629 unsigned int timeout; /* persistent timeout in ticks */
Julian Anastasov0a925862013-04-17 23:50:49 +0300630 __be32 netmask; /* grouping granularity, mask/plen */
Eric W. Biederman3109d2f2015-09-21 13:01:44 -0500631 struct netns_ipvs *ipvs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 struct list_head destinations; /* real server d-linked list */
634 __u32 num_dests; /* number of servers */
635 struct ip_vs_stats stats; /* statistics for the service */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
637 /* for scheduling */
Julian Anastasovceec4c32013-03-22 11:46:53 +0200638 struct ip_vs_scheduler __rcu *scheduler; /* bound scheduler object */
Julian Anastasovba3a3ce2013-03-22 11:46:51 +0200639 spinlock_t sched_lock; /* lock sched_data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 void *sched_data; /* scheduler application data */
Simon Horman85999282010-08-22 21:37:53 +0900641
642 /* alternate persistence engine */
Julian Anastasovceec4c32013-03-22 11:46:53 +0200643 struct ip_vs_pe __rcu *pe;
644
645 struct rcu_head rcu_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646};
647
Julian Anastasov026ace02013-03-21 11:58:06 +0200648/* Information for cached dst */
649struct ip_vs_dest_dst {
650 struct dst_entry *dst_cache; /* destination cache entry */
651 u32 dst_cookie;
652 union nf_inet_addr dst_saddr;
653 struct rcu_head rcu_head;
654};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Simon Horman07dcc682014-09-30 10:50:06 +0900656/* The real server destination forwarding entry with ip address, port number,
657 * and so on.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 */
659struct ip_vs_dest {
660 struct list_head n_list; /* for the dests in the service */
Julian Anastasov276472e2013-03-21 11:58:08 +0200661 struct hlist_node d_list; /* for table with all the dests */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Julius Volze7ade462008-09-02 15:55:33 +0200663 u16 af; /* address family */
Al Viro014d7302006-09-28 14:29:52 -0700664 __be16 port; /* port number of the server */
Hans Schillstromf2431e62011-01-03 14:45:00 +0100665 union nf_inet_addr addr; /* IP address of the server */
Eric Dumazet95c96172012-04-15 05:58:06 +0000666 volatile unsigned int flags; /* dest status flags */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 atomic_t conn_flags; /* flags to copy to conn */
668 atomic_t weight; /* server weight */
669
670 atomic_t refcnt; /* reference counter */
671 struct ip_vs_stats stats; /* statistics */
Julian Anastasovbcbde4c2013-09-12 11:21:07 +0300672 unsigned long idle_start; /* start time, jiffies */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
674 /* connection counters and thresholds */
675 atomic_t activeconns; /* active connections */
676 atomic_t inactconns; /* inactive connections */
677 atomic_t persistconns; /* persistent connections */
678 __u32 u_threshold; /* upper threshold */
679 __u32 l_threshold; /* lower threshold */
680
681 /* for destination cache */
682 spinlock_t dst_lock; /* lock of dst_cache */
Julian Anastasov026ace02013-03-21 11:58:06 +0200683 struct ip_vs_dest_dst __rcu *dest_dst; /* cached dst info */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
685 /* for virtual service */
Julian Anastasovbcbde4c2013-09-12 11:21:07 +0300686 struct ip_vs_service __rcu *svc; /* service it belongs to */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 __u16 protocol; /* which protocol (TCP/UDP) */
Al Viro014d7302006-09-28 14:29:52 -0700688 __be16 vport; /* virtual port number */
Hans Schillstromf2431e62011-01-03 14:45:00 +0100689 union nf_inet_addr vaddr; /* virtual IP address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 __u32 vfwmark; /* firewall mark of service */
Julian Anastasov276472e2013-03-21 11:58:08 +0200691
Julian Anastasov578bc3e2013-03-22 11:46:49 +0200692 struct list_head t_list; /* in dest_trash */
Julian Anastasov276472e2013-03-21 11:58:08 +0200693 unsigned int in_rs_table:1; /* we are in rs_table */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694};
695
Simon Horman07dcc682014-09-30 10:50:06 +0900696/* The scheduler object */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697struct ip_vs_scheduler {
698 struct list_head n_list; /* d-linked list head */
699 char *name; /* scheduler name */
700 atomic_t refcnt; /* reference counter */
701 struct module *module; /* THIS_MODULE/NULL */
702
703 /* scheduler initializing service */
704 int (*init_service)(struct ip_vs_service *svc);
705 /* scheduling service finish */
Julian Anastasoved3ffc42013-03-22 11:46:50 +0200706 void (*done_service)(struct ip_vs_service *svc);
Julian Anastasov6b6df462013-03-22 11:46:37 +0200707 /* dest is linked */
708 int (*add_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest);
709 /* dest is unlinked */
710 int (*del_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest);
711 /* dest is updated */
712 int (*upd_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 /* selecting a server from the given service */
715 struct ip_vs_dest* (*schedule)(struct ip_vs_service *svc,
Julian Anastasovbba54de2013-06-16 09:09:36 +0300716 const struct sk_buff *skb,
717 struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718};
719
Simon Horman85999282010-08-22 21:37:53 +0900720/* The persistence engine object */
721struct ip_vs_pe {
722 struct list_head n_list; /* d-linked list head */
723 char *name; /* scheduler name */
724 atomic_t refcnt; /* reference counter */
725 struct module *module; /* THIS_MODULE/NULL */
726
727 /* get the connection template, if any */
728 int (*fill_param)(struct ip_vs_conn_param *p, struct sk_buff *skb);
729 bool (*ct_match)(const struct ip_vs_conn_param *p,
730 struct ip_vs_conn *ct);
731 u32 (*hashkey_raw)(const struct ip_vs_conn_param *p, u32 initval,
732 bool inverse);
Simon Hormana3c918a2010-08-22 21:37:53 +0900733 int (*show_pe_data)(const struct ip_vs_conn *cp, char *buf);
Marco Angaroni39b97222016-04-05 18:26:29 +0200734 /* create connections for real-server outgoing packets */
735 struct ip_vs_conn* (*conn_out)(struct ip_vs_service *svc,
736 struct ip_vs_dest *dest,
737 struct sk_buff *skb,
738 const struct ip_vs_iphdr *iph,
739 __be16 dport, __be16 cport);
Simon Horman85999282010-08-22 21:37:53 +0900740};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Simon Horman07dcc682014-09-30 10:50:06 +0900742/* The application module object (a.k.a. app incarnation) */
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000743struct ip_vs_app {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 struct list_head a_list; /* member in app list */
745 int type; /* IP_VS_APP_TYPE_xxx */
746 char *name; /* application module name */
747 __u16 protocol;
748 struct module *module; /* THIS_MODULE/NULL */
749 struct list_head incs_list; /* list of incarnations */
750
751 /* members for application incarnations */
752 struct list_head p_list; /* member in proto app list */
753 struct ip_vs_app *app; /* its real application */
Al Viro014d7302006-09-28 14:29:52 -0700754 __be16 port; /* port number in net order */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 atomic_t usecnt; /* usage counter */
Julian Anastasov363c97d2013-03-21 11:58:07 +0200756 struct rcu_head rcu_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Simon Horman07dcc682014-09-30 10:50:06 +0900758 /* output hook: Process packet in inout direction, diff set for TCP.
