]>
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 */ |
1da177e4 | 15 | #include <linux/compiler.h> |
14c85021 | 16 | #include <linux/timer.h> |
187f1882 | 17 | #include <linux/bug.h> |
1da177e4 | 18 | |
14c85021 | 19 | #include <net/checksum.h> |
e7ade46a | 20 | #include <linux/netfilter.h> /* for union nf_inet_addr */ |
1668e010 | 21 | #include <linux/ip.h> |
e7ade46a | 22 | #include <linux/ipv6.h> /* for struct ipv6hdr */ |
4e3fd7a0 | 23 | #include <net/ipv6.h> |
54d83efa | 24 | #if IS_ENABLED(CONFIG_IP_VS_IPV6) |
63dca2c0 JDB |
25 | #include <linux/netfilter_ipv6/ip6_tables.h> |
26 | #endif | |
a638e514 | 27 | #if IS_ENABLED(CONFIG_NF_CONNTRACK) |
f4bc17cd JA |
28 | #include <net/netfilter/nf_conntrack.h> |
29 | #endif | |
61b1ab45 HS |
30 | #include <net/net_namespace.h> /* Netw namespace */ |
31 | ||
4fd9beef AG |
32 | #define IP_VS_HDR_INVERSE 1 |
33 | #define IP_VS_HDR_ICMP 2 | |
34 | ||
07dcc686 | 35 | /* Generic access of ipvs struct */ |
61b1ab45 HS |
36 | static inline struct netns_ipvs *net_ipvs(struct net* net) |
37 | { | |
38 | return net->ipvs; | |
39 | } | |
07dcc686 SH |
40 | |
41 | /* Get net ptr from skb in traffic cases | |
fc723250 HS |
42 | * use skb_sknet when call is from userland (ioctl or netlink) |
43 | */ | |
a0840e2e | 44 | static inline struct net *skb_net(const struct sk_buff *skb) |
fc723250 HS |
45 | { |
46 | #ifdef CONFIG_NET_NS | |
47 | #ifdef CONFIG_IP_VS_DEBUG | |
48 | /* | |
49 | * This is used for debug only. | |
50 | * Start with the most likely hit | |
51 | * End with BUG | |
52 | */ | |
0c5c9fb5 | 53 | if (likely(skb->dev && dev_net(skb->dev))) |
fc723250 | 54 | return dev_net(skb->dev); |
a09d1977 | 55 | if (skb_dst(skb) && skb_dst(skb)->dev) |
fc723250 HS |
56 | return dev_net(skb_dst(skb)->dev); |
57 | WARN(skb->sk, "Maybe skb_sknet should be used in %s() at line:%d\n", | |
58 | __func__, __LINE__); | |
0c5c9fb5 | 59 | if (likely(skb->sk && sock_net(skb->sk))) |
fc723250 HS |
60 | return sock_net(skb->sk); |
61 | pr_err("There is no net ptr to find in the skb in %s() line:%d\n", | |
62 | __func__, __LINE__); | |
63 | BUG(); | |
64 | #else | |
65 | return dev_net(skb->dev ? : skb_dst(skb)->dev); | |
66 | #endif | |
67 | #else | |
68 | return &init_net; | |
69 | #endif | |
70 | } | |
71 | ||
a0840e2e | 72 | static inline struct net *skb_sknet(const struct sk_buff *skb) |
fc723250 HS |
73 | { |
74 | #ifdef CONFIG_NET_NS | |
75 | #ifdef CONFIG_IP_VS_DEBUG | |
76 | /* Start with the most likely hit */ | |
0c5c9fb5 | 77 | if (likely(skb->sk && sock_net(skb->sk))) |
fc723250 HS |
78 | return sock_net(skb->sk); |
79 | WARN(skb->dev, "Maybe skb_net should be used instead in %s() line:%d\n", | |
80 | __func__, __LINE__); | |
0c5c9fb5 | 81 | if (likely(skb->dev && dev_net(skb->dev))) |
fc723250 HS |
