]>
Commit | Line | Data |
---|---|---|
07dcc686 SH |
1 | /* IP Virtual Server |
2 | * data structure and functionality definitions | |
1da177e4 LT |
3 | */ |
4 | ||
bc4768eb JV |
5 | #ifndef _NET_IP_VS_H |
6 | #define _NET_IP_VS_H | |
1da177e4 | 7 | |
bc4768eb | 8 | #include <linux/ip_vs.h> /* definitions shared with userland */ |
1da177e4 | 9 | |
bc4768eb JV |
10 | #include <asm/types.h> /* for __uXX types */ |
11 | ||
1da177e4 LT |
12 | #include <linux/list.h> /* for struct list_head */ |
13 | #include <linux/spinlock.h> /* for struct rwlock_t */ | |
07dcc686 | 14 | #include <linux/atomic.h> /* for struct atomic_t */ |
b54ab92b RE |
15 | #include <linux/refcount.h> /* for struct refcount_t */ |
16 | ||
1da177e4 | 17 | #include <linux/compiler.h> |
14c85021 | 18 | #include <linux/timer.h> |
187f1882 | 19 | #include <linux/bug.h> |
1da177e4 | 20 | |
14c85021 | 21 | #include <net/checksum.h> |
e7ade46a | 22 | #include <linux/netfilter.h> /* for union nf_inet_addr */ |
1668e010 | 23 | #include <linux/ip.h> |
e7ade46a | 24 | #include <linux/ipv6.h> /* for struct ipv6hdr */ |
4e3fd7a0 | 25 | #include <net/ipv6.h> |
54d83efa | 26 | #if IS_ENABLED(CONFIG_IP_VS_IPV6) |
63dca2c0 JDB |
27 | #include <linux/netfilter_ipv6/ip6_tables.h> |
28 | #endif | |
a638e514 | 29 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
f4bc17cd JA |
30 | #include <net/netfilter/nf_conntrack.h> |
31 | #endif | |
61b1ab45 HS |
32 | #include <net/net_namespace.h> /* Netw namespace */ |
33 | ||
4fd9beef AG |
34 | #define IP_VS_HDR_INVERSE 1 |
35 | #define IP_VS_HDR_ICMP 2 | |
36 | ||
07dcc686 | 37 | /* Generic access of ipvs struct */ |
61b1ab45 HS |
38 | static inline struct netns_ipvs *net_ipvs(struct net* net) |
39 | { | |
40 | return net->ipvs; | |
41 | } | |
07dcc686 | 42 | |
07dcc686 | 43 | /* This one needed for single_open_net since net is stored directly in |
25985edc | 44 | * private not as a struct i.e. seq_file_net can't be used. |
b17fc996 HS |
45 | */ |
46 | static inline struct net *seq_file_single_net(struct seq_file *seq) | |
47 | { | |
48 | #ifdef CONFIG_NET_NS | |
49 | return (struct net *)seq->private; | |
50 | #else | |
51 | return &init_net; | |
52 | #endif | |
53 | } | |
6f7edb48 CB |
54 | |
55 | /* Connections' size value needed by ip_vs_ctl.c */ | |
56 | extern int ip_vs_conn_tab_size; | |
57 | ||
64aae3cb | 58 | struct ip_vs_iphdr { |
4fd9beef | 59 | int hdr_flags; /* ipvs flags */ |
b0e010c5 | 60 | __u32 off; /* Where IP or IPv4 header starts */ |
63dca2c0 | 61 | __u32 len; /* IPv4 simply where L4 starts |
07dcc686 | 62 | * IPv6 where L4 Transport Header starts */ |
63dca2c0 JDB |
63 | __u16 fragoffs; /* IPv6 fragment offset, 0 if first frag (or not frag)*/ |
64 | __s16 protocol; | |
65 | __s32 flags; | |
64aae3cb JV |
66 | union nf_inet_addr saddr; |
67 | union nf_inet_addr daddr; | |
68 | }; | |
69 | ||
2f74713d JDB |
70 | static inline void *frag_safe_skb_hp(const struct sk_buff *skb, int offset, |
71 | int len, void *buffer, | |
72 | const struct ip_vs_iphdr *ipvsh) | |
73 | { | |
74 | return skb_header_pointer(skb, offset, len, buffer); | |
75 | } | |
2f74713d | 76 | |
63dca2c0 JDB |
77 | /* This function handles filling *ip_vs_iphdr, both for IPv4 and IPv6. |
78 | * IPv6 requires some extra work, as finding proper header position, | |
79 | * depend on the IPv6 extension headers. | |
80 | */ | |
b0e010c5 AG |
81 | static inline int |
82 | ip_vs_fill_iph_skb_off(int af, const struct sk_buff *skb, int offset, | |
4fd9beef | 83 | int hdr_flags, struct ip_vs_iphdr *iphdr) |
64aae3cb | 84 | { |
4fd9beef | 85 | iphdr->hdr_flags = hdr_flags; |
b0e010c5 | 86 | iphdr->off = offset; |
4fd9beef | 87 | |
64aae3cb JV |
88 | #ifdef CONFIG_IP_VS_IPV6 |
89 | if (af == AF_INET6) { | |
b0e010c5 AG |
90 | struct ipv6hdr _iph; |
91 | const struct ipv6hdr *iph = skb_header_pointer( | |
92 | skb, offset, sizeof(_iph), &_iph); | |
93 | if (!iph) | |
94 | return 0; | |
95 | ||
4e3fd7a0 AD |
96 | iphdr->saddr.in6 = iph->saddr; |
97 | iphdr->daddr.in6 = iph->daddr; | |
6aafeef0 | 98 | /* ipv6_find_hdr() updates len, flags */ |
b0e010c5 | 99 | iphdr->len = offset; |
63dca2c0 JDB |
100 | iphdr->flags = 0; |
101 | iphdr->protocol = ipv6_find_hdr(skb, &iphdr->len, -1, | |
102 | &iphdr->fragoffs, | |
103 | &iphdr->flags); | |
b0e010c5 AG |
104 | if (iphdr->protocol < 0) |
105 | return 0; | |
64aae3cb JV |
106 | } else |
107 | #endif | |
108 | { | |
b0e010c5 AG |
109 | struct iphdr _iph; |
110 | const struct iphdr *iph = skb_header_pointer( | |
111 | skb, offset, sizeof(_iph), &_iph); | |
112 | if (!iph) | |
113 | return 0; | |
114 | ||
115 | iphdr->len = offset + iph->ihl * 4; | |
63dca2c0 JDB |
116 | iphdr->fragoffs = 0; |
117 | iphdr->protocol = iph->protocol; | |
118 | iphdr->saddr.ip = iph->saddr; | |
119 | iphdr->daddr.ip = iph->daddr; | |
120 | } | |
b0e010c5 AG |
121 | |
122 | return 1; | |
123 | } | |
124 | ||
125 | static inline int | |
4fd9beef AG |
126 | ip_vs_fill_iph_skb_icmp(int af, const struct sk_buff *skb, int offset, |
127 | bool inverse, struct ip_vs_iphdr *iphdr) | |
128 | { | |
129 | int hdr_flags = IP_VS_HDR_ICMP; | |
130 | ||
131 | if (inverse) | |
132 | hdr_flags |= IP_VS_HDR_INVERSE; | |
133 | ||
134 | return ip_vs_fill_iph_skb_off(af, skb, offset, hdr_flags, iphdr); | |
135 | } | |
136 | ||
137 | static inline int | |
138 | ip_vs_fill_iph_skb(int af, const struct sk_buff *skb, bool inverse, | |
139 | struct ip_vs_iphdr *iphdr) | |
140 | { | |
141 | int hdr_flags = 0; | |
142 | ||
143 | if (inverse) | |
144 | hdr_flags |= IP_VS_HDR_INVERSE; | |
145 | ||
146 | return ip_vs_fill_iph_skb_off(af, skb, skb_network_offset(skb), | |
147 | hdr_flags, iphdr); | |
148 | } | |
149 | ||
150 | static inline bool | |
151 | ip_vs_iph_inverse(const struct ip_vs_iphdr *iph) | |
152 | { | |
153 | return !!(iph->hdr_flags & IP_VS_HDR_INVERSE); | |
154 | } | |
155 | ||
156 | static inline bool | |
157 | ip_vs_iph_icmp(const struct ip_vs_iphdr *iph) | |
b0e010c5 | 158 | { |
4fd9beef | 159 | return !!(iph->hdr_flags & IP_VS_HDR_ICMP); |
63dca2c0 JDB |
160 | } |
161 | ||
64aae3cb JV |
162 | static inline void ip_vs_addr_copy(int af, union nf_inet_addr *dst, |
163 | const union nf_inet_addr *src) | |
164 | { | |
165 | #ifdef CONFIG_IP_VS_IPV6 | |
166 | if (af == AF_INET6) | |
4e3fd7a0 | 167 | dst->in6 = src->in6; |
64aae3cb JV |
168 | else |
169 | #endif | |
170 | dst->ip = src->ip; | |
171 | } | |
172 | ||
9a05475c JA |
173 | static inline void ip_vs_addr_set(int af, union nf_inet_addr *dst, |
174 | const union nf_inet_addr *src) | |
175 | { | |
176 | #ifdef CONFIG_IP_VS_IPV6 | |
177 | if (af == AF_INET6) { | |
178 | dst->in6 = src->in6; | |
179 | return; | |
180 | } | |
181 | #endif | |
182 | dst->ip = src->ip; | |
183 | dst->all[1] = 0; | |
184 | dst->all[2] = 0; | |
185 | dst->all[3] = 0; | |
186 | } | |
187 | ||
64aae3cb JV |
188 | static inline int ip_vs_addr_equal(int af, const union nf_inet_addr *a, |
189 | const union nf_inet_addr *b) | |
190 | { | |
191 | #ifdef CONFIG_IP_VS_IPV6 | |
192 | if (af == AF_INET6) | |
193 | return ipv6_addr_equal(&a->in6, &b->in6); | |
194 | #endif | |
195 | return a->ip == b->ip; | |
196 | } | |
197 | ||
1da177e4 | 198 | #ifdef CONFIG_IP_VS_DEBUG |
14c85021 ACM |
199 | #include <linux/net.h> |
200 | ||
5c3a0fd7 | 201 | int ip_vs_get_debug_level(void); |
c842a3ad JV |
202 | |
203 | static inline const char *ip_vs_dbg_addr(int af, char *buf, size_t buf_len, | |
204 | const union nf_inet_addr *addr, | |
205 | int *idx) | |
206 | { | |
207 | int len; | |
208 | #ifdef CONFIG_IP_VS_IPV6 | |
209 | if (af == AF_INET6) | |
120b9c14 | 210 | len = snprintf(&buf[*idx], buf_len - *idx, "[%pI6c]", |
0c6ce78a | 211 | &addr->in6) + 1; |
c842a3ad JV |
212 | else |
213 | #endif | |
3685f25d HH |
214 | len = snprintf(&buf[*idx], buf_len - *idx, "%pI4", |
215 | &addr->ip) + 1; | |
c842a3ad JV |
216 | |
217 | *idx += len; | |
218 | BUG_ON(*idx > buf_len + 1); | |
219 | return &buf[*idx - len]; | |
220 | } | |
221 | ||
9aada7ac HE |
222 | #define IP_VS_DBG_BUF(level, msg, ...) \ |
223 | do { \ | |
224 | char ip_vs_dbg_buf[160]; \ | |
225 | int ip_vs_dbg_idx = 0; \ | |
226 | if (level <= ip_vs_get_debug_level()) \ | |
227 | printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \ | |
228 | } while (0) | |
229 | #define IP_VS_ERR_BUF(msg...) \ | |
230 | do { \ | |