Julian Anastasov8b27b102010-10-17 16:17:20 +0300759 * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok,
760 * 2=Mangled but checksum was not updated
761 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 int (*pkt_out)(struct ip_vs_app *, struct ip_vs_conn *,
Herbert Xu3db05fe2007-10-15 00:53:15 -0700763 struct sk_buff *, int *diff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Simon Horman07dcc682014-09-30 10:50:06 +0900765 /* input hook: Process packet in outin direction, diff set for TCP.
Julian Anastasov8b27b102010-10-17 16:17:20 +0300766 * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok,
767 * 2=Mangled but checksum was not updated
768 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 int (*pkt_in)(struct ip_vs_app *, struct ip_vs_conn *,
Herbert Xu3db05fe2007-10-15 00:53:15 -0700770 struct sk_buff *, int *diff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772 /* ip_vs_app initializer */
773 int (*init_conn)(struct ip_vs_app *, struct ip_vs_conn *);
774
775 /* ip_vs_app finish */
776 int (*done_conn)(struct ip_vs_app *, struct ip_vs_conn *);
777
778
779 /* not used now */
780 int (*bind_conn)(struct ip_vs_app *, struct ip_vs_conn *,
781 struct ip_vs_protocol *);
782
783 void (*unbind_conn)(struct ip_vs_app *, struct ip_vs_conn *);
784
785 int * timeout_table;
786 int * timeouts;
787 int timeouts_size;
788
789 int (*conn_schedule)(struct sk_buff *skb, struct ip_vs_app *app,
790 int *verdict, struct ip_vs_conn **cpp);
791
792 struct ip_vs_conn *
793 (*conn_in_get)(const struct sk_buff *skb, struct ip_vs_app *app,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200794 const struct iphdr *iph, int inverse);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
796 struct ip_vs_conn *
797 (*conn_out_get)(const struct sk_buff *skb, struct ip_vs_app *app,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +0200798 const struct iphdr *iph, int inverse);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800 int (*state_transition)(struct ip_vs_conn *cp, int direction,
801 const struct sk_buff *skb,
802 struct ip_vs_app *app);
803
804 void (*timeout_change)(struct ip_vs_app *app, int flags);
805};
806
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200807struct ipvs_master_sync_state {
808 struct list_head sync_queue;
809 struct ip_vs_sync_buff *sync_buff;
Zhang Yanfei07995672013-04-29 11:55:10 -0700810 unsigned long sync_queue_len;
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200811 unsigned int sync_queue_delay;
812 struct task_struct *master_thread;
813 struct delayed_work master_wakeup_work;
814 struct netns_ipvs *ipvs;
815};
816
Julian Anastasov578bc3e2013-03-22 11:46:49 +0200817/* How much time to keep dests in trash */
818#define IP_VS_DEST_TRASH_PERIOD (120 * HZ)
819
Julian Anastasove4ff6752015-07-26 15:03:27 +0300820struct ipvs_sync_daemon_cfg {
Julian Anastasovd3328812015-07-26 15:03:28 +0300821 union nf_inet_addr mcast_group;
Julian Anastasove4ff6752015-07-26 15:03:27 +0300822 int syncid;
823 u16 sync_maxlen;
Julian Anastasovd3328812015-07-26 15:03:28 +0300824 u16 mcast_port;
825 u8 mcast_af;
826 u8 mcast_ttl;
Julian Anastasove4ff6752015-07-26 15:03:27 +0300827 /* multicast interface name */
828 char mcast_ifn[IP_VS_IFNAME_MAXLEN];
829};
830
Julian Anastasov2553d062011-03-04 12:18:07 +0200831/* IPVS in network namespace */
832struct netns_ipvs {
833 int gen; /* Generation */
Hans Schillstrom7a4f0762011-05-03 22:09:31 +0200834 int enable; /* enable like nf_hooks do */
Simon Horman07dcc682014-09-30 10:50:06 +0900835 /* Hash table: for real service lookups */
Julian Anastasov2553d062011-03-04 12:18:07 +0200836 #define IP_VS_RTAB_BITS 4
837 #define IP_VS_RTAB_SIZE (1 << IP_VS_RTAB_BITS)
838 #define IP_VS_RTAB_MASK (IP_VS_RTAB_SIZE - 1)
839
Julian Anastasov276472e2013-03-21 11:58:08 +0200840 struct hlist_head rs_table[IP_VS_RTAB_SIZE];
Julian Anastasov2553d062011-03-04 12:18:07 +0200841 /* ip_vs_app */
842 struct list_head app_list;
Julian Anastasov2553d062011-03-04 12:18:07 +0200843 /* ip_vs_proto */
844 #define IP_VS_PROTO_TAB_SIZE 32 /* must be power of 2 */
845 struct ip_vs_proto_data *proto_data_table[IP_VS_PROTO_TAB_SIZE];
846 /* ip_vs_proto_tcp */
847#ifdef CONFIG_IP_VS_PROTO_TCP
848 #define TCP_APP_TAB_BITS 4
849 #define TCP_APP_TAB_SIZE (1 << TCP_APP_TAB_BITS)
850 #define TCP_APP_TAB_MASK (TCP_APP_TAB_SIZE - 1)
851 struct list_head tcp_apps[TCP_APP_TAB_SIZE];
Julian Anastasov2553d062011-03-04 12:18:07 +0200852#endif
853 /* ip_vs_proto_udp */
854#ifdef CONFIG_IP_VS_PROTO_UDP
855 #define UDP_APP_TAB_BITS 4
856 #define UDP_APP_TAB_SIZE (1 << UDP_APP_TAB_BITS)
857 #define UDP_APP_TAB_MASK (UDP_APP_TAB_SIZE - 1)
858 struct list_head udp_apps[UDP_APP_TAB_SIZE];
Julian Anastasov2553d062011-03-04 12:18:07 +0200859#endif
860 /* ip_vs_proto_sctp */
861#ifdef CONFIG_IP_VS_PROTO_SCTP
862 #define SCTP_APP_TAB_BITS 4
863 #define SCTP_APP_TAB_SIZE (1 << SCTP_APP_TAB_BITS)