82 | return dev_net(skb->dev); |
83 | pr_err("There is no net ptr to find in the skb in %s() line:%d\n", | |
84 | __func__, __LINE__); | |
85 | BUG(); | |
86 | #else | |
87 | return sock_net(skb->sk); | |
88 | #endif | |
89 | #else | |
90 | return &init_net; | |
91 | #endif | |
92 | } | |
07dcc686 SH |
93 | |
94 | /* This one needed for single_open_net since net is stored directly in | |
25985edc | 95 | * private not as a struct i.e. seq_file_net can't be used. |
b17fc996 HS |
96 | */ |
97 | static inline struct net *seq_file_single_net(struct seq_file *seq) | |
98 | { | |
99 | #ifdef CONFIG_NET_NS | |
100 | return (struct net *)seq->private; | |
101 | #else | |
102 | return &init_net; | |
103 | #endif | |
104 | } | |
6f7edb48 CB |
105 | |
106 | /* Connections' size value needed by ip_vs_ctl.c */ | |
107 | extern int ip_vs_conn_tab_size; | |
108 | ||
64aae3cb | 109 | struct ip_vs_iphdr { |
4fd9beef | 110 | int hdr_flags; /* ipvs flags */ |
b0e010c5 | 111 | __u32 off; /* Where IP or IPv4 header starts */ |
63dca2c0 | 112 | __u32 len; /* IPv4 simply where L4 starts |
07dcc686 | 113 | * IPv6 where L4 Transport Header starts */ |
63dca2c0 JDB |
114 | __u16 fragoffs; /* IPv6 fragment offset, 0 if first frag (or not frag)*/ |
115 | __s16 protocol; | |
116 | __s32 flags; | |
64aae3cb JV |
117 | union nf_inet_addr saddr; |
118 | union nf_inet_addr daddr; | |
119 | }; | |
120 | ||
2f74713d JDB |
121 | static inline void *frag_safe_skb_hp(const struct sk_buff *skb, int offset, |
122 | int len, void *buffer, | |
123 | const struct ip_vs_iphdr *ipvsh) | |
124 | { | |
125 | return skb_header_pointer(skb, offset, len, buffer); | |
126 | } | |
2f74713d | 127 | |
63dca2c0 JDB |
128 | /* This function handles filling *ip_vs_iphdr, both for IPv4 and IPv6. |
129 | * IPv6 requires some extra work, as finding proper header position, | |
130 | * depend on the IPv6 extension headers. | |
131 | */ | |
b0e010c5 AG |
132 | static inline int |
133 | ip_vs_fill_iph_skb_off(int af, const struct sk_buff *skb, int offset, | |
4fd9beef | 134 | int hdr_flags, struct ip_vs_iphdr *iphdr) |
64aae3cb | 135 | { |
4fd9beef | 136 | iphdr->hdr_flags = hdr_flags; |
b0e010c5 | 137 | iphdr->off = offset; |
4fd9beef | 138 | |
64aae3cb JV |
139 | #ifdef CONFIG_IP_VS_IPV6 |
140 | if (af == AF_INET6) { | |
b0e010c5 AG |
141 | struct ipv6hdr _iph; |
142 | const struct ipv6hdr *iph = skb_header_pointer( | |
143 | skb, offset, sizeof(_iph), &_iph); | |
144 | if (!iph) | |
145 | return 0; | |
146 | ||
4e3fd7a0 AD |
147 | iphdr->saddr.in6 = iph->saddr; |
148 | iphdr->daddr.in6 = iph->daddr; | |
6aafeef0 | 149 | /* ipv6_find_hdr() updates len, flags */ |
b0e010c5 | 150 | iphdr->len = offset; |
63dca2c0 JDB |
151 | iphdr->flags = 0; |
152 | iphdr->protocol = ipv6_find_hdr(skb, &iphdr->len, -1, | |