231 | char ip_vs_dbg_buf[160]; \ | |
232 | int ip_vs_dbg_idx = 0; \ | |
233 | pr_err(msg); \ | |
234 | } while (0) | |
c842a3ad JV |
235 | |
236 | /* Only use from within IP_VS_DBG_BUF() or IP_VS_ERR_BUF macros */ | |
9aada7ac HE |
237 | #define IP_VS_DBG_ADDR(af, addr) \ |
238 | ip_vs_dbg_addr(af, ip_vs_dbg_buf, \ | |
239 | sizeof(ip_vs_dbg_buf), addr, \ | |
240 | &ip_vs_dbg_idx) | |
241 | ||
242 | #define IP_VS_DBG(level, msg, ...) \ | |
243 | do { \ | |
244 | if (level <= ip_vs_get_debug_level()) \ | |
245 | printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \ | |
246 | } while (0) | |
247 | #define IP_VS_DBG_RL(msg, ...) \ | |
248 | do { \ | |
249 | if (net_ratelimit()) \ | |
250 | printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \ | |
251 | } while (0) | |
0d79641a | 252 | #define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) \ |
9aada7ac HE |
253 | do { \ |
254 | if (level <= ip_vs_get_debug_level()) \ | |
0d79641a | 255 | pp->debug_packet(af, pp, skb, ofs, msg); \ |
9aada7ac | 256 | } while (0) |
0d79641a | 257 | #define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) \ |
9aada7ac HE |
258 | do { \ |
259 | if (level <= ip_vs_get_debug_level() && \ | |
260 | net_ratelimit()) \ | |
0d79641a | 261 | pp->debug_packet(af, pp, skb, ofs, msg); \ |
9aada7ac | 262 | } while (0) |
1da177e4 | 263 | #else /* NO DEBUGGING at ALL */ |
c842a3ad JV |
264 | #define IP_VS_DBG_BUF(level, msg...) do {} while (0) |
265 | #define IP_VS_ERR_BUF(msg...) do {} while (0) | |
1da177e4 LT |
266 | #define IP_VS_DBG(level, msg...) do {} while (0) |
267 | #define IP_VS_DBG_RL(msg...) do {} while (0) | |
0d79641a JA |
268 | #define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) do {} while (0) |
269 | #define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) do {} while (0) | |
1da177e4 LT |
270 | #endif |
271 | ||
272 | #define IP_VS_BUG() BUG() | |
1e3e238e | 273 | #define IP_VS_ERR_RL(msg, ...) \ |
9aada7ac HE |
274 | do { \ |
275 | if (net_ratelimit()) \ | |
1e3e238e | 276 | pr_err(msg, ##__VA_ARGS__); \ |
9aada7ac | 277 | } while (0) |
1da177e4 LT |
278 | |
279 | #ifdef CONFIG_IP_VS_DEBUG | |
280 | #define EnterFunction(level) \ | |
9aada7ac HE |
281 | do { \ |
282 | if (level <= ip_vs_get_debug_level()) \ | |
283 | printk(KERN_DEBUG \ | |
284 | pr_fmt("Enter: %s, %s line %i\n"), \ | |
285 | __func__, __FILE__, __LINE__); \ | |
286 | } while (0) | |
287 | #define LeaveFunction(level) \ | |
288 | do { \ | |
289 | if (level <= ip_vs_get_debug_level()) \ | |
290 | printk(KERN_DEBUG \ | |
291 | pr_fmt("Leave: %s, %s line %i\n"), \ | |
292 | __func__, __FILE__, __LINE__); \ | |
293 | } while (0) | |
1da177e4 LT |
294 | #else |
295 | #define EnterFunction(level) do {} while (0) | |
296 | #define LeaveFunction(level) do {} while (0) | |
297 | #endif | |
298 | ||
07dcc686 | 299 | /* The port number of FTP service (in network order). */ |
f3a7c66b HH |
300 | #define FTPPORT cpu_to_be16(21) |
301 | #define FTPDATA cpu_to_be16(20) | |
1da177e4 | 302 | |
07dcc686 | 303 | /* TCP State Values */ |
1da177e4 LT |
304 | enum { |
305 | IP_VS_TCP_S_NONE = 0, | |
306 | IP_VS_TCP_S_ESTABLISHED, | |
307 | IP_VS_TCP_S_SYN_SENT, | |
308 | IP_VS_TCP_S_SYN_RECV, | |
309 | IP_VS_TCP_S_FIN_WAIT, | |
310 | IP_VS_TCP_S_TIME_WAIT, | |
311 | IP_VS_TCP_S_CLOSE, | |
312 | IP_VS_TCP_S_CLOSE_WAIT, | |
313 | IP_VS_TCP_S_LAST_ACK, | |
314 | IP_VS_TCP_S_LISTEN, | |
315 | IP_VS_TCP_S_SYNACK, | |
316 | IP_VS_TCP_S_LAST | |
317 | }; | |
318 | ||
07dcc686 | 319 | /* UDP State Values */ |
1da177e4 LT |
320 | enum { |
321 | IP_VS_UDP_S_NORMAL, | |
322 | IP_VS_UDP_S_LAST, | |
323 | }; | |
324 | ||
07dcc686 | 325 | /* ICMP State Values */ |
1da177e4 LT |
326 | enum { |
327 | IP_VS_ICMP_S_NORMAL, | |
328 | IP_VS_ICMP_S_LAST, | |
329 | }; | |
330 | ||
07dcc686 | 331 | /* SCTP State Values */ |
2906f66a VMR |
332 | enum ip_vs_sctp_states { |
333 | IP_VS_SCTP_S_NONE, | |
61e7c420 JA |
334 | IP_VS_SCTP_S_INIT1, |
335 | IP_VS_SCTP_S_INIT, | |
336 | IP_VS_SCTP_S_COOKIE_SENT, | |
337 | IP_VS_SCTP_S_COOKIE_REPLIED, | |
338 | IP_VS_SCTP_S_COOKIE_WAIT, | |
339 | IP_VS_SCTP_S_COOKIE, | |
340 | IP_VS_SCTP_S_COOKIE_ECHOED, | |
2906f66a | 341 | IP_VS_SCTP_S_ESTABLISHED, |
61e7c420 JA |
342 | IP_VS_SCTP_S_SHUTDOWN_SENT, |
343 | IP_VS_SCTP_S_SHUTDOWN_RECEIVED, | |
344 | IP_VS_SCTP_S_SHUTDOWN_ACK_SENT, | |
345 | IP_VS_SCTP_S_REJECTED, | |
2906f66a VMR |
346 | IP_VS_SCTP_S_CLOSED, |
347 | IP_VS_SCTP_S_LAST | |
348 | }; | |
349 | ||
07dcc686 SH |
350 | /* Delta sequence info structure |
351 | * Each ip_vs_conn has 2 (output AND input seq. changes). | |
352 | * Only used in the VS/NAT. | |
1da177e4 LT |
353 | */ |
354 | struct ip_vs_seq { | |
355 | __u32 init_seq; /* Add delta from this seq */ | |
356 | __u32 delta; /* Delta in sequence numbers */ | |
357 | __u32 previous_delta; /* Delta in sequence numbers | |
07dcc686 | 358 | * before last resized pkt */ |
1da177e4 LT |
359 | }; |
360 | ||
07dcc686 | 361 | /* counters per cpu */ |
b17fc996 | 362 | struct ip_vs_counters { |
cd67cd5e JA |
363 | __u64 conns; /* connections scheduled */ |
364 | __u64 inpkts; /* incoming packets */ | |
365 | __u64 outpkts; /* outgoing packets */ | |
b17fc996 HS |
366 | __u64 inbytes; /* incoming bytes */ |
367 | __u64 outbytes; /* outgoing bytes */ | |
368 | }; | |
07dcc686 | 369 | /* Stats per cpu */ |
b17fc996 | 370 | struct ip_vs_cpu_stats { |
cd67cd5e | 371 | struct ip_vs_counters cnt; |
b17fc996 HS |
372 | struct u64_stats_sync syncp; |
373 | }; | |
1da177e4 | 374 | |
07dcc686 | 375 | /* IPVS statistics objects */ |
3a14a313 SW |
376 | struct ip_vs_estimator { |
377 | struct list_head list; | |
378 | ||
379 | u64 last_inbytes; | |
380 | u64 last_outbytes; | |
cd67cd5e JA |
381 | u64 last_conns; |
382 | u64 last_inpkts; | |
383 | u64 last_outpkts; | |
384 | ||
385 | u64 cps; | |
386 | u64 inpps; | |
387 | u64 outpps; | |
388 | u64 inbps; | |
389 | u64 outbps; | |
390 | }; | |
391 | ||
392 | /* | |
393 | * IPVS statistics object, 64-bit kernel version of struct ip_vs_stats_user | |
394 | */ | |
395 | struct ip_vs_kstats { | |
396 | u64 conns; /* connections scheduled */ | |
397 | u64 inpkts; /* incoming packets */ | |
398 | u64 outpkts; /* outgoing packets */ | |
399 | u64 inbytes; /* incoming bytes */ | |
400 | u64 outbytes; /* outgoing bytes */ | |
401 | ||
402 | u64 cps; /* current connection rate */ | |
403 | u64 inpps; /* current in packet rate */ | |
404 | u64 outpps; /* current out packet rate */ | |
405 | u64 inbps; /* current in byte rate */ | |
406 | u64 outbps; /* current out byte rate */ | |
3a14a313 SW |
407 | }; |
408 | ||
fd2c3ef7 | 409 | struct ip_vs_stats { |
cd67cd5e | 410 | struct ip_vs_kstats kstats; /* kernel statistics */ |
e9c0ce23 | 411 | struct ip_vs_estimator est; /* estimator */ |
b962abdc | 412 | struct ip_vs_cpu_stats __percpu *cpustats; /* per cpu counters */ |
b17fc996 | 413 | spinlock_t lock; /* spin lock */ |
cd67cd5e | 414 | struct ip_vs_kstats kstats0; /* reset values */ |
1da177e4 LT |
415 | }; |
416 | ||
14c85021 ACM |
417 | struct dst_entry; |
418 | struct iphdr; | |
1da177e4 LT |
419 | struct ip_vs_conn; |
420 | struct ip_vs_app; | |
14c85021 | 421 | struct sk_buff; |
252c6410 | 422 | struct ip_vs_proto_data; |
1da177e4 LT |
423 | |
424 | struct ip_vs_protocol { | |
425 | struct ip_vs_protocol *next; | |
426 | char *name; | |
2ad17def JA |
427 | u16 protocol; |
428 | u16 num_states; | |
1da177e4 | 429 | int dont_defrag; |
1da177e4 LT |
430 | |
431 | void (*init)(struct ip_vs_protocol *pp); | |
432 | ||
433 | void (*exit)(struct ip_vs_protocol *pp); | |
434 | ||
1281a9c2 | 435 | int (*init_netns)(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd); |
252c6410 | 436 | |
1281a9c2 | 437 | void (*exit_netns)(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd); |
252c6410 | 438 | |
d8f44c33 EB |
439 | int (*conn_schedule)(struct netns_ipvs *ipvs, |
440 | int af, struct sk_buff *skb, | |
9330419d | 441 | struct ip_vs_proto_data *pd, |
d4383f04 JDB |
442 | int *verdict, struct ip_vs_conn **cpp, |
443 | struct ip_vs_iphdr *iph); | |
1da177e4 LT |
444 | |
445 | struct ip_vs_conn * | |
ab161976 EB |
446 | (*conn_in_get)(struct netns_ipvs *ipvs, |
447 | int af, | |
51ef348b | 448 | const struct sk_buff *skb, |
802c41ad | 449 | const struct ip_vs_iphdr *iph); |
1da177e4 LT |