864 #define SCTP_APP_TAB_MASK (SCTP_APP_TAB_SIZE - 1)
865 /* Hash table for SCTP application incarnations */
866 struct list_head sctp_apps[SCTP_APP_TAB_SIZE];
Julian Anastasov2553d062011-03-04 12:18:07 +0200867#endif
868 /* ip_vs_conn */
Simon Horman07dcc682014-09-30 10:50:06 +0900869 atomic_t conn_count; /* connection counter */
Julian Anastasov2553d062011-03-04 12:18:07 +0200870
871 /* ip_vs_ctl */
Julian Anastasov2a0751a2011-03-04 12:20:35 +0200872 struct ip_vs_stats tot_stats; /* Statistics & est. */
Julian Anastasov2553d062011-03-04 12:18:07 +0200873
874 int num_services; /* no of virtual services */
Simon Hormanf2247fb2011-02-04 18:33:02 +0900875
Julian Anastasov2553d062011-03-04 12:18:07 +0200876 /* Trash for destinations */
877 struct list_head dest_trash;
Julian Anastasov578bc3e2013-03-22 11:46:49 +0200878 spinlock_t dest_trash_lock;
879 struct timer_list dest_trash_timer; /* expiration timer */
Julian Anastasov2553d062011-03-04 12:18:07 +0200880 /* Service counters */
881 atomic_t ftpsvc_counter;
882 atomic_t nullsvc_counter;
Marco Angaroni39b97222016-04-05 18:26:29 +0200883 atomic_t conn_out_counter;
Julian Anastasov2553d062011-03-04 12:18:07 +0200884
Simon Hormanf2247fb2011-02-04 18:33:02 +0900885#ifdef CONFIG_SYSCTL
886 /* 1/rate drop and drop-entry variables */
887 struct delayed_work defense_work; /* Work handler */
888 int drop_rate;
889 int drop_counter;
890 atomic_t dropentry;
891 /* locks in ctl.c */
892 spinlock_t dropentry_lock; /* drop entry handling */
893 spinlock_t droppacket_lock; /* drop packet handling */
894 spinlock_t securetcp_lock; /* state and timeout tables */
895
Julian Anastasov2553d062011-03-04 12:18:07 +0200896 /* sys-ctl struct */
897 struct ctl_table_header *sysctl_hdr;
898 struct ctl_table *sysctl_tbl;
Simon Hormanf2247fb2011-02-04 18:33:02 +0900899#endif
900
Julian Anastasov2553d062011-03-04 12:18:07 +0200901 /* sysctl variables */
902 int sysctl_amemthresh;
903 int sysctl_am_droprate;
904 int sysctl_drop_entry;
905 int sysctl_drop_packet;
906 int sysctl_secure_tcp;
907#ifdef CONFIG_IP_VS_NFCT
908 int sysctl_conntrack;
909#endif
910 int sysctl_snat_reroute;
911 int sysctl_sync_ver;
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200912 int sysctl_sync_ports;
Julian Anastasov4d0c8752013-06-24 22:44:41 +0300913 int sysctl_sync_persist_mode;
Zhang Yanfei07995672013-04-29 11:55:10 -0700914 unsigned long sysctl_sync_qlen_max;
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +0200915 int sysctl_sync_sock_size;
Julian Anastasov2553d062011-03-04 12:18:07 +0200916 int sysctl_cache_bypass;
917 int sysctl_expire_nodest_conn;
Alexander Frolkinc6c96c12013-06-13 08:56:15 +0100918 int sysctl_sloppy_tcp;
919 int sysctl_sloppy_sctp;
Julian Anastasov2553d062011-03-04 12:18:07 +0200920 int sysctl_expire_quiescent_template;
921 int sysctl_sync_threshold[2];
Julian Anastasov749c42b2012-04-24 23:46:40 +0300922 unsigned int sysctl_sync_refresh_period;
923 int sysctl_sync_retries;
Julian Anastasov2553d062011-03-04 12:18:07 +0200924 int sysctl_nat_icmp_send;
Julian Anastasov3654e612012-07-20 11:59:53 +0300925 int sysctl_pmtu_disc;
Julian Anastasov0c125822013-03-09 23:25:04 +0200926 int sysctl_backup_only;
Marcelo Ricardo Leitnerd752c362015-02-23 15:02:34 -0300927 int sysctl_conn_reuse_mode;
Alex Gartrell94485fe2015-08-26 09:40:34 -0700928 int sysctl_schedule_icmp;
Alex Gartrell4e478092015-09-14 23:23:05 -0700929 int sysctl_ignore_tunneled;
Julian Anastasov2553d062011-03-04 12:18:07 +0200930
931 /* ip_vs_lblc */
932 int sysctl_lblc_expiration;
933 struct ctl_table_header *lblc_ctl_header;
934 struct ctl_table *lblc_ctl_table;
935 /* ip_vs_lblcr */
936 int sysctl_lblcr_expiration;
937 struct ctl_table_header *lblcr_ctl_header;
938 struct ctl_table *lblcr_ctl_table;
939 /* ip_vs_est */
940 struct list_head est_list; /* estimator list */
941 spinlock_t est_lock;
942 struct timer_list est_timer; /* Estimation timer */
943 /* ip_vs_sync */
Julian Anastasov2553d062011-03-04 12:18:07 +0200944 spinlock_t sync_lock;
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200945 struct ipvs_master_sync_state *ms;
Julian Anastasov2553d062011-03-04 12:18:07 +0200946 spinlock_t sync_buff_lock;
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200947 struct task_struct **backup_threads;
948 int threads_mask;
Julian Anastasov2553d062011-03-04 12:18:07 +0200949 volatile int sync_state;
Hans Schillstromae1d48b2011-10-11 10:54:35 +0900950 struct mutex sync_mutex;
Julian Anastasove4ff6752015-07-26 15:03:27 +0300951 struct ipvs_sync_daemon_cfg mcfg; /* Master Configuration */
952 struct ipvs_sync_daemon_cfg bcfg; /* Backup Configuration */
Julian Anastasov2553d062011-03-04 12:18:07 +0200953 /* net name space ptr */
954 struct net *net; /* Needed by timer routines */
Simon Horman07dcc682014-09-30 10:50:06 +0900955 /* Number of heterogeneous destinations, needed becaus heterogeneous
956 * are not supported when synchronization is enabled.