153 | &iphdr->fragoffs, | |
154 | &iphdr->flags); | |
b0e010c5 AG |
155 | if (iphdr->protocol < 0) |
156 | return 0; | |
64aae3cb JV |
157 | } else |
158 | #endif | |
159 | { | |
b0e010c5 AG |
160 | struct iphdr _iph; |
161 | const struct iphdr *iph = skb_header_pointer( | |
162 | skb, offset, sizeof(_iph), &_iph); | |
163 | if (!iph) | |
164 | return 0; | |
165 | ||
166 | iphdr->len = offset + iph->ihl * 4; | |
63dca2c0 JDB |
167 | iphdr->fragoffs = 0; |
168 | iphdr->protocol = iph->protocol; | |
169 | iphdr->saddr.ip = iph->saddr; | |
170 | iphdr->daddr.ip = iph->daddr; | |
171 | } | |
b0e010c5 AG |
172 | |
173 | return 1; | |
174 | } | |
175 | ||
176 | static inline int | |
4fd9beef AG |
177 | ip_vs_fill_iph_skb_icmp(int af, const struct sk_buff *skb, int offset, |
178 | bool inverse, struct ip_vs_iphdr *iphdr) | |
179 | { | |
180 | int hdr_flags = IP_VS_HDR_ICMP; | |
181 | ||
182 | if (inverse) | |
183 | hdr_flags |= IP_VS_HDR_INVERSE; | |
184 | ||
185 | return ip_vs_fill_iph_skb_off(af, skb, offset, hdr_flags, iphdr); | |
186 | } | |
187 | ||
188 | static inline int | |
189 | ip_vs_fill_iph_skb(int af, const struct sk_buff *skb, bool inverse, | |
190 | struct ip_vs_iphdr *iphdr) | |
191 | { | |
192 | int hdr_flags = 0; | |
193 | ||
194 | if (inverse) | |
195 | hdr_flags |= IP_VS_HDR_INVERSE; | |
196 | ||
197 | return ip_vs_fill_iph_skb_off(af, skb, skb_network_offset(skb), | |
198 | hdr_flags, iphdr); | |
199 | } | |
200 | ||
201 | static inline bool | |
202 | ip_vs_iph_inverse(const struct ip_vs_iphdr *iph) | |
203 | { | |
204 | return !!(iph->hdr_flags & IP_VS_HDR_INVERSE); | |
205 | } | |
206 | ||
207 | static inline bool | |
208 | ip_vs_iph_icmp(const struct ip_vs_iphdr *iph) | |
b0e010c5 | 209 | { |
4fd9beef | 210 | return !!(iph->hdr_flags & IP_VS_HDR_ICMP); |
63dca2c0 JDB |
211 | } |
212 | ||
64aae3cb JV |
213 | static inline void ip_vs_addr_copy(int af, union nf_inet_addr *dst, |
214 | const union nf_inet_addr *src) | |
215 | { | |
216 | #ifdef CONFIG_IP_VS_IPV6 | |
217 | if (af == AF_INET6) | |
4e3fd7a0 | 218 | dst->in6 = src->in6; |
64aae3cb JV |
219 | else |
220 | #endif | |
221 | dst->ip = src->ip; | |
222 | } | |
223 | ||
9a05475c JA |
224 | static inline void ip_vs_addr_set(int af, union nf_inet_addr *dst, |
225 | const union nf_inet_addr *src) | |
226 | { | |
227 | #ifdef CONFIG_IP_VS_IPV6 | |
228 | if (af == AF_INET6) { | |
229 | dst->in6 = src->in6; | |
230 | return; | |
231 | } | |
232 | #endif | |
233 | dst->ip = src->ip; | |
234 | dst->all[1] = 0; | |
235 | dst->all[2] = 0; | |
236 | dst->all[3] = 0; | |
237 | } | |
238 | ||
64aae3cb JV |
239 | static inline int ip_vs_addr_equal(int af, const union nf_inet_addr *a, |
240 | const union nf_inet_addr *b) | |
241 | { | |
242 | #ifdef CONFIG_IP_VS_IPV6 | |
243 | if (af == AF_INET6) | |