450 | |
451 | struct ip_vs_conn * | |
0cf705c8 EB |
452 | (*conn_out_get)(struct netns_ipvs *ipvs, |
453 | int af, | |
51ef348b | 454 | const struct sk_buff *skb, |
802c41ad | 455 | const struct ip_vs_iphdr *iph); |
1da177e4 | 456 | |
d4383f04 JDB |
457 | int (*snat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp, |
458 | struct ip_vs_conn *cp, struct ip_vs_iphdr *iph); | |
1da177e4 | 459 | |
d4383f04 JDB |
460 | int (*dnat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp, |
461 | struct ip_vs_conn *cp, struct ip_vs_iphdr *iph); | |
1da177e4 | 462 | |
51ef348b JV |
463 | int (*csum_check)(int af, struct sk_buff *skb, |
464 | struct ip_vs_protocol *pp); | |
1da177e4 LT |
465 | |
466 | const char *(*state_name)(int state); | |
467 | ||
4a516f11 SH |
468 | void (*state_transition)(struct ip_vs_conn *cp, int direction, |
469 | const struct sk_buff *skb, | |
470 | struct ip_vs_proto_data *pd); | |
1da177e4 | 471 | |
19648918 | 472 | int (*register_app)(struct netns_ipvs *ipvs, struct ip_vs_app *inc); |
1da177e4 | 473 | |
19648918 | 474 | void (*unregister_app)(struct netns_ipvs *ipvs, struct ip_vs_app *inc); |
1da177e4 LT |
475 | |
476 | int (*app_conn_bind)(struct ip_vs_conn *cp); | |
477 | ||
0d79641a | 478 | void (*debug_packet)(int af, struct ip_vs_protocol *pp, |
1da177e4 LT |
479 | const struct sk_buff *skb, |
480 | int offset, | |
481 | const char *msg); | |
482 | ||
9330419d | 483 | void (*timeout_change)(struct ip_vs_proto_data *pd, int flags); |
1da177e4 LT |
484 | }; |
485 | ||
07dcc686 | 486 | /* protocol data per netns */ |
252c6410 HS |
487 | struct ip_vs_proto_data { |
488 | struct ip_vs_proto_data *next; | |
489 | struct ip_vs_protocol *pp; | |
490 | int *timeout_table; /* protocol timeout table */ | |
491 | atomic_t appcnt; /* counter of proto app incs. */ | |
492 | struct tcp_states_t *tcp_state_table; | |
493 | }; | |
494 | ||
5c3a0fd7 | 495 | struct ip_vs_protocol *ip_vs_proto_get(unsigned short proto); |
18d6ade6 | 496 | struct ip_vs_proto_data *ip_vs_proto_data_get(struct netns_ipvs *ipvs, |
5c3a0fd7 | 497 | unsigned short proto); |
1da177e4 | 498 | |
f11017ec | 499 | struct ip_vs_conn_param { |
e64e2b46 | 500 | struct netns_ipvs *ipvs; |
f11017ec SH |
501 | const union nf_inet_addr *caddr; |
502 | const union nf_inet_addr *vaddr; | |
503 | __be16 cport; | |
504 | __be16 vport; | |
505 | __u16 protocol; | |
506 | u16 af; | |
85999283 SH |
507 | |
508 | const struct ip_vs_pe *pe; | |
509 | char *pe_data; | |
510 | __u8 pe_data_len; | |
f11017ec SH |
511 | }; |
512 | ||
07dcc686 | 513 | /* IP_VS structure allocated for each dynamically scheduled connection */ |
1da177e4 | 514 | struct ip_vs_conn { |
731109e7 | 515 | struct hlist_node c_list; /* hashed list heads */ |
1da177e4 | 516 | /* Protocol, addresses and port numbers */ |
6e67e586 | 517 | __be16 cport; |
6e67e586 | 518 | __be16 dport; |
1845ed0b JA |
519 | __be16 vport; |
520 | u16 af; /* address family */ | |
6e67e586 HS |
521 | union nf_inet_addr caddr; /* client address */ |
522 | union nf_inet_addr vaddr; /* virtual address */ | |
523 | union nf_inet_addr daddr; /* destination address */ | |
524 | volatile __u32 flags; /* status flags */ | |
1da177e4 | 525 | __u16 protocol; /* Which protocol (TCP/UDP) */ |
ba38528a | 526 | __u16 daf; /* Address family of the dest */ |
58dbc6f2 | 527 | struct netns_ipvs *ipvs; |
1da177e4 LT |
528 | |
529 | /* counter and timer */ | |
b54ab92b | 530 | refcount_t refcnt; /* reference count */ |
1da177e4 LT |
531 | struct timer_list timer; /* Expiration timer */ |
532 | volatile unsigned long timeout; /* timeout */ | |
533 | ||
534 | /* Flags and state transition */ | |
535 | spinlock_t lock; /* lock for state transition */ | |
1da177e4 | 536 | volatile __u16 state; /* state info */ |
efac5276 RB |
537 | volatile __u16 old_state; /* old state, to be used for |
538 | * state transition triggerd | |
539 | * synchronization | |
540 | */ | |
1845ed0b | 541 | __u32 fwmark; /* Fire wall mark from skb */ |
749c42b6 | 542 | unsigned long sync_endtime; /* jiffies + sent_retries */ |
1da177e4 LT |
543 | |
544 | /* Control members */ | |
545 | struct ip_vs_conn *control; /* Master control connection */ | |
546 | atomic_t n_control; /* Number of controlled ones */ | |
547 | struct ip_vs_dest *dest; /* real server */ | |
548 | atomic_t in_pkts; /* incoming packet counter */ | |
549 | ||
07dcc686 SH |
550 | /* Packet transmitter for different forwarding methods. If it |
551 | * mangles the packet, it must return NF_DROP or better NF_STOLEN, | |
552 | * otherwise this must be changed to a sk_buff **. | |
553 | * NF_ACCEPT can be returned when destination is local. | |
1da177e4 LT |
554 | */ |
555 | int (*packet_xmit)(struct sk_buff *skb, struct ip_vs_conn *cp, | |
d4383f04 | 556 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); |
1da177e4 LT |
557 | |
558 | /* Note: we can group the following members into a structure, | |
07dcc686 SH |
559 | * in order to save more space, and the following members are |
560 | * only used in VS/NAT anyway | |
561 | */ | |
1da177e4 LT |
562 | struct ip_vs_app *app; /* bound ip_vs_app object */ |
563 | void *app_data; /* Application private data */ | |
564 | struct ip_vs_seq in_seq; /* incoming seq. struct */ | |
565 | struct ip_vs_seq out_seq; /* outgoing seq. struct */ | |
85999283 | 566 | |
e9e5eee8 | 567 | const struct ip_vs_pe *pe; |
85999283 SH |
568 | char *pe_data; |
569 | __u8 pe_data_len; | |
088339a5 JA |
570 | |
571 | struct rcu_head rcu_head; | |
1da177e4 LT |
572 | }; |
573 | ||
07dcc686 SH |
574 | /* Extended internal versions of struct ip_vs_service_user and ip_vs_dest_user |
575 | * for IPv6 support. | |
c860c6b1 | 576 | * |
07dcc686 SH |
577 | * We need these to conveniently pass around service and destination |
578 | * options, but unfortunately, we also need to keep the old definitions to | |
579 | * maintain userspace backwards compatibility for the setsockopt interface. | |
c860c6b1 JV |
580 | */ |
581 | struct ip_vs_service_user_kern { | |
582 | /* virtual service addresses */ | |
583 | u16 af; | |
584 | u16 protocol; | |
585 | union nf_inet_addr addr; /* virtual ip address */ | |
0a925864 | 586 | __be16 port; |
c860c6b1 JV |
587 | u32 fwmark; /* firwall mark of service */ |
588 | ||
589 | /* virtual service options */ | |
590 | char *sched_name; | |
0d1e71b0 | 591 | char *pe_name; |
95c96174 ED |
592 | unsigned int flags; /* virtual service flags */ |
593 | unsigned int timeout; /* persistent timeout in sec */ | |
0a925864 | 594 | __be32 netmask; /* persistent netmask or plen */ |
c860c6b1 JV |
595 | }; |
596 | ||
597 | ||
598 | struct ip_vs_dest_user_kern { | |
599 | /* destination server address */ | |
600 | union nf_inet_addr addr; | |
0a925864 | 601 | __be16 port; |
c860c6b1 JV |
602 | |
603 | /* real server options */ | |
95c96174 | 604 | unsigned int conn_flags; /* connection flags */ |
c860c6b1 JV |
605 | int weight; /* destination weight */ |
606 | ||
607 | /* thresholds for active connections */ | |
608 | u32 u_threshold; /* upper threshold */ | |
609 | u32 l_threshold; /* lower threshold */ | |
6cff339b AG |
610 | |
611 | /* Address family of addr */ | |
612 | u16 af; | |
c860c6b1 JV |
613 | }; |
614 | ||
615 | ||
1da177e4 | 616 | /* |
07dcc686 SH |
617 | * The information about the virtual service offered to the net and the |
618 | * forwarding entries. | |
1da177e4 LT |
619 | */ |
620 | struct ip_vs_service { | |
ceec4c38 JA |
621 | struct hlist_node s_list; /* for normal service table */ |
622 | struct hlist_node f_list; /* for fwmark-based service table */ | |
1da177e4 | 623 | atomic_t refcnt; /* reference counter */ |
1da177e4 | 624 | |
e7ade46a | 625 | u16 af; /* address family */ |
1da177e4 | 626 | __u16 protocol; /* which protocol (TCP/UDP) */ |
e7ade46a | 627 | union nf_inet_addr addr; /* IP address for virtual service */ |
014d730d | 628 | __be16 port; /* port number for the service */ |
1da177e4 | 629 | __u32 fwmark; /* firewall mark of the service */ |
95c96174 ED |
630 | unsigned int flags; /* service status flags */ |
631 | unsigned int timeout; /* persistent timeout in ticks */ | |
0a925864 | 632 | __be32 netmask; /* grouping granularity, mask/plen */ |
3109d2f2 | 633 | struct netns_ipvs *ipvs; |
1da177e4 LT |
634 | |
635 | struct list_head destinations; /* real server d-linked list */ | |