957 */
Alex Gartrell391f503d2014-09-09 16:40:24 -0700958 unsigned int mixed_address_family_dests;
Julian Anastasov2553d062011-03-04 12:18:07 +0200959};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Simon Horman59e03502011-02-04 18:33:01 +0900961#define DEFAULT_SYNC_THRESHOLD 3
962#define DEFAULT_SYNC_PERIOD 50
Simon Horman7532e8d2011-02-04 18:33:01 +0900963#define DEFAULT_SYNC_VER 1
Alexander Frolkinc6c96c12013-06-13 08:56:15 +0100964#define DEFAULT_SLOPPY_TCP 0
965#define DEFAULT_SLOPPY_SCTP 0
Julian Anastasov749c42b2012-04-24 23:46:40 +0300966#define DEFAULT_SYNC_REFRESH_PERIOD (0U * HZ)
967#define DEFAULT_SYNC_RETRIES 0
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +0200968#define IPVS_SYNC_WAKEUP_RATE 8
969#define IPVS_SYNC_QLEN_MAX (IPVS_SYNC_WAKEUP_RATE * 4)
970#define IPVS_SYNC_SEND_DELAY (HZ / 50)
971#define IPVS_SYNC_CHECK_PERIOD HZ
Julian Anastasov749c42b2012-04-24 23:46:40 +0300972#define IPVS_SYNC_FLUSH_TIME (HZ * 2)
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +0200973#define IPVS_SYNC_PORTS_MAX (1 << 6)
Simon Horman59e03502011-02-04 18:33:01 +0900974
975#ifdef CONFIG_SYSCTL
976
977static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs)
978{
979 return ipvs->sysctl_sync_threshold[0];
980}
981
982static inline int sysctl_sync_period(struct netns_ipvs *ipvs)
983{
Julian Anastasov749c42b2012-04-24 23:46:40 +0300984 return ACCESS_ONCE(ipvs->sysctl_sync_threshold[1]);
985}
986
987static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs)
988{
989 return ACCESS_ONCE(ipvs->sysctl_sync_refresh_period);
990}
991
992static inline int sysctl_sync_retries(struct netns_ipvs *ipvs)
993{
994 return ipvs->sysctl_sync_retries;
Simon Horman59e03502011-02-04 18:33:01 +0900995}
996
Simon Horman7532e8d2011-02-04 18:33:01 +0900997static inline int sysctl_sync_ver(struct netns_ipvs *ipvs)
998{
999 return ipvs->sysctl_sync_ver;
1000}
1001
Alexander Frolkinc6c96c12013-06-13 08:56:15 +01001002static inline int sysctl_sloppy_tcp(struct netns_ipvs *ipvs)
1003{
1004 return ipvs->sysctl_sloppy_tcp;
1005}
1006
1007static inline int sysctl_sloppy_sctp(struct netns_ipvs *ipvs)
1008{
1009 return ipvs->sysctl_sloppy_sctp;
1010}
1011
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +02001012static inline int sysctl_sync_ports(struct netns_ipvs *ipvs)
1013{
1014 return ACCESS_ONCE(ipvs->sysctl_sync_ports);
1015}
1016
Julian Anastasov4d0c8752013-06-24 22:44:41 +03001017static inline int sysctl_sync_persist_mode(struct netns_ipvs *ipvs)
1018{
1019 return ipvs->sysctl_sync_persist_mode;
1020}
1021
Zhang Yanfei07995672013-04-29 11:55:10 -07001022static inline unsigned long sysctl_sync_qlen_max(struct netns_ipvs *ipvs)
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +02001023{
1024 return ipvs->sysctl_sync_qlen_max;
1025}
1026
1027static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs)
1028{
1029 return ipvs->sysctl_sync_sock_size;
1030}
1031
Julian Anastasov3654e612012-07-20 11:59:53 +03001032static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs)
1033{
1034 return ipvs->sysctl_pmtu_disc;
1035}
1036
Julian Anastasov0c125822013-03-09 23:25:04 +02001037static inline int sysctl_backup_only(struct netns_ipvs *ipvs)
1038{
1039 return ipvs->sync_state & IP_VS_STATE_BACKUP &&
1040 ipvs->sysctl_backup_only;
1041}
1042
Marcelo Ricardo Leitnerd752c362015-02-23 15:02:34 -03001043static inline int sysctl_conn_reuse_mode(struct netns_ipvs *ipvs)
1044{
1045 return ipvs->sysctl_conn_reuse_mode;
1046}
1047
Alex Gartrell94485fe2015-08-26 09:40:34 -07001048static inline int sysctl_schedule_icmp(struct netns_ipvs *ipvs)
1049{
1050 return ipvs->sysctl_schedule_icmp;
1051}
1052
Alex Gartrell4e478092015-09-14 23:23:05 -07001053static inline int sysctl_ignore_tunneled(struct netns_ipvs *ipvs)
1054{
1055 return ipvs->sysctl_ignore_tunneled;
1056}
1057
Eric W. Biederman57032942015-09-21 13:02:49 -05001058static inline int sysctl_cache_bypass(struct netns_ipvs *ipvs)
1059{
1060 return ipvs->sysctl_cache_bypass;
1061}
1062
Simon Horman59e03502011-02-04 18:33:01 +09001063#else
1064
1065static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs)
1066{
1067 return DEFAULT_SYNC_THRESHOLD;
1068}
1069
1070static inline int sysctl_sync_period(struct netns_ipvs *ipvs)
1071{
1072 return DEFAULT_SYNC_PERIOD;
1073}
1074
Julian Anastasov749c42b2012-04-24 23:46:40 +03001075static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs)
1076{
1077 return DEFAULT_SYNC_REFRESH_PERIOD;
1078}
1079
1080static inline int sysctl_sync_retries(struct netns_ipvs *ipvs)
1081{
1082 return DEFAULT_SYNC_RETRIES & 3;
1083}
1084
Simon Horman7532e8d2011-02-04 18:33:01 +09001085static inline int sysctl_sync_ver(struct netns_ipvs *ipvs)
1086{
1087 return DEFAULT_SYNC_VER;
1088}
1089
Alexander Frolkinc6c96c12013-06-13 08:56:15 +01001090static inline int sysctl_sloppy_tcp(struct netns_ipvs *ipvs)
1091{
1092 return DEFAULT_SLOPPY_TCP;
1093}
1094
1095static inline int sysctl_sloppy_sctp(struct netns_ipvs *ipvs)
1096{
1097 return DEFAULT_SLOPPY_SCTP;
1098}
1099
Pablo Neira Ayusof73181c2012-05-08 19:40:30 +02001100static inline int sysctl_sync_ports(struct netns_ipvs *ipvs)
1101{
1102 return 1;
1103}
1104
Julian Anastasov4d0c8752013-06-24 22:44:41 +03001105static inline int sysctl_sync_persist_mode(struct netns_ipvs *ipvs)
1106{
1107 return 0;
1108}
1109
Zhang Yanfei07995672013-04-29 11:55:10 -07001110static inline unsigned long sysctl_sync_qlen_max(struct netns_ipvs *ipvs)
Pablo Neira Ayuso1c003b12012-05-08 19:39:49 +02001111{
1112 return IPVS_SYNC_QLEN_MAX;
1113}
1114
1115static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs)
1116{
1117 return 0;
1118}
1119
Julian Anastasov3654e612012-07-20 11:59:53 +03001120static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs)
1121{
1122 return 1;
1123}
1124
Julian Anastasov0c125822013-03-09 23:25:04 +02001125static inline int sysctl_backup_only(struct netns_ipvs *ipvs)
1126{
1127 return 0;
1128}
1129
Marcelo Ricardo Leitnerd752c362015-02-23 15:02:34 -03001130static inline int sysctl_conn_reuse_mode(struct netns_ipvs *ipvs)
1131{
1132 return 1;
1133}
1134
Alex Gartrell94485fe2015-08-26 09:40:34 -07001135static inline int sysctl_schedule_icmp(struct netns_ipvs *ipvs)
1136{
1137 return 0;
1138}
1139
Alex Gartrell4e478092015-09-14 23:23:05 -07001140static inline int sysctl_ignore_tunneled(struct netns_ipvs *ipvs)
1141{
1142 return 0;
1143}
1144
Eric W. Biederman57032942015-09-21 13:02:49 -05001145static inline int sysctl_cache_bypass(struct netns_ipvs *ipvs)
1146{
1147 return 0;
1148}
1149