244 | return ipv6_addr_equal(&a->in6, &b->in6); | |
245 | #endif | |
246 | return a->ip == b->ip; | |
247 | } | |
248 | ||
1da177e4 | 249 | #ifdef CONFIG_IP_VS_DEBUG |
14c85021 ACM |
250 | #include <linux/net.h> |
251 | ||
5c3a0fd7 | 252 | int ip_vs_get_debug_level(void); |
c842a3ad JV |
253 | |
254 | static inline const char *ip_vs_dbg_addr(int af, char *buf, size_t buf_len, | |
255 | const union nf_inet_addr *addr, | |
256 | int *idx) | |
257 | { | |
258 | int len; | |
259 | #ifdef CONFIG_IP_VS_IPV6 | |
260 | if (af == AF_INET6) | |
120b9c14 | 261 | len = snprintf(&buf[*idx], buf_len - *idx, "[%pI6c]", |
0c6ce78a | 262 | &addr->in6) + 1; |
c842a3ad JV |
263 | else |
264 | #endif | |
3685f25d HH |
265 | len = snprintf(&buf[*idx], buf_len - *idx, "%pI4", |
266 | &addr->ip) + 1; | |
c842a3ad JV |
267 | |
268 | *idx += len; | |
269 | BUG_ON(*idx > buf_len + 1); | |
270 | return &buf[*idx - len]; | |
271 | } | |
272 | ||
9aada7ac HE |
273 | #define IP_VS_DBG_BUF(level, msg, ...) \ |
274 | do { \ | |
275 | char ip_vs_dbg_buf[160]; \ | |
276 | int ip_vs_dbg_idx = 0; \ | |
277 | if (level <= ip_vs_get_debug_level()) \ | |
278 | printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \ | |
279 | } while (0) | |
280 | #define IP_VS_ERR_BUF(msg...) \ | |
281 | do { \ | |
282 | char ip_vs_dbg_buf[160]; \ | |
283 | int ip_vs_dbg_idx = 0; \ | |
284 | pr_err(msg); \ | |
285 | } while (0) | |
c842a3ad JV |
286 | |
287 | /* Only use from within IP_VS_DBG_BUF() or IP_VS_ERR_BUF macros */ | |
9aada7ac HE |
288 | #define IP_VS_DBG_ADDR(af, addr) \ |
289 | ip_vs_dbg_addr(af, ip_vs_dbg_buf, \ | |
290 | sizeof(ip_vs_dbg_buf), addr, \ | |
291 | &ip_vs_dbg_idx) | |
292 | ||
293 | #define IP_VS_DBG(level, msg, ...) \ | |
294 | do { \ | |
295 | if (level <= ip_vs_get_debug_level()) \ | |
296 | printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \ | |
297 | } while (0) | |
298 | #define IP_VS_DBG_RL(msg, ...) \ | |
299 | do { \ | |
300 | if (net_ratelimit()) \ | |
301 | printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \ | |
302 | } while (0) | |
0d79641a | 303 | #define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) \ |
9aada7ac HE |
304 | do { \ |
305 | if (level <= ip_vs_get_debug_level()) \ | |
0d79641a | 306 | pp->debug_packet(af, pp, skb, ofs, msg); \ |
9aada7ac | 307 | } while (0) |
0d79641a | 308 | #define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) \ |
9aada7ac HE |
309 | do { \ |
310 | if (level <= ip_vs_get_debug_level() && \ | |
311 | net_ratelimit()) \ | |
0d79641a | 312 | pp->debug_packet(af, pp, skb, ofs, msg); \ |
9aada7ac | 313 | } while (0) |
1da177e4 | 314 | #else /* NO DEBUGGING at ALL */ |
c842a3ad JV |
315 | #define IP_VS_DBG_BUF(level, msg...) do {} while (0) |
316 | #define IP_VS_ERR_BUF(msg...) do {} while (0) | |