636 | __u32 num_dests; /* number of servers */ | |
637 | struct ip_vs_stats stats; /* statistics for the service */ | |
1da177e4 LT |
638 | |
639 | /* for scheduling */ | |
ceec4c38 | 640 | struct ip_vs_scheduler __rcu *scheduler; /* bound scheduler object */ |
ba3a3ce1 | 641 | spinlock_t sched_lock; /* lock sched_data */ |
1da177e4 | 642 | void *sched_data; /* scheduler application data */ |
85999283 SH |
643 | |
644 | /* alternate persistence engine */ | |
ceec4c38 JA |
645 | struct ip_vs_pe __rcu *pe; |
646 | ||
647 | struct rcu_head rcu_head; | |
1da177e4 LT |
648 | }; |
649 | ||
026ace06 JA |
650 | /* Information for cached dst */ |
651 | struct ip_vs_dest_dst { | |
652 | struct dst_entry *dst_cache; /* destination cache entry */ | |
653 | u32 dst_cookie; | |
654 | union nf_inet_addr dst_saddr; | |
655 | struct rcu_head rcu_head; | |
656 | }; | |
1da177e4 | 657 | |
07dcc686 SH |
658 | /* The real server destination forwarding entry with ip address, port number, |
659 | * and so on. | |
1da177e4 LT |
660 | */ |
661 | struct ip_vs_dest { | |
662 | struct list_head n_list; /* for the dests in the service */ | |
276472ea | 663 | struct hlist_node d_list; /* for table with all the dests */ |
1da177e4 | 664 | |
e7ade46a | 665 | u16 af; /* address family */ |
014d730d | 666 | __be16 port; /* port number of the server */ |
f2431e6e | 667 | union nf_inet_addr addr; /* IP address of the server */ |
95c96174 | 668 | volatile unsigned int flags; /* dest status flags */ |
1da177e4 LT |
669 | atomic_t conn_flags; /* flags to copy to conn */ |
670 | atomic_t weight; /* server weight */ | |
671 | ||
b54ab92b | 672 | refcount_t refcnt; /* reference counter */ |
1da177e4 | 673 | struct ip_vs_stats stats; /* statistics */ |
bcbde4c0 | 674 | unsigned long idle_start; /* start time, jiffies */ |
1da177e4 LT |
675 | |
676 | /* connection counters and thresholds */ | |
677 | atomic_t activeconns; /* active connections */ | |
678 | atomic_t inactconns; /* inactive connections */ | |
679 | atomic_t persistconns; /* persistent connections */ | |
680 | __u32 u_threshold; /* upper threshold */ | |
681 | __u32 l_threshold; /* lower threshold */ | |
682 | ||
683 | /* for destination cache */ | |
684 | spinlock_t dst_lock; /* lock of dst_cache */ | |
026ace06 | 685 | struct ip_vs_dest_dst __rcu *dest_dst; /* cached dst info */ |
1da177e4 LT |
686 | |
687 | /* for virtual service */ | |
bcbde4c0 | 688 | struct ip_vs_service __rcu *svc; /* service it belongs to */ |
1da177e4 | 689 | __u16 protocol; /* which protocol (TCP/UDP) */ |
014d730d | 690 | __be16 vport; /* virtual port number */ |
f2431e6e | 691 | union nf_inet_addr vaddr; /* virtual IP address */ |
1da177e4 | 692 | __u32 vfwmark; /* firewall mark of service */ |
276472ea | 693 | |
578bc3ef | 694 | struct list_head t_list; /* in dest_trash */ |
276472ea | 695 | unsigned int in_rs_table:1; /* we are in rs_table */ |
1da177e4 LT |
696 | }; |
697 | ||
07dcc686 | 698 | /* The scheduler object */ |
1da177e4 LT |
699 | struct ip_vs_scheduler { |
700 | struct list_head n_list; /* d-linked list head */ | |
701 | char *name; /* scheduler name */ | |
702 | atomic_t refcnt; /* reference counter */ | |
703 | struct module *module; /* THIS_MODULE/NULL */ | |
704 | ||
705 | /* scheduler initializing service */ | |
706 | int (*init_service)(struct ip_vs_service *svc); | |
707 | /* scheduling service finish */ | |
ed3ffc4e | 708 | void (*done_service)(struct ip_vs_service *svc); |
6b6df466 JA |
709 | /* dest is linked */ |
710 | int (*add_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest); | |
711 | /* dest is unlinked */ | |
712 | int (*del_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest); | |
713 | /* dest is updated */ | |
714 | int (*upd_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest); | |
1da177e4 LT |
715 | |
716 | /* selecting a server from the given service */ | |
717 | struct ip_vs_dest* (*schedule)(struct ip_vs_service *svc, | |
bba54de5 JA |
718 | const struct sk_buff *skb, |
719 | struct ip_vs_iphdr *iph); | |
1da177e4 LT |
720 | }; |
721 | ||
85999283 SH |
722 | /* The persistence engine object */ |
723 | struct ip_vs_pe { | |
724 | struct list_head n_list; /* d-linked list head */ | |
725 | char *name; /* scheduler name */ | |
726 | atomic_t refcnt; /* reference counter */ | |
727 | struct module *module; /* THIS_MODULE/NULL */ | |
728 | ||
729 | /* get the connection template, if any */ | |
730 | int (*fill_param)(struct ip_vs_conn_param *p, struct sk_buff *skb); | |
731 | bool (*ct_match)(const struct ip_vs_conn_param *p, | |
732 | struct ip_vs_conn *ct); | |
733 | u32 (*hashkey_raw)(const struct ip_vs_conn_param *p, u32 initval, | |
734 | bool inverse); | |
a3c918ac | 735 | int (*show_pe_data)(const struct ip_vs_conn *cp, char *buf); |
39b97223 MA |
736 | /* create connections for real-server outgoing packets */ |
737 | struct ip_vs_conn* (*conn_out)(struct ip_vs_service *svc, | |
738 | struct ip_vs_dest *dest, | |
739 | struct sk_buff *skb, | |
740 | const struct ip_vs_iphdr *iph, | |
741 | __be16 dport, __be16 cport); | |
85999283 | 742 | }; |
1da177e4 | 743 | |
07dcc686 | 744 | /* The application module object (a.k.a. app incarnation) */ |
fd2c3ef7 | 745 | struct ip_vs_app { |
1da177e4 LT |
746 | struct list_head a_list; /* member in app list */ |
747 | int type; /* IP_VS_APP_TYPE_xxx */ | |
748 | char *name; /* application module name */ | |
749 | __u16 protocol; | |
750 | struct module *module; /* THIS_MODULE/NULL */ | |
751 | struct list_head incs_list; /* list of incarnations */ | |
752 | ||
753 | /* members for application incarnations */ | |
754 | struct list_head p_list; /* member in proto app list */ | |
755 | struct ip_vs_app *app; /* its real application */ | |
014d730d | 756 | __be16 port; /* port number in net order */ |
1da177e4 | 757 | atomic_t usecnt; /* usage counter */ |
363c97d7 | 758 | struct rcu_head rcu_head; |
1da177e4 | 759 | |
07dcc686 | 760 | /* output hook: Process packet in inout direction, diff set for TCP. |
8b27b10f JA |
761 | * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok, |
762 | * 2=Mangled but checksum was not updated | |
763 | */ | |
1da177e4 | 764 | int (*pkt_out)(struct ip_vs_app *, struct ip_vs_conn *, |
3db05fea | 765 | struct sk_buff *, int *diff); |
1da177e4 | 766 | |
07dcc686 | 767 | /* input hook: Process packet in outin direction, diff set for TCP. |
8b27b10f JA |
768 | * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok, |
769 | * 2=Mangled but checksum was not updated | |
770 | */ | |
1da177e4 | 771 | int (*pkt_in)(struct ip_vs_app *, struct ip_vs_conn *, |
3db05fea | 772 | struct sk_buff *, int *diff); |
1da177e4 LT |
773 | |
774 | /* ip_vs_app initializer */ | |
775 | int (*init_conn)(struct ip_vs_app *, struct ip_vs_conn *); | |
776 | ||
777 | /* ip_vs_app finish */ | |
778 | int (*done_conn)(struct ip_vs_app *, struct ip_vs_conn *); | |
779 | ||
780 | ||
781 | /* not used now */ | |
782 | int (*bind_conn)(struct ip_vs_app *, struct ip_vs_conn *, | |
783 | struct ip_vs_protocol *); | |
784 | ||
785 | void (*unbind_conn)(struct ip_vs_app *, struct ip_vs_conn *); | |
786 | ||
787 | int * timeout_table; | |
788 | int * timeouts; | |
789 | int timeouts_size; | |
790 | ||
791 | int (*conn_schedule)(struct sk_buff *skb, struct ip_vs_app *app, | |
792 | int *verdict, struct ip_vs_conn **cpp); | |
793 | ||
794 | struct ip_vs_conn * | |
795 | (*conn_in_get)(const struct sk_buff *skb, struct ip_vs_app *app, | |
d4383f04 | 796 | const struct iphdr *iph, int inverse); |
1da177e4 LT |
797 | |
798 | struct ip_vs_conn * | |
799 | (*conn_out_get)(const struct sk_buff *skb, struct ip_vs_app *app, | |
d4383f04 | 800 | const struct iphdr *iph, int inverse); |
1da177e4 LT |
801 | |
802 | int (*state_transition)(struct ip_vs_conn *cp, int direction, | |
803 | const struct sk_buff *skb, | |
804 | struct ip_vs_app *app); | |
805 | ||
806 | void (*timeout_change)(struct ip_vs_app *app, int flags); | |
807 | }; | |
808 | ||
f73181c8 PNA |
809 | struct ipvs_master_sync_state { |
810 | struct list_head sync_queue; | |
811 | struct ip_vs_sync_buff *sync_buff; | |
07995674 | 812 | unsigned long sync_queue_len; |
f73181c8 PNA |
813 | unsigned int sync_queue_delay; |
814 | struct task_struct *master_thread; | |
815 | struct delayed_work master_wakeup_work; | |