Simon Horman59e03502011-02-04 18:33:01 +09001150#endif
1151
Simon Horman07dcc682014-09-30 10:50:06 +09001152/* IPVS core functions
1153 * (from ip_vs_core.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 */
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001155const char *ip_vs_proto_name(unsigned int proto);
1156void ip_vs_init_hash_table(struct list_head *table, int rows);
Marco Angaroni39b97222016-04-05 18:26:29 +02001157struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1158 struct ip_vs_dest *dest,
1159 struct sk_buff *skb,
1160 const struct ip_vs_iphdr *iph,
1161 __be16 dport,
1162 __be16 cport);
Sven Wegenerafdd6142008-08-10 09:18:01 +00001163#define IP_VS_INIT_HASH_TABLE(t) ip_vs_init_hash_table((t), ARRAY_SIZE((t)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165#define IP_VS_APP_TYPE_FTP 1
1166
Simon Horman07dcc682014-09-30 10:50:06 +09001167/* ip_vs_conn handling functions
1168 * (from ip_vs_conn.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170enum {
1171 IP_VS_DIR_INPUT = 0,
1172 IP_VS_DIR_OUTPUT,
1173 IP_VS_DIR_INPUT_ONLY,
1174 IP_VS_DIR_LAST,
1175};
1176
Eric W. Biederman19913de2015-09-21 13:01:43 -05001177static inline void ip_vs_conn_fill_param(struct netns_ipvs *ipvs, int af, int protocol,
Simon Hormanf11017e2010-08-22 21:37:52 +09001178 const union nf_inet_addr *caddr,
1179 __be16 cport,
1180 const union nf_inet_addr *vaddr,
1181 __be16 vport,
1182 struct ip_vs_conn_param *p)
1183{
Eric W. Biederman19913de2015-09-21 13:01:43 -05001184 p->ipvs = ipvs;
Simon Hormanf11017e2010-08-22 21:37:52 +09001185 p->af = af;
1186 p->protocol = protocol;
1187 p->caddr = caddr;
1188 p->cport = cport;
1189 p->vaddr = vaddr;
1190 p->vport = vport;
Simon Horman85999282010-08-22 21:37:53 +09001191 p->pe = NULL;
1192 p->pe_data = NULL;
Simon Hormanf11017e2010-08-22 21:37:52 +09001193}
Julius Volz28364a52008-09-02 15:55:43 +02001194
Simon Hormanf11017e2010-08-22 21:37:52 +09001195struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p);
1196struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p);
Julius Volz28364a52008-09-02 15:55:43 +02001197
Eric W. Biedermanab161972015-09-21 13:02:38 -05001198struct ip_vs_conn * ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af,
1199 const struct sk_buff *skb,
Alex Gartrell802c41a2015-08-26 09:40:32 -07001200 const struct ip_vs_iphdr *iph);
Simon Horman5c0d2372010-08-02 17:12:44 +02001201
Simon Hormanf11017e2010-08-22 21:37:52 +09001202struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Eric W. Biederman0cf705c8c2015-09-21 13:02:39 -05001204struct ip_vs_conn * ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af,
1205 const struct sk_buff *skb,
Alex Gartrell802c41a2015-08-26 09:40:32 -07001206 const struct ip_vs_iphdr *iph);
Simon Horman5c0d2372010-08-02 17:12:44 +02001207
Julian Anastasov088339a2013-03-21 11:58:10 +02001208/* Get reference to gain full access to conn.
1209 * By default, RCU read-side critical sections have access only to
1210 * conn fields and its PE data, see ip_vs_conn_rcu_free() for reference.
1211 */
1212static inline bool __ip_vs_conn_get(struct ip_vs_conn *cp)
1213{
1214 return atomic_inc_not_zero(&cp->refcnt);
1215}
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217/* put back the conn without restarting its timer */
1218static inline void __ip_vs_conn_put(struct ip_vs_conn *cp)
1219{
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001220 smp_mb__before_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 atomic_dec(&cp->refcnt);
1222}
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001223void ip_vs_conn_put(struct ip_vs_conn *cp);
1224void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Alex Gartrellba385282014-09-09 16:40:23 -07001226struct ip_vs_conn *ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
Simon Hormanf11017e2010-08-22 21:37:52 +09001227 const union nf_inet_addr *daddr,
Eric Dumazet95c96172012-04-15 05:58:06 +00001228 __be16 dport, unsigned int flags,
Hans Schillstrom0e051e62010-11-19 14:25:07 +01001229 struct ip_vs_dest *dest, __u32 fwmark);
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001230void ip_vs_conn_expire_now(struct ip_vs_conn *cp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001232const char *ip_vs_state_name(__u16 proto, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Eric W. Biederman69f39092015-09-21 13:02:57 -05001234void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp);
Marco Angaroni3ec10d32016-05-16 19:18:09 +02001235int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest);
Eric W. Biederman423b5592015-09-21 13:02:24 -05001236void ip_vs_random_dropentry(struct netns_ipvs *ipvs);
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001237int ip_vs_conn_init(void);
1238void ip_vs_conn_cleanup(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
1240static inline void ip_vs_control_del(struct ip_vs_conn *cp)
1241{
1242 struct ip_vs_conn *ctl_cp = cp->control;
1243 if (!ctl_cp) {
Julius Volzcfc78c52008-09-02 15:55:53 +02001244 IP_VS_ERR_BUF("request control DEL for uncontrolled: "
1245 "%s:%d to %s:%d\n",
1246 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1247 ntohs(cp->cport),
1248 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
1249 ntohs(cp->vport));
1250
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 return;
1252 }
1253
Julius Volzcfc78c52008-09-02 15:55:53 +02001254 IP_VS_DBG_BUF(7, "DELeting control for: "
1255 "cp.dst=%s:%d ctl_cp.dst=%s:%d\n",
1256 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1257 ntohs(cp->cport),
1258 IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr),
1259 ntohs(ctl_cp->cport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261 cp->control = NULL;
1262 if (atomic_read(&ctl_cp->n_control) == 0) {
Julius Volzcfc78c52008-09-02 15:55:53 +02001263 IP_VS_ERR_BUF("BUG control DEL with n=0 : "
1264 "%s:%d to %s:%d\n",
1265 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1266 ntohs(cp->cport),
1267 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
1268 ntohs(cp->vport));
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 return;
1271 }
1272 atomic_dec(&ctl_cp->n_control);
1273}
1274
1275static inline void
1276ip_vs_control_add(struct ip_vs_conn *cp, struct ip_vs_conn *ctl_cp)
1277{
1278 if (cp->control) {
Julius Volzcfc78c52008-09-02 15:55:53 +02001279 IP_VS_ERR_BUF("request control ADD for already controlled: "
1280 "%s:%d to %s:%d\n",
1281 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1282 ntohs(cp->cport),
1283 IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