1da177e4 LT |
317 | #define IP_VS_DBG(level, msg...) do {} while (0) |
318 | #define IP_VS_DBG_RL(msg...) do {} while (0) | |
0d79641a JA |
319 | #define IP_VS_DBG_PKT(level, af, pp, skb, ofs, msg) do {} while (0) |
320 | #define IP_VS_DBG_RL_PKT(level, af, pp, skb, ofs, msg) do {} while (0) | |
1da177e4 LT |
321 | #endif |
322 | ||
323 | #define IP_VS_BUG() BUG() | |
1e3e238e | 324 | #define IP_VS_ERR_RL(msg, ...) \ |
9aada7ac HE |
325 | do { \ |
326 | if (net_ratelimit()) \ | |
1e3e238e | 327 | pr_err(msg, ##__VA_ARGS__); \ |
9aada7ac | 328 | } while (0) |
1da177e4 LT |
329 | |
330 | #ifdef CONFIG_IP_VS_DEBUG | |
331 | #define EnterFunction(level) \ | |
9aada7ac HE |
332 | do { \ |
333 | if (level <= ip_vs_get_debug_level()) \ | |
334 | printk(KERN_DEBUG \ | |
335 | pr_fmt("Enter: %s, %s line %i\n"), \ | |
336 | __func__, __FILE__, __LINE__); \ | |
337 | } while (0) | |
338 | #define LeaveFunction(level) \ | |
339 | do { \ | |
340 | if (level <= ip_vs_get_debug_level()) \ | |
341 | printk(KERN_DEBUG \ | |
342 | pr_fmt("Leave: %s, %s line %i\n"), \ | |
343 | __func__, __FILE__, __LINE__); \ | |
344 | } while (0) | |
1da177e4 LT |
345 | #else |
346 | #define EnterFunction(level) do {} while (0) | |
347 | #define LeaveFunction(level) do {} while (0) | |
348 | #endif | |
349 | ||
07dcc686 | 350 | /* The port number of FTP service (in network order). */ |
f3a7c66b HH |
351 | #define FTPPORT cpu_to_be16(21) |
352 | #define FTPDATA cpu_to_be16(20) | |
1da177e4 | 353 | |
07dcc686 | 354 | /* TCP State Values */ |
1da177e4 LT |
355 | enum { |
356 | IP_VS_TCP_S_NONE = 0, | |
357 | IP_VS_TCP_S_ESTABLISHED, | |
358 | IP_VS_TCP_S_SYN_SENT, | |
359 | IP_VS_TCP_S_SYN_RECV, | |
360 | IP_VS_TCP_S_FIN_WAIT, | |
361 | IP_VS_TCP_S_TIME_WAIT, | |
362 | IP_VS_TCP_S_CLOSE, | |
363 | IP_VS_TCP_S_CLOSE_WAIT, | |
364 | IP_VS_TCP_S_LAST_ACK, | |
365 | IP_VS_TCP_S_LISTEN, | |
366 | IP_VS_TCP_S_SYNACK, | |
367 | IP_VS_TCP_S_LAST | |
368 | }; | |
369 | ||
07dcc686 | 370 | /* UDP State Values */ |
1da177e4 LT |
371 | enum { |
372 | IP_VS_UDP_S_NORMAL, | |
373 | IP_VS_UDP_S_LAST, | |
374 | }; | |
375 | ||
07dcc686 | 376 | /* ICMP State Values */ |
1da177e4 LT |
377 | enum { |
378 | IP_VS_ICMP_S_NORMAL, | |
379 | IP_VS_ICMP_S_LAST, | |
380 | }; | |
381 | ||
07dcc686 | 382 | /* SCTP State Values */ |
2906f66a VMR |
383 | enum ip_vs_sctp_states { |
384 | IP_VS_SCTP_S_NONE, | |
61e7c420 JA |
385 | IP_VS_SCTP_S_INIT1, |
386 | IP_VS_SCTP_S_INIT, | |
387 | IP_VS_SCTP_S_COOKIE_SENT, | |
388 | IP_VS_SCTP_S_COOKIE_REPLIED, | |
389 | IP_VS_SCTP_S_COOKIE_WAIT, | |
390 | IP_VS_SCTP_S_COOKIE, | |
391 | IP_VS_SCTP_S_COOKIE_ECHOED, | |
2906f66a | 392 | IP_VS_SCTP_S_ESTABLISHED, |
61e7c420 JA |
393 | IP_VS_SCTP_S_SHUTDOWN_SENT, |
394 | IP_VS_SCTP_S_SHUTDOWN_RECEIVED, | |