816 | struct netns_ipvs *ipvs; | |
817 | }; | |
818 | ||
578bc3ef JA |
819 | /* How much time to keep dests in trash */ |
820 | #define IP_VS_DEST_TRASH_PERIOD (120 * HZ) | |
821 | ||
e4ff6751 | 822 | struct ipvs_sync_daemon_cfg { |
d3328817 | 823 | union nf_inet_addr mcast_group; |
e4ff6751 JA |
824 | int syncid; |
825 | u16 sync_maxlen; | |
d3328817 JA |
826 | u16 mcast_port; |
827 | u8 mcast_af; | |
828 | u8 mcast_ttl; | |
e4ff6751 JA |
829 | /* multicast interface name */ |
830 | char mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
831 | }; | |
832 | ||
2553d064 JA |
833 | /* IPVS in network namespace */ |
834 | struct netns_ipvs { | |
835 | int gen; /* Generation */ | |
7a4f0761 | 836 | int enable; /* enable like nf_hooks do */ |
07dcc686 | 837 | /* Hash table: for real service lookups */ |
2553d064 JA |
838 | #define IP_VS_RTAB_BITS 4 |
839 | #define IP_VS_RTAB_SIZE (1 << IP_VS_RTAB_BITS) | |
840 | #define IP_VS_RTAB_MASK (IP_VS_RTAB_SIZE - 1) | |
841 | ||
276472ea | 842 | struct hlist_head rs_table[IP_VS_RTAB_SIZE]; |
2553d064 JA |
843 | /* ip_vs_app */ |
844 | struct list_head app_list; | |
2553d064 JA |
845 | /* ip_vs_proto */ |
846 | #define IP_VS_PROTO_TAB_SIZE 32 /* must be power of 2 */ | |
847 | struct ip_vs_proto_data *proto_data_table[IP_VS_PROTO_TAB_SIZE]; | |
848 | /* ip_vs_proto_tcp */ | |
849 | #ifdef CONFIG_IP_VS_PROTO_TCP | |
850 | #define TCP_APP_TAB_BITS 4 | |
851 | #define TCP_APP_TAB_SIZE (1 << TCP_APP_TAB_BITS) | |
852 | #define TCP_APP_TAB_MASK (TCP_APP_TAB_SIZE - 1) | |
853 | struct list_head tcp_apps[TCP_APP_TAB_SIZE]; | |
2553d064 JA |
854 | #endif |
855 | /* ip_vs_proto_udp */ | |
856 | #ifdef CONFIG_IP_VS_PROTO_UDP | |
857 | #define UDP_APP_TAB_BITS 4 | |
858 | #define UDP_APP_TAB_SIZE (1 << UDP_APP_TAB_BITS) | |
859 | #define UDP_APP_TAB_MASK (UDP_APP_TAB_SIZE - 1) | |
860 | struct list_head udp_apps[UDP_APP_TAB_SIZE]; | |
2553d064 JA |
861 | #endif |
862 | /* ip_vs_proto_sctp */ | |
863 | #ifdef CONFIG_IP_VS_PROTO_SCTP | |
864 | #define SCTP_APP_TAB_BITS 4 | |
865 | #define SCTP_APP_TAB_SIZE (1 << SCTP_APP_TAB_BITS) | |
866 | #define SCTP_APP_TAB_MASK (SCTP_APP_TAB_SIZE - 1) | |
867 | /* Hash table for SCTP application incarnations */ | |
868 | struct list_head sctp_apps[SCTP_APP_TAB_SIZE]; | |
2553d064 JA |
869 | #endif |
870 | /* ip_vs_conn */ | |
07dcc686 | 871 | atomic_t conn_count; /* connection counter */ |
2553d064 JA |
872 | |
873 | /* ip_vs_ctl */ | |
2a0751af | 874 | struct ip_vs_stats tot_stats; /* Statistics & est. */ |
2553d064 JA |
875 | |
876 | int num_services; /* no of virtual services */ | |
f2247fbd | 877 | |
f2247fbd SH |
878 | /* Trash for destinations */ |
879 | struct list_head dest_trash; | |
578bc3ef JA |
880 | spinlock_t dest_trash_lock; |
881 | struct timer_list dest_trash_timer; /* expiration timer */ | |
f2247fbd SH |
882 | /* Service counters */ |
883 | atomic_t ftpsvc_counter; | |
884 | atomic_t nullsvc_counter; | |
39b97223 | 885 | atomic_t conn_out_counter; |
f2247fbd SH |
886 | |
887 | #ifdef CONFIG_SYSCTL | |
2553d064 JA |
888 | /* 1/rate drop and drop-entry variables */ |
889 | struct delayed_work defense_work; /* Work handler */ | |
890 | int drop_rate; | |
891 | int drop_counter; | |
892 | atomic_t dropentry; | |
893 | /* locks in ctl.c */ | |
894 | spinlock_t dropentry_lock; /* drop entry handling */ | |
895 | spinlock_t droppacket_lock; /* drop packet handling */ | |
896 | spinlock_t securetcp_lock; /* state and timeout tables */ | |
2553d064 JA |
897 | |
898 | /* sys-ctl struct */ | |
899 | struct ctl_table_header *sysctl_hdr; | |
900 | struct ctl_table *sysctl_tbl; | |
f2247fbd SH |
901 | #endif |
902 | ||
2553d064 JA |
903 | /* sysctl variables */ |
904 | int sysctl_amemthresh; | |
905 | int sysctl_am_droprate; | |
906 | int sysctl_drop_entry; | |
907 | int sysctl_drop_packet; | |
908 | int sysctl_secure_tcp; | |
909 | #ifdef CONFIG_IP_VS_NFCT | |
910 | int sysctl_conntrack; | |
911 | #endif | |
912 | int sysctl_snat_reroute; | |
913 | int sysctl_sync_ver; | |
f73181c8 | 914 | int sysctl_sync_ports; |
4d0c875d | 915 | int sysctl_sync_persist_mode; |
07995674 | 916 | unsigned long sysctl_sync_qlen_max; |
1c003b15 | 917 | int sysctl_sync_sock_size; |
2553d064 JA |
918 | int sysctl_cache_bypass; |
919 | int sysctl_expire_nodest_conn; | |
c6c96c18 AF |
920 | int sysctl_sloppy_tcp; |
921 | int sysctl_sloppy_sctp; | |
2553d064 JA |
922 | int sysctl_expire_quiescent_template; |
923 | int sysctl_sync_threshold[2]; | |
749c42b6 JA |
924 | unsigned int sysctl_sync_refresh_period; |
925 | int sysctl_sync_retries; | |
2553d064 | 926 | int sysctl_nat_icmp_send; |
3654e611 | 927 | int sysctl_pmtu_disc; |
0c12582f | 928 | int sysctl_backup_only; |
d752c364 | 929 | int sysctl_conn_reuse_mode; |
94485fed | 930 | int sysctl_schedule_icmp; |
4e478098 | 931 | int sysctl_ignore_tunneled; |
2553d064 JA |
932 | |
933 | /* ip_vs_lblc */ | |
934 | int sysctl_lblc_expiration; | |
935 | struct ctl_table_header *lblc_ctl_header; | |
936 | struct ctl_table *lblc_ctl_table; | |
937 | /* ip_vs_lblcr */ | |
938 | int sysctl_lblcr_expiration; | |
939 | struct ctl_table_header *lblcr_ctl_header; | |
940 | struct ctl_table *lblcr_ctl_table; | |
941 | /* ip_vs_est */ | |
942 | struct list_head est_list; /* estimator list */ | |
943 | spinlock_t est_lock; | |
944 | struct timer_list est_timer; /* Estimation timer */ | |
945 | /* ip_vs_sync */ | |
2553d064 | 946 | spinlock_t sync_lock; |
f73181c8 | 947 | struct ipvs_master_sync_state *ms; |
2553d064 | 948 | spinlock_t sync_buff_lock; |
f73181c8 PNA |
949 | struct task_struct **backup_threads; |
950 | int threads_mask; | |
2553d064 | 951 | volatile int sync_state; |
ae1d48b2 | 952 | struct mutex sync_mutex; |
e4ff6751 JA |
953 | struct ipvs_sync_daemon_cfg mcfg; /* Master Configuration */ |
954 | struct ipvs_sync_daemon_cfg bcfg; /* Backup Configuration */ | |
2553d064 JA |
955 | /* net name space ptr */ |
956 | struct net *net; /* Needed by timer routines */ | |
07dcc686 SH |
957 | /* Number of heterogeneous destinations, needed becaus heterogeneous |
958 | * are not supported when synchronization is enabled. | |
959 | */ | |
391f503d | 960 | unsigned int mixed_address_family_dests; |
2553d064 | 961 | }; |
1da177e4 | 962 | |
59e0350e SH |
963 | #define DEFAULT_SYNC_THRESHOLD 3 |
964 | #define DEFAULT_SYNC_PERIOD 50 | |
7532e8d4 | 965 | #define DEFAULT_SYNC_VER 1 |
c6c96c18 AF |
966 | #define DEFAULT_SLOPPY_TCP 0 |
967 | #define DEFAULT_SLOPPY_SCTP 0 | |
749c42b6 JA |
968 | #define DEFAULT_SYNC_REFRESH_PERIOD (0U * HZ) |
969 | #define DEFAULT_SYNC_RETRIES 0 | |
1c003b15 PNA |
970 | #define IPVS_SYNC_WAKEUP_RATE 8 |
971 | #define IPVS_SYNC_QLEN_MAX (IPVS_SYNC_WAKEUP_RATE * 4) | |
972 | #define IPVS_SYNC_SEND_DELAY (HZ / 50) | |
973 | #define IPVS_SYNC_CHECK_PERIOD HZ | |
749c42b6 | 974 | #define IPVS_SYNC_FLUSH_TIME (HZ * 2) |
f73181c8 | 975 | #define IPVS_SYNC_PORTS_MAX (1 << 6) |
59e0350e SH |
976 | |
977 | #ifdef CONFIG_SYSCTL | |
978 | ||
979 | static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs) | |
980 | { | |
981 | return ipvs->sysctl_sync_threshold[0]; | |
982 | } | |
983 | ||
984 | static inline int sysctl_sync_period(struct netns_ipvs *ipvs) | |
985 | { | |
749c42b6 JA |
986 | return ACCESS_ONCE(ipvs->sysctl_sync_threshold[1]); |
987 | } | |
988 | ||
989 | static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs) | |
990 | { | |
991 | return ACCESS_ONCE(ipvs->sysctl_sync_refresh_period); | |
992 | } | |
993 | ||
994 | static inline int sysctl_sync_retries(struct netns_ipvs *ipvs) | |
995 | { | |
996 | return ipvs->sysctl_sync_retries; | |
59e0350e SH |
997 | } |
998 | ||
7532e8d4 SH |
999 | static inline int sysctl_sync_ver(struct netns_ipvs *ipvs) |
1000 | { | |
1001 | return ipvs->sysctl_sync_ver; | |
1002 | } | |
1003 | ||
c6c96c18 AF |
1004 | static inline int sysctl_sloppy_tcp(struct netns_ipvs *ipvs) |
1005 | { | |
1006 | return ipvs->sysctl_sloppy_tcp; | |
1007 | } | |
1008 | ||
1009 | static inline int sysctl_sloppy_sctp(struct netns_ipvs *ipvs) | |
1010 | { | |
1011 | return ipvs->sysctl_sloppy_sctp; | |
1012 | } | |
1013 | ||
f73181c8 PNA |
1014 | static inline int sysctl_sync_ports(struct netns_ipvs *ipvs) |
1015 | { | |
1016 | return ACCESS_ONCE(ipvs->sysctl_sync_ports); | |
1017 | } | |
1018 | ||