1284 ntohs(cp->vport));
1285
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 ip_vs_control_del(cp);
1287 }
1288
Julius Volzcfc78c52008-09-02 15:55:53 +02001289 IP_VS_DBG_BUF(7, "ADDing control for: "
1290 "cp.dst=%s:%d ctl_cp.dst=%s:%d\n",
1291 IP_VS_DBG_ADDR(cp->af, &cp->caddr),
1292 ntohs(cp->cport),
1293 IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr),
1294 ntohs(ctl_cp->cport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 cp->control = ctl_cp;
1297 atomic_inc(&ctl_cp->n_control);
1298}
1299
Simon Horman07dcc682014-09-30 10:50:06 +09001300/* IPVS netns init & cleanup functions */
Eric W. Biedermana4dd0362015-09-21 13:02:28 -05001301int ip_vs_estimator_net_init(struct netns_ipvs *ipvs);
Eric W. Biederman3d993762015-09-21 13:02:26 -05001302int ip_vs_control_net_init(struct netns_ipvs *ipvs);
Eric W. Biederman7d1f88e2015-09-21 13:02:58 -05001303int ip_vs_protocol_net_init(struct netns_ipvs *ipvs);
Eric W. Biedermanb5dd2122015-09-21 13:02:34 -05001304int ip_vs_app_net_init(struct netns_ipvs *ipvs);
Eric W. Biederman2f3edc6a2015-09-21 13:02:42 -05001305int ip_vs_conn_net_init(struct netns_ipvs *ipvs);
Eric W. Biederman802cb432015-09-21 13:02:20 -05001306int ip_vs_sync_net_init(struct netns_ipvs *ipvs);
Eric W. Biederman2f3edc6a2015-09-21 13:02:42 -05001307void ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs);
Eric W. Biedermanb5dd2122015-09-21 13:02:34 -05001308void ip_vs_app_net_cleanup(struct netns_ipvs *ipvs);
Eric W. Biederman7d1f88e2015-09-21 13:02:58 -05001309void ip_vs_protocol_net_cleanup(struct netns_ipvs *ipvs);
Eric W. Biederman3d993762015-09-21 13:02:26 -05001310void ip_vs_control_net_cleanup(struct netns_ipvs *ipvs);
Eric W. Biedermana4dd0362015-09-21 13:02:28 -05001311void ip_vs_estimator_net_cleanup(struct netns_ipvs *ipvs);
Eric W. Biedermanebea1f72015-09-21 13:02:21 -05001312void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs);
Eric W. Biederman56d21692015-09-21 13:01:58 -05001313void ip_vs_service_net_cleanup(struct netns_ipvs *ipvs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Simon Horman07dcc682014-09-30 10:50:06 +09001315/* IPVS application functions
1316 * (from ip_vs_app.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 */
1318#define IP_VS_APP_MAX_PORTS 8
Eric W. Biederman9f8128a2015-09-21 13:02:32 -05001319struct ip_vs_app *register_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app);
1320void unregister_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app);
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001321int ip_vs_bind_app(struct ip_vs_conn *cp, struct ip_vs_protocol *pp);
1322void ip_vs_unbind_app(struct ip_vs_conn *cp);
Eric W. Biederman3250dc92015-09-21 13:02:31 -05001323int register_ip_vs_app_inc(struct netns_ipvs *ipvs, struct ip_vs_app *app, __u16 proto,
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001324 __u16 port);
1325int ip_vs_app_inc_get(struct ip_vs_app *inc);
1326void ip_vs_app_inc_put(struct ip_vs_app *inc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001328int ip_vs_app_pkt_out(struct ip_vs_conn *, struct sk_buff *skb);
1329int ip_vs_app_pkt_in(struct ip_vs_conn *, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Simon Horman8be67a62010-08-22 21:37:54 +09001331int register_ip_vs_pe(struct ip_vs_pe *pe);
1332int unregister_ip_vs_pe(struct ip_vs_pe *pe);
Simon Hormane9e5eee2010-11-08 20:05:57 +09001333struct ip_vs_pe *ip_vs_pe_getbyname(const char *name);
Hans Schillstromfe5e7a12010-11-19 14:25:12 +01001334struct ip_vs_pe *__ip_vs_pe_getbyname(const char *pe_name);
Simon Hormane9e5eee2010-11-08 20:05:57 +09001335
Simon Horman07dcc682014-09-30 10:50:06 +09001336/* Use a #define to avoid all of module.h just for these trivial ops */
Paul Gortmaker69e7dae2011-08-29 12:40:15 -04001337#define ip_vs_pe_get(pe) \
1338 if (pe && pe->module) \
Simon Hormane9e5eee2010-11-08 20:05:57 +09001339 __module_get(pe->module);
Simon Hormane9e5eee2010-11-08 20:05:57 +09001340
Paul Gortmaker69e7dae2011-08-29 12:40:15 -04001341#define ip_vs_pe_put(pe) \
1342 if (pe && pe->module) \
Simon Hormane9e5eee2010-11-08 20:05:57 +09001343 module_put(pe->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Simon Horman07dcc682014-09-30 10:50:06 +09001345/* IPVS protocol functions (from ip_vs_proto.c) */
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001346int ip_vs_protocol_init(void);
1347void ip_vs_protocol_cleanup(void);
1348void ip_vs_protocol_timeout_change(struct netns_ipvs *ipvs, int flags);
1349int *ip_vs_create_timeout_table(int *table, int size);
1350int ip_vs_set_state_timeout(int *table, int num, const char *const *names,
1351 const char *name, int to);
1352void ip_vs_tcpudp_debug_packet(int af, struct ip_vs_protocol *pp,
1353 const struct sk_buff *skb, int offset,
1354 const char *msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
1356extern struct ip_vs_protocol ip_vs_protocol_tcp;
1357extern struct ip_vs_protocol ip_vs_protocol_udp;
1358extern struct ip_vs_protocol ip_vs_protocol_icmp;
1359extern struct ip_vs_protocol ip_vs_protocol_esp;
1360extern struct ip_vs_protocol ip_vs_protocol_ah;
Venkata Mohan Reddy2906f662010-02-18 12:31:05 +01001361extern struct ip_vs_protocol ip_vs_protocol_sctp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
Simon Horman07dcc682014-09-30 10:50:06 +09001363/* Registering/unregistering scheduler functions
1364 * (from ip_vs_sched.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 */
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001366int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
1367int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler);
1368int ip_vs_bind_scheduler(struct ip_vs_service *svc,
1369 struct ip_vs_scheduler *scheduler);
1370void ip_vs_unbind_scheduler(struct ip_vs_service *svc,
1371 struct ip_vs_scheduler *sched);
1372struct ip_vs_scheduler *ip_vs_scheduler_get(const char *sched_name);
1373void ip_vs_scheduler_put(struct ip_vs_scheduler *scheduler);
1374struct ip_vs_conn *
Julian Anastasov190ecd22010-10-17 16:24:37 +03001375ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
Jesper Dangaard Brouerd4383f02012-09-26 14:07:17 +02001376 struct ip_vs_proto_data *pd, int *ignored,
1377 struct ip_vs_iphdr *iph);
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001378int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
1379 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001381void ip_vs_scheduler_err(struct ip_vs_service *svc, const char *msg);
Patrick Schaaf41ac51e2011-02-11 14:01:12 +01001382