395 | IP_VS_SCTP_S_SHUTDOWN_ACK_SENT, | |
396 | IP_VS_SCTP_S_REJECTED, | |
2906f66a VMR |
397 | IP_VS_SCTP_S_CLOSED, |
398 | IP_VS_SCTP_S_LAST | |
399 | }; | |
400 | ||
07dcc686 SH |
401 | /* Delta sequence info structure |
402 | * Each ip_vs_conn has 2 (output AND input seq. changes). | |
403 | * Only used in the VS/NAT. | |
1da177e4 LT |
404 | */ |
405 | struct ip_vs_seq { | |
406 | __u32 init_seq; /* Add delta from this seq */ | |
407 | __u32 delta; /* Delta in sequence numbers */ | |
408 | __u32 previous_delta; /* Delta in sequence numbers | |
07dcc686 | 409 | * before last resized pkt */ |
1da177e4 LT |
410 | }; |
411 | ||
07dcc686 | 412 | /* counters per cpu */ |
b17fc996 | 413 | struct ip_vs_counters { |
cd67cd5e JA |
414 | __u64 conns; /* connections scheduled */ |
415 | __u64 inpkts; /* incoming packets */ | |
416 | __u64 outpkts; /* outgoing packets */ | |
b17fc996 HS |
417 | __u64 inbytes; /* incoming bytes */ |
418 | __u64 outbytes; /* outgoing bytes */ | |
419 | }; | |
07dcc686 | 420 | /* Stats per cpu */ |
b17fc996 | 421 | struct ip_vs_cpu_stats { |
cd67cd5e | 422 | struct ip_vs_counters cnt; |
b17fc996 HS |
423 | struct u64_stats_sync syncp; |
424 | }; | |
1da177e4 | 425 | |
07dcc686 | 426 | /* IPVS statistics objects */ |
3a14a313 SW |
427 | struct ip_vs_estimator { |
428 | struct list_head list; | |
429 | ||
430 | u64 last_inbytes; | |
431 | u64 last_outbytes; | |
cd67cd5e JA |
432 | u64 last_conns; |
433 | u64 last_inpkts; | |
434 | u64 last_outpkts; | |
435 | ||
436 | u64 cps; | |
437 | u64 inpps; | |
438 | u64 outpps; | |
439 | u64 inbps; | |
440 | u64 outbps; | |
441 | }; | |
442 | ||
443 | /* | |
444 | * IPVS statistics object, 64-bit kernel version of struct ip_vs_stats_user | |
445 | */ | |
446 | struct ip_vs_kstats { | |
447 | u64 conns; /* connections scheduled */ | |
448 | u64 inpkts; /* incoming packets */ | |
449 | u64 outpkts; /* outgoing packets */ | |
450 | u64 inbytes; /* incoming bytes */ | |
451 | u64 outbytes; /* outgoing bytes */ | |
452 | ||
453 | u64 cps; /* current connection rate */ | |
454 | u64 inpps; /* current in packet rate */ | |
455 | u64 outpps; /* current out packet rate */ | |
456 | u64 inbps; /* current in byte rate */ | |
457 | u64 outbps; /* current out byte rate */ | |
3a14a313 SW |
458 | }; |
459 | ||
fd2c3ef7 | 460 | struct ip_vs_stats { |
cd67cd5e | 461 | struct ip_vs_kstats kstats; /* kernel statistics */ |
e9c0ce23 | 462 | struct ip_vs_estimator est; /* estimator */ |
b962abdc | 463 | struct ip_vs_cpu_stats __percpu *cpustats; /* per cpu counters */ |
b17fc996 | 464 | spinlock_t lock; /* spin lock */ |
cd67cd5e | 465 | struct ip_vs_kstats kstats0; /* reset values */ |
1da177e4 LT |
466 | }; |
467 | ||
14c85021 ACM |
468 | struct dst_entry; |
469 | struct iphdr; | |
1da177e4 LT |
470 | struct ip_vs_conn; |