4d0c875d JA |
1019 | static inline int sysctl_sync_persist_mode(struct netns_ipvs *ipvs) |
1020 | { | |
1021 | return ipvs->sysctl_sync_persist_mode; | |
1022 | } | |
1023 | ||
07995674 | 1024 | static inline unsigned long sysctl_sync_qlen_max(struct netns_ipvs *ipvs) |
1c003b15 PNA |
1025 | { |
1026 | return ipvs->sysctl_sync_qlen_max; | |
1027 | } | |
1028 | ||
1029 | static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs) | |
1030 | { | |
1031 | return ipvs->sysctl_sync_sock_size; | |
1032 | } | |
1033 | ||
3654e611 JA |
1034 | static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs) |
1035 | { | |
1036 | return ipvs->sysctl_pmtu_disc; | |
1037 | } | |
1038 | ||
0c12582f JA |
1039 | static inline int sysctl_backup_only(struct netns_ipvs *ipvs) |
1040 | { | |
1041 | return ipvs->sync_state & IP_VS_STATE_BACKUP && | |
1042 | ipvs->sysctl_backup_only; | |
1043 | } | |
1044 | ||
d752c364 MRL |
1045 | static inline int sysctl_conn_reuse_mode(struct netns_ipvs *ipvs) |
1046 | { | |
1047 | return ipvs->sysctl_conn_reuse_mode; | |
1048 | } | |
1049 | ||
94485fed AG |
1050 | static inline int sysctl_schedule_icmp(struct netns_ipvs *ipvs) |
1051 | { | |
1052 | return ipvs->sysctl_schedule_icmp; | |
1053 | } | |
1054 | ||
4e478098 AG |
1055 | static inline int sysctl_ignore_tunneled(struct netns_ipvs *ipvs) |
1056 | { | |
1057 | return ipvs->sysctl_ignore_tunneled; | |
1058 | } | |
1059 | ||
57032948 EB |
1060 | static inline int sysctl_cache_bypass(struct netns_ipvs *ipvs) |
1061 | { | |
1062 | return ipvs->sysctl_cache_bypass; | |
1063 | } | |
1064 | ||
59e0350e SH |
1065 | #else |
1066 | ||
1067 | static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs) | |
1068 | { | |
1069 | return DEFAULT_SYNC_THRESHOLD; | |
1070 | } | |
1071 | ||
1072 | static inline int sysctl_sync_period(struct netns_ipvs *ipvs) | |
1073 | { | |
1074 | return DEFAULT_SYNC_PERIOD; | |
1075 | } | |
1076 | ||
749c42b6 JA |
1077 | static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs) |
1078 | { | |
1079 | return DEFAULT_SYNC_REFRESH_PERIOD; | |
1080 | } | |
1081 | ||
1082 | static inline int sysctl_sync_retries(struct netns_ipvs *ipvs) | |
1083 | { | |
1084 | return DEFAULT_SYNC_RETRIES & 3; | |
1085 | } | |
1086 | ||
7532e8d4 SH |
1087 | static inline int sysctl_sync_ver(struct netns_ipvs *ipvs) |
1088 | { | |
1089 | return DEFAULT_SYNC_VER; | |
1090 | } | |
1091 | ||
c6c96c18 AF |
1092 | static inline int sysctl_sloppy_tcp(struct netns_ipvs *ipvs) |
1093 | { | |
1094 | return DEFAULT_SLOPPY_TCP; | |
1095 | } | |
1096 | ||
1097 | static inline int sysctl_sloppy_sctp(struct netns_ipvs *ipvs) | |
1098 | { | |
1099 | return DEFAULT_SLOPPY_SCTP; | |
1100 | } | |
1101 | ||
f73181c8 PNA |
1102 | static inline int sysctl_sync_ports(struct netns_ipvs *ipvs) |
1103 | { | |
1104 | return 1; | |
1105 | } | |
1106 | ||
4d0c875d JA |
1107 | static inline int sysctl_sync_persist_mode(struct netns_ipvs *ipvs) |
1108 | { | |
1109 | return 0; | |
1110 | } | |
1111 | ||
07995674 | 1112 | static inline unsigned long sysctl_sync_qlen_max(struct netns_ipvs *ipvs) |
1c003b15 PNA |
1113 | { |
1114 | return IPVS_SYNC_QLEN_MAX; | |
1115 | } | |
1116 | ||
1117 | static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs) | |
1118 | { | |
1119 | return 0; | |
1120 | } | |
1121 | ||
3654e611 JA |
1122 | static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs) |
1123 | { | |
1124 | return 1; | |
1125 | } | |
1126 | ||
0c12582f JA |
1127 | static inline int sysctl_backup_only(struct netns_ipvs *ipvs) |
1128 | { | |
1129 | return 0; | |
1130 | } | |
1131 | ||
d752c364 MRL |
1132 | static inline int sysctl_conn_reuse_mode(struct netns_ipvs *ipvs) |
1133 | { | |
1134 | return 1; | |
1135 | } | |
1136 | ||
94485fed AG |
1137 | static inline int sysctl_schedule_icmp(struct netns_ipvs *ipvs) |
1138 | { | |
1139 | return 0; | |
1140 | } | |
1141 | ||
4e478098 AG |
1142 | static inline int sysctl_ignore_tunneled(struct netns_ipvs *ipvs) |
1143 | { | |
1144 | return 0; | |
1145 | } | |
1146 | ||
57032948 EB |
1147 | static inline int sysctl_cache_bypass(struct netns_ipvs *ipvs) |
1148 | { | |
1149 | return 0; | |
1150 | } | |
1151 | ||
59e0350e SH |
1152 | #endif |
1153 | ||
07dcc686 SH |
1154 | /* IPVS core functions |
1155 | * (from ip_vs_core.c) | |
1da177e4 | 1156 | */ |
5c3a0fd7 JP |
1157 | const char *ip_vs_proto_name(unsigned int proto); |
1158 | void ip_vs_init_hash_table(struct list_head *table, int rows); | |
39b97223 MA |
1159 | struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc, |
1160 | struct ip_vs_dest *dest, | |
1161 | struct sk_buff *skb, | |
1162 | const struct ip_vs_iphdr *iph, | |
1163 | __be16 dport, | |
1164 | __be16 cport); | |
afdd6140 | 1165 | #define IP_VS_INIT_HASH_TABLE(t) ip_vs_init_hash_table((t), ARRAY_SIZE((t))) |
1da177e4 | 1166 | |
1da177e4 LT |
1167 | #define IP_VS_APP_TYPE_FTP 1 |
1168 | ||
07dcc686 SH |
1169 | /* ip_vs_conn handling functions |
1170 | * (from ip_vs_conn.c) | |
1da177e4 | 1171 | */ |
1da177e4 LT |
1172 | enum { |
1173 | IP_VS_DIR_INPUT = 0, | |
1174 | IP_VS_DIR_OUTPUT, | |
1175 | IP_VS_DIR_INPUT_ONLY, | |
1176 | IP_VS_DIR_LAST, | |
1177 | }; | |
1178 | ||
19913dec | 1179 | static inline void ip_vs_conn_fill_param(struct netns_ipvs *ipvs, int af, int protocol, |
f11017ec SH |
1180 | const union nf_inet_addr *caddr, |
1181 | __be16 cport, | |
1182 | const union nf_inet_addr *vaddr, | |
1183 | __be16 vport, | |
1184 | struct ip_vs_conn_param *p) | |
1185 | { | |
19913dec | 1186 | p->ipvs = ipvs; |
f11017ec SH |
1187 | p->af = af; |
1188 | p->protocol = protocol; | |
1189 | p->caddr = caddr; | |
1190 | p->cport = cport; | |
1191 | p->vaddr = vaddr; | |
1192 | p->vport = vport; | |
85999283 SH |
1193 | p->pe = NULL; |
1194 | p->pe_data = NULL; | |
f11017ec | 1195 | } |
28364a59 | 1196 | |
f11017ec SH |
1197 | struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p); |
1198 | struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p); | |
28364a59 | 1199 | |
ab161976 EB |
1200 | struct ip_vs_conn * ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af, |
1201 | const struct sk_buff *skb, | |
802c41ad | 1202 | const struct ip_vs_iphdr *iph); |
5c0d2374 | 1203 | |
f11017ec | 1204 | struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p); |
1da177e4 | 1205 | |
0cf705c8 EB |
1206 | struct ip_vs_conn * ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af, |
1207 | const struct sk_buff *skb, | |
802c41ad | 1208 | const struct ip_vs_iphdr *iph); |
5c0d2374 | 1209 | |
088339a5 JA |
1210 | /* Get reference to gain full access to conn. |
1211 | * By default, RCU read-side critical sections have access only to | |
1212 | * conn fields and its PE data, see ip_vs_conn_rcu_free() for reference. | |
1213 | */ | |
1214 | static inline bool __ip_vs_conn_get(struct ip_vs_conn *cp) | |
1215 | { | |
b54ab92b | 1216 | return refcount_inc_not_zero(&cp->refcnt); |
088339a5 JA |
1217 | } |
1218 | ||
1da177e4 LT |
1219 | /* put back the conn without restarting its timer */ |
1220 | static inline void __ip_vs_conn_put(struct ip_vs_conn *cp) | |
1221 | { | |
4e857c58 | 1222 | smp_mb__before_atomic(); |
b54ab92b | 1223 | refcount_dec(&cp->refcnt); |
1da177e4 | 1224 | } |
5c3a0fd7 JP |
1225 | void ip_vs_conn_put(struct ip_vs_conn *cp); |
1226 | void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport); | |
1da177e4 | 1227 | |
ba38528a | 1228 | struct ip_vs_conn *ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af, |
f11017ec | 1229 | const union nf_inet_addr *daddr, |
95c96174 | 1230 | __be16 dport, unsigned int flags, |
0e051e68 | 1231 | struct ip_vs_dest *dest, __u32 fwmark); |
5c3a0fd7 | 1232 | void ip_vs_conn_expire_now(struct ip_vs_conn *cp); |
1da177e4 | 1233 | |
5c3a0fd7 | 1234 | const char *ip_vs_state_name(__u16 proto, int state); |
1da177e4 | 1235 | |
69f39093 | 1236 | void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp); |
3ec10d3a | 1237 | int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest); |
423b5595 | 1238 | void ip_vs_random_dropentry(struct netns_ipvs *ipvs); |
5c3a0fd7 JP |
1239 | int ip_vs_conn_init(void); |
1240 | void ip_vs_conn_cleanup(void); | |
1da177e4 LT |
1241 | |
1242 | static inline void ip_vs_control_del(struct ip_vs_conn *cp) | |
1243 | { | |
1244 | struct ip_vs_conn *ctl_cp = cp->control; | |