Simon Horman07dcc682014-09-30 10:50:06 +09001383/* IPVS control data and functions (from ip_vs_ctl.c) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384extern struct ip_vs_stats ip_vs_stats;
Hans Schillstromb880c1f2010-11-19 14:25:14 +01001385extern int sysctl_ip_vs_sync_ver;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001387struct ip_vs_service *
Eric W. Biederman0a4fd6c2015-09-21 13:01:49 -05001388ip_vs_service_find(struct netns_ipvs *ipvs, int af, __u32 fwmark, __u16 protocol,
Julius Volz3c2e0502008-09-02 15:55:38 +02001389 const union nf_inet_addr *vaddr, __be16 vport);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
Eric W. Biederman48aed1b2015-09-21 13:01:50 -05001391bool ip_vs_has_real_service(struct netns_ipvs *ipvs, int af, __u16 protocol,
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001392 const union nf_inet_addr *daddr, __be16 dport);
Julius Volz7937df12008-09-02 15:55:48 +02001393
Marco Angaroni39b97222016-04-05 18:26:29 +02001394struct ip_vs_dest *
1395ip_vs_find_real_service(struct netns_ipvs *ipvs, int af, __u16 protocol,
1396 const union nf_inet_addr *daddr, __be16 dport);
1397
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001398int ip_vs_use_count_inc(void);
1399void ip_vs_use_count_dec(void);
1400int ip_vs_register_nl_ioctl(void);
1401void ip_vs_unregister_nl_ioctl(void);
1402int ip_vs_control_init(void);
1403void ip_vs_control_cleanup(void);
1404struct ip_vs_dest *
Eric W. Biedermandc2add62015-09-21 13:01:51 -05001405ip_vs_find_dest(struct netns_ipvs *ipvs, int svc_af, int dest_af,
Alex Gartrell655eef12014-09-09 16:40:21 -07001406 const union nf_inet_addr *daddr, __be16 dport,
1407 const union nf_inet_addr *vaddr, __be16 vport,
Julian Anastasov52793db2011-12-30 14:19:02 +09001408 __u16 protocol, __u32 fwmark, __u32 flags);
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001409void ip_vs_try_bind_dest(struct ip_vs_conn *cp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Julian Anastasovfca9c202013-03-22 11:46:38 +02001411static inline void ip_vs_dest_hold(struct ip_vs_dest *dest)
1412{
1413 atomic_inc(&dest->refcnt);
1414}
1415
1416static inline void ip_vs_dest_put(struct ip_vs_dest *dest)
1417{
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001418 smp_mb__before_atomic();
Julian Anastasovfca9c202013-03-22 11:46:38 +02001419 atomic_dec(&dest->refcnt);
1420}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Julian Anastasov9e4e9482013-10-09 09:24:27 +03001422static inline void ip_vs_dest_put_and_free(struct ip_vs_dest *dest)
1423{
1424 if (atomic_dec_return(&dest->refcnt) < 0)
1425 kfree(dest);
1426}
1427
Simon Horman07dcc682014-09-30 10:50:06 +09001428/* IPVS sync daemon data and function prototypes
1429 * (from ip_vs_sync.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 */
Eric W. Biederman6ac121d2015-09-21 13:02:10 -05001431int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *cfg,
Julian Anastasove4ff6752015-07-26 15:03:27 +03001432 int state);
Eric W. Biedermanb3cf3cb2015-09-21 13:02:11 -05001433int stop_sync_thread(struct netns_ipvs *ipvs, int state);
Eric W. Biedermanb61a8c12015-09-21 13:02:17 -05001434void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Simon Horman07dcc682014-09-30 10:50:06 +09001436/* IPVS rate estimator prototypes (from ip_vs_est.c) */
Eric W. Biederman0f34d542015-09-21 13:02:23 -05001437void ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats);
1438void ip_vs_stop_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats);
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001439void ip_vs_zero_estimator(struct ip_vs_stats *stats);
Julian Anastasovcd67cd52015-02-06 09:44:44 +02001440void ip_vs_read_estimator(struct ip_vs_kstats *dst, struct ip_vs_stats *stats);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
Simon Horman07dcc682014-09-30 10:50:06 +09001442/* Various IPVS packet transmitters (from ip_vs_xmit.c) */
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001443int ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1444 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1445int ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1446 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1447int ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1448 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1449int ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1450 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1451int ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1452 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1453int ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1454 struct ip_vs_protocol *pp, int offset,
1455 unsigned int hooknum, struct ip_vs_iphdr *iph);
1456void ip_vs_dest_dst_rcu_free(struct rcu_head *head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Julius Volzb3cdd2a72008-09-02 15:55:45 +02001458#ifdef CONFIG_IP_VS_IPV6
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001459int ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1460 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1461int ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1462 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1463int ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1464 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1465int ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1466 struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph);
1467int ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1468 struct ip_vs_protocol *pp, int offset,
1469 unsigned int hooknum, struct ip_vs_iphdr *iph);
Julius Volzb3cdd2a72008-09-02 15:55:45 +02001470#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Simon Horman3a1bbf12011-02-04 18:33:02 +09001472#ifdef CONFIG_SYSCTL
Simon Horman07dcc682014-09-30 10:50:06 +09001473/* This is a simple mechanism to ignore packets when
1474 * we are loaded. Just set ip_vs_drop_rate to 'n' and
1475 * we start to drop 1/rate of the packets
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 */
Hans Schillstroma0840e22011-01-03 14:44:58 +01001477static inline int ip_vs_todrop(struct netns_ipvs *ipvs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478{
Hans Schillstroma0840e22011-01-03 14:44:58 +01001479 if (!ipvs->drop_rate)
1480 return 0;
1481 if (--ipvs->drop_counter > 0)
1482 return 0;
1483 ipvs->drop_counter = ipvs->drop_rate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 return 1;
1485}
Simon Horman3a1bbf12011-02-04 18:33:02 +09001486#else
1487static inline int ip_vs_todrop(struct netns_ipvs *ipvs) { return 0; }