471 | struct ip_vs_app; | |
14c85021 | 472 | struct sk_buff; |
252c6410 | 473 | struct ip_vs_proto_data; |
1da177e4 LT |
474 | |
475 | struct ip_vs_protocol { | |
476 | struct ip_vs_protocol *next; | |
477 | char *name; | |
2ad17def JA |
478 | u16 protocol; |
479 | u16 num_states; | |
1da177e4 | 480 | int dont_defrag; |
1da177e4 LT |
481 | |
482 | void (*init)(struct ip_vs_protocol *pp); | |
483 | ||
484 | void (*exit)(struct ip_vs_protocol *pp); | |
485 | ||
1281a9c2 | 486 | int (*init_netns)(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd); |
252c6410 | 487 | |
1281a9c2 | 488 | void (*exit_netns)(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd); |
252c6410 | 489 | |
51ef348b | 490 | int (*conn_schedule)(int af, struct sk_buff *skb, |
9330419d | 491 | struct ip_vs_proto_data *pd, |
d4383f04 JDB |
492 | int *verdict, struct ip_vs_conn **cpp, |
493 | struct ip_vs_iphdr *iph); | |
1da177e4 LT |
494 | |
495 | struct ip_vs_conn * | |
ab161976 EB |
496 | (*conn_in_get)(struct netns_ipvs *ipvs, |
497 | int af, | |
51ef348b | 498 | const struct sk_buff *skb, |
802c41ad | 499 | const struct ip_vs_iphdr *iph); |
1da177e4 LT |
500 | |
501 | struct ip_vs_conn * | |
0cf705c8 EB |
502 | (*conn_out_get)(struct netns_ipvs *ipvs, |
503 | int af, | |
51ef348b | 504 | const struct sk_buff *skb, |
802c41ad | 505 | const struct ip_vs_iphdr *iph); |
1da177e4 | 506 | |
d4383f04 JDB |
507 | int (*snat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp, |
508 | struct ip_vs_conn *cp, struct ip_vs_iphdr *iph); | |
1da177e4 | 509 | |
d4383f04 JDB |
510 | int (*dnat_handler)(struct sk_buff *skb, struct ip_vs_protocol *pp, |
511 | struct ip_vs_conn *cp, struct ip_vs_iphdr *iph); | |
1da177e4 | 512 | |
51ef348b JV |
513 | int (*csum_check)(int af, struct sk_buff *skb, |
514 | struct ip_vs_protocol *pp); | |
1da177e4 LT |
515 | |
516 | const char *(*state_name)(int state); | |
517 | ||
4a516f11 SH |
518 | void (*state_transition)(struct ip_vs_conn *cp, int direction, |
519 | const struct sk_buff *skb, | |
520 | struct ip_vs_proto_data *pd); | |
1da177e4 | 521 | |
19648918 | 522 | int (*register_app)(struct netns_ipvs *ipvs, struct ip_vs_app *inc); |
1da177e4 | 523 | |
19648918 | 524 | void (*unregister_app)(struct netns_ipvs *ipvs, struct ip_vs_app *inc); |
1da177e4 LT |
525 | |
526 | int (*app_conn_bind)(struct ip_vs_conn *cp); | |
527 | ||
0d79641a | 528 | void (*debug_packet)(int af, struct ip_vs_protocol *pp, |
1da177e4 LT |
529 | const struct sk_buff *skb, |
530 | int offset, | |
531 | const char *msg); | |
532 | ||
9330419d | 533 | void (*timeout_change)(struct ip_vs_proto_data *pd, int flags); |
1da177e4 LT |
534 | }; |
535 | ||
07dcc686 | 536 | /* protocol data per netns */ |
252c6410 HS |
537 | struct ip_vs_proto_data { |
538 | struct ip_vs_proto_data *next; | |