1245 | if (!ctl_cp) { | |
cfc78c5a JV |
1246 | IP_VS_ERR_BUF("request control DEL for uncontrolled: " |
1247 | "%s:%d to %s:%d\n", | |
1248 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1249 | ntohs(cp->cport), | |
1250 | IP_VS_DBG_ADDR(cp->af, &cp->vaddr), | |
1251 | ntohs(cp->vport)); | |
1252 | ||
1da177e4 LT |
1253 | return; |
1254 | } | |
1255 | ||
cfc78c5a JV |
1256 | IP_VS_DBG_BUF(7, "DELeting control for: " |
1257 | "cp.dst=%s:%d ctl_cp.dst=%s:%d\n", | |
1258 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1259 | ntohs(cp->cport), | |
1260 | IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr), | |
1261 | ntohs(ctl_cp->cport)); | |
1da177e4 LT |
1262 | |
1263 | cp->control = NULL; | |
1264 | if (atomic_read(&ctl_cp->n_control) == 0) { | |
cfc78c5a JV |
1265 | IP_VS_ERR_BUF("BUG control DEL with n=0 : " |
1266 | "%s:%d to %s:%d\n", | |
1267 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1268 | ntohs(cp->cport), | |
1269 | IP_VS_DBG_ADDR(cp->af, &cp->vaddr), | |
1270 | ntohs(cp->vport)); | |
1271 | ||
1da177e4 LT |
1272 | return; |
1273 | } | |
1274 | atomic_dec(&ctl_cp->n_control); | |
1275 | } | |
1276 | ||
1277 | static inline void | |
1278 | ip_vs_control_add(struct ip_vs_conn *cp, struct ip_vs_conn *ctl_cp) | |
1279 | { | |
1280 | if (cp->control) { | |
cfc78c5a JV |
1281 | IP_VS_ERR_BUF("request control ADD for already controlled: " |
1282 | "%s:%d to %s:%d\n", | |
1283 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1284 | ntohs(cp->cport), | |
1285 | IP_VS_DBG_ADDR(cp->af, &cp->vaddr), | |
1286 | ntohs(cp->vport)); | |
1287 | ||
1da177e4 LT |
1288 | ip_vs_control_del(cp); |
1289 | } | |
1290 | ||
cfc78c5a JV |
1291 | IP_VS_DBG_BUF(7, "ADDing control for: " |
1292 | "cp.dst=%s:%d ctl_cp.dst=%s:%d\n", | |
1293 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1294 | ntohs(cp->cport), | |
1295 | IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr), | |
1296 | ntohs(ctl_cp->cport)); | |
1da177e4 LT |
1297 | |
1298 | cp->control = ctl_cp; | |
1299 | atomic_inc(&ctl_cp->n_control); | |
1300 | } | |
1301 | ||
07dcc686 | 1302 | /* IPVS netns init & cleanup functions */ |
a4dd0360 | 1303 | int ip_vs_estimator_net_init(struct netns_ipvs *ipvs); |
3d993766 | 1304 | int ip_vs_control_net_init(struct netns_ipvs *ipvs); |
7d1f88ec | 1305 | int ip_vs_protocol_net_init(struct netns_ipvs *ipvs); |
b5dd212c | 1306 | int ip_vs_app_net_init(struct netns_ipvs *ipvs); |
2f3edc6a | 1307 | int ip_vs_conn_net_init(struct netns_ipvs *ipvs); |
802cb437 | 1308 | int ip_vs_sync_net_init(struct netns_ipvs *ipvs); |
2f3edc6a | 1309 | void ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs); |
b5dd212c | 1310 | void ip_vs_app_net_cleanup(struct netns_ipvs *ipvs); |
7d1f88ec | 1311 | void ip_vs_protocol_net_cleanup(struct netns_ipvs *ipvs); |
3d993766 | 1312 | void ip_vs_control_net_cleanup(struct netns_ipvs *ipvs); |
a4dd0360 | 1313 | void ip_vs_estimator_net_cleanup(struct netns_ipvs *ipvs); |
ebea1f7c | 1314 | void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs); |
56d2169b | 1315 | void ip_vs_service_net_cleanup(struct netns_ipvs *ipvs); |
1da177e4 | 1316 | |
07dcc686 SH |
1317 | /* IPVS application functions |
1318 | * (from ip_vs_app.c) | |
1da177e4 LT |
1319 | */ |
1320 | #define IP_VS_APP_MAX_PORTS 8 | |
9f8128a5 EB |
1321 | struct ip_vs_app *register_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app); |
1322 | void unregister_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app); | |
5c3a0fd7 JP |
1323 | int ip_vs_bind_app(struct ip_vs_conn *cp, struct ip_vs_protocol *pp); |
1324 | void ip_vs_unbind_app(struct ip_vs_conn *cp); | |
3250dc9c | 1325 | int register_ip_vs_app_inc(struct netns_ipvs *ipvs, struct ip_vs_app *app, __u16 proto, |
5c3a0fd7 JP |
1326 | __u16 port); |
1327 | int ip_vs_app_inc_get(struct ip_vs_app *inc); | |
1328 | void ip_vs_app_inc_put(struct ip_vs_app *inc); | |
1329 | ||
1330 | int ip_vs_app_pkt_out(struct ip_vs_conn *, struct sk_buff *skb); | |
1331 | int ip_vs_app_pkt_in(struct ip_vs_conn *, struct sk_buff *skb); | |
1da177e4 | 1332 | |
8be67a66 SH |
1333 | int register_ip_vs_pe(struct ip_vs_pe *pe); |
1334 | int unregister_ip_vs_pe(struct ip_vs_pe *pe); | |
e9e5eee8 | 1335 | struct ip_vs_pe *ip_vs_pe_getbyname(const char *name); |
fe5e7a1e | 1336 | struct ip_vs_pe *__ip_vs_pe_getbyname(const char *pe_name); |
e9e5eee8 | 1337 | |
07dcc686 | 1338 | /* Use a #define to avoid all of module.h just for these trivial ops */ |
69e7dae4 PG |
1339 | #define ip_vs_pe_get(pe) \ |
1340 | if (pe && pe->module) \ | |
e9e5eee8 | 1341 | __module_get(pe->module); |
e9e5eee8 | 1342 | |
69e7dae4 PG |
1343 | #define ip_vs_pe_put(pe) \ |
1344 | if (pe && pe->module) \ | |
e9e5eee8 | 1345 | module_put(pe->module); |
1da177e4 | 1346 | |
07dcc686 | 1347 | /* IPVS protocol functions (from ip_vs_proto.c) */ |
5c3a0fd7 JP |
1348 | int ip_vs_protocol_init(void); |
1349 | void ip_vs_protocol_cleanup(void); | |
1350 | void ip_vs_protocol_timeout_change(struct netns_ipvs *ipvs, int flags); | |
1351 | int *ip_vs_create_timeout_table(int *table, int size); | |
5c3a0fd7 JP |
1352 | void ip_vs_tcpudp_debug_packet(int af, struct ip_vs_protocol *pp, |
1353 | const struct sk_buff *skb, int offset, | |
1354 | const char *msg); | |
1da177e4 LT |
1355 | |
1356 | extern struct ip_vs_protocol ip_vs_protocol_tcp; | |
1357 | extern struct ip_vs_protocol ip_vs_protocol_udp; | |
1358 | extern struct ip_vs_protocol ip_vs_protocol_icmp; | |
1359 | extern struct ip_vs_protocol ip_vs_protocol_esp; | |
1360 | extern struct ip_vs_protocol ip_vs_protocol_ah; | |
2906f66a | 1361 | extern struct ip_vs_protocol ip_vs_protocol_sctp; |
1da177e4 | 1362 | |
07dcc686 SH |
1363 | /* Registering/unregistering scheduler functions |
1364 | * (from ip_vs_sched.c) | |
1da177e4 | 1365 | */ |
5c3a0fd7 JP |
1366 | int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler); |
1367 | int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler); | |
1368 | int ip_vs_bind_scheduler(struct ip_vs_service *svc, | |
1369 | struct ip_vs_scheduler *scheduler); | |
1370 | void ip_vs_unbind_scheduler(struct ip_vs_service *svc, | |
1371 | struct ip_vs_scheduler *sched); | |
1372 | struct ip_vs_scheduler *ip_vs_scheduler_get(const char *sched_name); | |
1373 | void ip_vs_scheduler_put(struct ip_vs_scheduler *scheduler); | |
1374 | struct ip_vs_conn * | |
190ecd27 | 1375 | ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb, |
d4383f04 JDB |
1376 | struct ip_vs_proto_data *pd, int *ignored, |
1377 | struct ip_vs_iphdr *iph); | |
5c3a0fd7 JP |
1378 | int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb, |
1379 | struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph); | |
1da177e4 | 1380 | |
5c3a0fd7 | 1381 | void ip_vs_scheduler_err(struct ip_vs_service *svc, const char *msg); |
41ac51ee | 1382 | |
07dcc686 | 1383 | /* IPVS control data and functions (from ip_vs_ctl.c) */ |
1da177e4 | 1384 | extern struct ip_vs_stats ip_vs_stats; |
b880c1f0 | 1385 | extern int sysctl_ip_vs_sync_ver; |
1da177e4 | 1386 | |
5c3a0fd7 | 1387 | struct ip_vs_service * |
0a4fd6ce | 1388 | ip_vs_service_find(struct netns_ipvs *ipvs, int af, __u32 fwmark, __u16 protocol, |
3c2e0505 | 1389 | const union nf_inet_addr *vaddr, __be16 vport); |
1da177e4 | 1390 | |
48aed1b0 | 1391 | bool ip_vs_has_real_service(struct netns_ipvs *ipvs, int af, __u16 protocol, |
5c3a0fd7 JP |
1392 | const union nf_inet_addr *daddr, __be16 dport); |
1393 | ||
39b97223 MA |
1394 | struct ip_vs_dest * |
1395 | ip_vs_find_real_service(struct netns_ipvs *ipvs, int af, __u16 protocol, | |
1396 | const union nf_inet_addr *daddr, __be16 dport); | |
1397 | ||
5c3a0fd7 JP |
1398 | int ip_vs_use_count_inc(void); |
1399 | void ip_vs_use_count_dec(void); | |
1400 | int ip_vs_register_nl_ioctl(void); | |
1401 | void ip_vs_unregister_nl_ioctl(void); | |
1402 | int ip_vs_control_init(void); | |
1403 | void ip_vs_control_cleanup(void); | |
1404 | struct ip_vs_dest * | |
dc2add6f | 1405 | ip_vs_find_dest(struct netns_ipvs *ipvs, int svc_af, int dest_af, |
655eef10 AG |
1406 | const union nf_inet_addr *daddr, __be16 dport, |
1407 | const union nf_inet_addr *vaddr, __be16 vport, | |
52793dbe | 1408 | __u16 protocol, __u32 fwmark, __u32 flags); |
5c3a0fd7 | 1409 | void ip_vs_try_bind_dest(struct ip_vs_conn *cp); |
1da177e4 | 1410 | |