1488#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Simon Horman07dcc682014-09-30 10:50:06 +09001490/* ip_vs_fwd_tag returns the forwarding tag of the connection */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491#define IP_VS_FWD_METHOD(cp) (cp->flags & IP_VS_CONN_F_FWD_MASK)
1492
Adrian Bunk732db652005-09-01 17:40:26 -07001493static inline char ip_vs_fwd_tag(struct ip_vs_conn *cp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494{
1495 char fwd;
1496
1497 switch (IP_VS_FWD_METHOD(cp)) {
1498 case IP_VS_CONN_F_MASQ:
1499 fwd = 'M'; break;
1500 case IP_VS_CONN_F_LOCALNODE:
1501 fwd = 'L'; break;
1502 case IP_VS_CONN_F_TUNNEL:
1503 fwd = 'T'; break;
1504 case IP_VS_CONN_F_DROUTE:
1505 fwd = 'R'; break;
1506 case IP_VS_CONN_F_BYPASS:
1507 fwd = 'B'; break;
1508 default:
1509 fwd = '?'; break;
1510 }
1511 return fwd;
1512}
1513
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001514void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
1515 struct ip_vs_conn *cp, int dir);
Julius Volzb3cdd2a72008-09-02 15:55:45 +02001516
1517#ifdef CONFIG_IP_VS_IPV6
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001518void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
1519 struct ip_vs_conn *cp, int dir);
Julius Volzb3cdd2a72008-09-02 15:55:45 +02001520#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001522__sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Al Virof9214b22006-11-16 02:41:18 -08001524static inline __wsum ip_vs_check_diff4(__be32 old, __be32 new, __wsum oldsum)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
Al Virof9214b22006-11-16 02:41:18 -08001526 __be32 diff[2] = { ~old, new };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
Joe Perches07f07572008-11-19 15:44:53 -08001528 return csum_partial(diff, sizeof(diff), oldsum);
Al Virof9214b22006-11-16 02:41:18 -08001529}
1530
Julius Volz0bbdd422008-09-02 15:55:42 +02001531#ifdef CONFIG_IP_VS_IPV6
1532static inline __wsum ip_vs_check_diff16(const __be32 *old, const __be32 *new,
1533 __wsum oldsum)
1534{
1535 __be32 diff[8] = { ~old[3], ~old[2], ~old[1], ~old[0],
1536 new[3], new[2], new[1], new[0] };
1537
Joe Perches07f07572008-11-19 15:44:53 -08001538 return csum_partial(diff, sizeof(diff), oldsum);
Julius Volz0bbdd422008-09-02 15:55:42 +02001539}
1540#endif
1541
Al Virof9214b22006-11-16 02:41:18 -08001542static inline __wsum ip_vs_check_diff2(__be16 old, __be16 new, __wsum oldsum)
1543{
1544 __be16 diff[2] = { ~old, new };
1545
Joe Perches07f07572008-11-19 15:44:53 -08001546 return csum_partial(diff, sizeof(diff), oldsum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547}
1548
Simon Horman07dcc682014-09-30 10:50:06 +09001549/* Forget current conntrack (unconfirmed) and attach notrack entry */
Julian Anastasovcf356d62010-10-17 16:21:07 +03001550static inline void ip_vs_notrack(struct sk_buff *skb)
1551{
1552#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1553 enum ip_conntrack_info ctinfo;
Jesper Juhl06b69392011-03-09 22:55:05 +01001554 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
Julian Anastasovcf356d62010-10-17 16:21:07 +03001555
1556 if (!ct || !nf_ct_is_untracked(ct)) {
Lin Ming9e33ce42012-07-07 18:26:10 +08001557 nf_conntrack_put(skb->nfct);
Julian Anastasovcf356d62010-10-17 16:21:07 +03001558 skb->nfct = &nf_ct_untracked_get()->ct_general;
1559 skb->nfctinfo = IP_CT_NEW;
1560 nf_conntrack_get(skb->nfct);
1561 }
1562#endif
1563}
1564
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001565#ifdef CONFIG_IP_VS_NFCT
Simon Horman07dcc682014-09-30 10:50:06 +09001566/* Netfilter connection tracking
1567 * (from ip_vs_nfct.c)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001568 */
Hans Schillstroma0840e22011-01-03 14:44:58 +01001569static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001570{
Simon Hormana4e2f5a2011-02-04 18:33:02 +09001571#ifdef CONFIG_SYSCTL
Hans Schillstroma0840e22011-01-03 14:44:58 +01001572 return ipvs->sysctl_conntrack;
Simon Hormana4e2f5a2011-02-04 18:33:02 +09001573#else
1574 return 0;
1575#endif
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001576}
1577
Joe Perches5c3a0fd2013-09-21 10:22:42 -07001578void ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp,
1579 int outin);
1580int ip_vs_confirm_conntrack(struct sk_buff *skb);
1581void ip_vs_nfct_expect_related(struct sk_buff *skb, struct nf_conn *ct,
1582 struct ip_vs_conn *cp, u_int8_t proto,
1583 const __be16 port, int from_rs);
1584void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp);
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001585
1586#else
1587
Hans Schillstroma0840e22011-01-03 14:44:58 +01001588static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001589{
1590 return 0;
1591}
1592
1593static inline void ip_vs_update_conntrack(struct sk_buff *skb,
1594 struct ip_vs_conn *cp, int outin)
1595{
1596}
1597
Krzysztof Wilczynskie23ebf02011-10-20 13:18:04 +01001598static inline int ip_vs_confirm_conntrack(struct sk_buff *skb)
Julian Anastasovf4bc17c2010-09-21 17:35:41 +02001599{
1600 return NF_ACCEPT;
1601}
1602
1603static inline void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp)
1604{
1605}
Simon Horman07dcc682014-09-30 10:50:06 +09001606#endif /* CONFIG_IP_VS_NFCT */
Julian Anastasov6523ce12010-09-05 18:02:29 +00001607
Julian Anastasovf719e372016-03-05 15:03:22 +02001608/* Really using conntrack? */
1609static inline bool ip_vs_conn_uses_conntrack(struct ip_vs_conn *cp,
1610 struct sk_buff *skb)
1611{
1612#ifdef CONFIG_IP_VS_NFCT
1613 enum ip_conntrack_info ctinfo;
1614 struct nf_conn *ct;
1615
1616 if (!(cp->flags & IP_VS_CONN_F_NFCT))
1617 return false;
1618 ct = nf_ct_get(skb, &ctinfo);
1619 if (ct && !nf_ct_is_untracked(ct))
1620 return true;
1621#endif
1622 return false;
1623}
1624
Simon Kirbyc16526a2013-08-10 01:26:18 -07001625static inline int
Changli Gaob552f7e2011-02-19 17:32:28 +08001626ip_vs_dest_conn_overhead(struct ip_vs_dest *dest)
1627{
Simon Horman07dcc682014-09-30 10:50:06 +09001628 /* We think the overhead of processing active connections is 256
Changli Gaob552f7e2011-02-19 17:32:28 +08001629 * times higher than that of inactive connections in average. (This
1630 * 256 times might not be accurate, we will change it later) We
1631 * use the following formula to estimate the overhead now:
1632 * dest->activeconns*256 + dest->inactconns
1633 */
1634 return (atomic_read(&dest->activeconns) << 8) +
1635 atomic_read(&dest->inactconns);
1636}
1637
Julius Volzbc4768e2008-07-31 20:45:24 -07001638#endif /* _NET_IP_VS_H */