fca9c20a JA |
1411 | static inline void ip_vs_dest_hold(struct ip_vs_dest *dest) |
1412 | { | |
b54ab92b | 1413 | refcount_inc(&dest->refcnt); |
fca9c20a JA |
1414 | } |
1415 | ||
1416 | static inline void ip_vs_dest_put(struct ip_vs_dest *dest) | |
1417 | { | |
4e857c58 | 1418 | smp_mb__before_atomic(); |
b54ab92b | 1419 | refcount_dec(&dest->refcnt); |
fca9c20a | 1420 | } |
1da177e4 | 1421 | |
9e4e948a JA |
1422 | static inline void ip_vs_dest_put_and_free(struct ip_vs_dest *dest) |
1423 | { | |
b54ab92b | 1424 | if (refcount_dec_and_test(&dest->refcnt)) |
9e4e948a JA |
1425 | kfree(dest); |
1426 | } | |
1427 | ||
07dcc686 SH |
1428 | /* IPVS sync daemon data and function prototypes |
1429 | * (from ip_vs_sync.c) | |
1da177e4 | 1430 | */ |
6ac121d7 | 1431 | int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *cfg, |
e4ff6751 | 1432 | int state); |
b3cf3cbf | 1433 | int stop_sync_thread(struct netns_ipvs *ipvs, int state); |
b61a8c1a | 1434 | void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts); |
1da177e4 | 1435 | |
07dcc686 | 1436 | /* IPVS rate estimator prototypes (from ip_vs_est.c) */ |
0f34d54b EB |
1437 | void ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats); |
1438 | void ip_vs_stop_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats); | |
5c3a0fd7 | 1439 | void ip_vs_zero_estimator(struct ip_vs_stats *stats); |
cd67cd5e | 1440 | void ip_vs_read_estimator(struct ip_vs_kstats *dst, struct ip_vs_stats *stats); |
1da177e4 | 1441 | |
07dcc686 | 1442 | /* Various IPVS packet transmitters (from ip_vs_xmit.c) */ |
5c3a0fd7 JP |
1443 | int ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, |
1444 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1445 | int ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1446 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1447 | int ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1448 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1449 | int ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1450 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1451 | int ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1452 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1453 | int 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); | |
1456 | void ip_vs_dest_dst_rcu_free(struct rcu_head *head); | |
1da177e4 | 1457 | |
b3cdd2a7 | 1458 | #ifdef CONFIG_IP_VS_IPV6 |
5c3a0fd7 JP |
1459 | int 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); | |
1461 | int 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); | |
1463 | int 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); | |
1465 | int 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); | |
1467 | int 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); | |
b3cdd2a7 | 1470 | #endif |
1da177e4 | 1471 | |
3a1bbf18 | 1472 | #ifdef CONFIG_SYSCTL |
07dcc686 SH |
1473 | /* 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 | |
1da177e4 | 1476 | */ |
a0840e2e | 1477 | static inline int ip_vs_todrop(struct netns_ipvs *ipvs) |
1da177e4 | 1478 | { |
a0840e2e HS |
1479 | if (!ipvs->drop_rate) |
1480 | return 0; | |
1481 | if (--ipvs->drop_counter > 0) | |
1482 | return 0; | |
1483 | ipvs->drop_counter = ipvs->drop_rate; | |
1da177e4 LT |
1484 | return 1; |
1485 | } | |
3a1bbf18 SH |
1486 | #else |
1487 | static inline int ip_vs_todrop(struct netns_ipvs *ipvs) { return 0; } | |
1488 | #endif | |
1da177e4 | 1489 | |
07dcc686 | 1490 | /* ip_vs_fwd_tag returns the forwarding tag of the connection */ |
1da177e4 LT |
1491 | #define IP_VS_FWD_METHOD(cp) (cp->flags & IP_VS_CONN_F_FWD_MASK) |
1492 | ||
732db659 | 1493 | static inline char ip_vs_fwd_tag(struct ip_vs_conn *cp) |
1da177e4 LT |
1494 | { |
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 | ||
5c3a0fd7 JP |
1514 | void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp, |
1515 | struct ip_vs_conn *cp, int dir); | |
b3cdd2a7 JV |
1516 | |
1517 | #ifdef CONFIG_IP_VS_IPV6 | |
5c3a0fd7 JP |
1518 | void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp, |
1519 | struct ip_vs_conn *cp, int dir); | |
b3cdd2a7 | 1520 | #endif |
1da177e4 | 1521 | |
5c3a0fd7 | 1522 | __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset); |
1da177e4 | 1523 | |
f9214b26 | 1524 | static inline __wsum ip_vs_check_diff4(__be32 old, __be32 new, __wsum oldsum) |
1da177e4 | 1525 | { |
f9214b26 | 1526 | __be32 diff[2] = { ~old, new }; |
1da177e4 | 1527 | |
07f0757a | 1528 | return csum_partial(diff, sizeof(diff), oldsum); |
f9214b26 AV |
1529 | } |
1530 | ||
0bbdd42b JV |
1531 | #ifdef CONFIG_IP_VS_IPV6 |
1532 | static 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 | ||
07f0757a | 1538 | return csum_partial(diff, sizeof(diff), oldsum); |
0bbdd42b JV |
1539 | } |
1540 | #endif | |
1541 | ||
f9214b26 AV |
1542 | static inline __wsum ip_vs_check_diff2(__be16 old, __be16 new, __wsum oldsum) |
1543 | { | |
1544 | __be16 diff[2] = { ~old, new }; | |
1545 | ||
07f0757a | 1546 | return csum_partial(diff, sizeof(diff), oldsum); |
1da177e4 LT |
1547 | } |
1548 | ||
07dcc686 | 1549 | /* Forget current conntrack (unconfirmed) and attach notrack entry */ |
cf356d69 JA |
1550 | static 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; | |
06b69390 | 1554 | struct nf_conn *ct = nf_ct_get(skb, &ctinfo); |
cf356d69 | 1555 | |
ab8bc7ed | 1556 | if (ct) { |
97a6ad13 | 1557 | nf_conntrack_put(&ct->ct_general); |
cc41c84b | 1558 | nf_ct_set(skb, NULL, IP_CT_UNTRACKED); |
cf356d69 JA |
1559 | } |
1560 | #endif | |
1561 | } | |
1562 | ||
f4bc17cd | 1563 | #ifdef CONFIG_IP_VS_NFCT |
07dcc686 SH |
1564 | /* Netfilter connection tracking |
1565 | * (from ip_vs_nfct.c) | |
f4bc17cd | 1566 | */ |
a0840e2e | 1567 | static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs) |
f4bc17cd | 1568 | { |
a4e2f5a7 | 1569 | #ifdef CONFIG_SYSCTL |
a0840e2e | 1570 | return ipvs->sysctl_conntrack; |
a4e2f5a7 SH |
1571 | #else |
1572 | return 0; | |
1573 | #endif | |
f4bc17cd JA |
1574 | } |
1575 | ||
5c3a0fd7 JP |
1576 | void ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp, |
1577 | int outin); | |
1578 | int ip_vs_confirm_conntrack(struct sk_buff *skb); | |
1579 | void ip_vs_nfct_expect_related(struct sk_buff *skb, struct nf_conn *ct, | |
1580 | struct ip_vs_conn *cp, u_int8_t proto, | |
1581 | const __be16 port, int from_rs); | |
1582 | void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp); | |
f4bc17cd JA |
1583 | |
1584 | #else | |
1585 | ||
a0840e2e | 1586 | static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs) |
f4bc17cd JA |
1587 | { |
1588 | return 0; | |
1589 | } | |
1590 | ||
1591 | static inline void ip_vs_update_conntrack(struct sk_buff *skb, | |
1592 | struct ip_vs_conn *cp, int outin) | |
1593 | { | |
1594 | } | |
1595 | ||
e23ebf0f | 1596 | static inline int ip_vs_confirm_conntrack(struct sk_buff *skb) |
f4bc17cd JA |
1597 | { |
1598 | return NF_ACCEPT; | |
1599 | } | |
1600 | ||
1601 | static inline void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp) | |
1602 | { | |
1603 | } | |
07dcc686 | 1604 | #endif /* CONFIG_IP_VS_NFCT */ |
6523ce15 | 1605 | |
f719e375 JA |
1606 | /* Really using conntrack? */ |
1607 | static inline bool ip_vs_conn_uses_conntrack(struct ip_vs_conn *cp, | |
1608 | struct sk_buff *skb) | |
1609 | { | |
1610 | #ifdef CONFIG_IP_VS_NFCT | |
1611 | enum ip_conntrack_info ctinfo; | |
1612 | struct nf_conn *ct; | |
1613 | ||
1614 | if (!(cp->flags & IP_VS_CONN_F_NFCT)) | |
1615 | return false; | |
1616 | ct = nf_ct_get(skb, &ctinfo); | |
ab8bc7ed | 1617 | if (ct) |
f719e375 JA |
1618 | return true; |
1619 | #endif | |
1620 | return false; | |
1621 | } | |
1622 | ||
c16526a7 | 1623 | static inline int |
b552f7e3 CG |
1624 | ip_vs_dest_conn_overhead(struct ip_vs_dest *dest) |
1625 | { | |
07dcc686 | 1626 | /* We think the overhead of processing active connections is 256 |
b552f7e3 CG |
1627 | * times higher than that of inactive connections in average. (This |
1628 | * 256 times might not be accurate, we will change it later) We | |
1629 | * use the following formula to estimate the overhead now: | |
1630 | * dest->activeconns*256 + dest->inactconns | |
1631 | */ | |
1632 | return (atomic_read(&dest->activeconns) << 8) + | |
1633 | atomic_read(&dest->inactconns); | |
1634 | } | |
1635 | ||
bc4768eb | 1636 | #endif /* _NET_IP_VS_H */ |