539 | struct ip_vs_protocol *pp; | |
540 | int *timeout_table; /* protocol timeout table */ | |
541 | atomic_t appcnt; /* counter of proto app incs. */ | |
542 | struct tcp_states_t *tcp_state_table; | |
543 | }; | |
544 | ||
5c3a0fd7 | 545 | struct ip_vs_protocol *ip_vs_proto_get(unsigned short proto); |
18d6ade6 | 546 | struct ip_vs_proto_data *ip_vs_proto_data_get(struct netns_ipvs *ipvs, |
5c3a0fd7 | 547 | unsigned short proto); |
1da177e4 | 548 | |
f11017ec | 549 | struct ip_vs_conn_param { |
e64e2b46 | 550 | struct netns_ipvs *ipvs; |
f11017ec SH |
551 | const union nf_inet_addr *caddr; |
552 | const union nf_inet_addr *vaddr; | |
553 | __be16 cport; | |
554 | __be16 vport; | |
555 | __u16 protocol; | |
556 | u16 af; | |
85999283 SH |
557 | |
558 | const struct ip_vs_pe *pe; | |
559 | char *pe_data; | |
560 | __u8 pe_data_len; | |
f11017ec SH |
561 | }; |
562 | ||
07dcc686 | 563 | /* IP_VS structure allocated for each dynamically scheduled connection */ |
1da177e4 | 564 | struct ip_vs_conn { |
731109e7 | 565 | struct hlist_node c_list; /* hashed list heads */ |
1da177e4 | 566 | /* Protocol, addresses and port numbers */ |
6e67e586 | 567 | __be16 cport; |
6e67e586 | 568 | __be16 dport; |
1845ed0b JA |
569 | __be16 vport; |
570 | u16 af; /* address family */ | |
6e67e586 HS |
571 | union nf_inet_addr caddr; /* client address */ |
572 | union nf_inet_addr vaddr; /* virtual address */ | |
573 | union nf_inet_addr daddr; /* destination address */ | |
574 | volatile __u32 flags; /* status flags */ | |
1da177e4 | 575 | __u16 protocol; /* Which protocol (TCP/UDP) */ |
ba38528a | 576 | __u16 daf; /* Address family of the dest */ |
58dbc6f2 | 577 | struct netns_ipvs *ipvs; |
1da177e4 LT |
578 | |
579 | /* counter and timer */ | |
580 | atomic_t refcnt; /* reference count */ | |
581 | struct timer_list timer; /* Expiration timer */ | |
582 | volatile unsigned long timeout; /* timeout */ | |
583 | ||
584 | /* Flags and state transition */ | |
585 | spinlock_t lock; /* lock for state transition */ | |
1da177e4 | 586 | volatile __u16 state; /* state info */ |
efac5276 RB |
587 | volatile __u16 old_state; /* old state, to be used for |
588 | * state transition triggerd | |
589 | * synchronization | |
590 | */ | |
1845ed0b | 591 | __u32 fwmark; /* Fire wall mark from skb */ |
749c42b6 | 592 | unsigned long sync_endtime; /* jiffies + sent_retries */ |
1da177e4 LT |
593 | |
594 | /* Control members */ | |
595 | struct ip_vs_conn *control; /* Master control connection */ | |
596 | atomic_t n_control; /* Number of controlled ones */ | |
597 | struct ip_vs_dest *dest; /* real server */ | |
598 | atomic_t in_pkts; /* incoming packet counter */ | |
599 | ||
07dcc686 SH |
600 | /* Packet transmitter for different forwarding methods. If it |
601 | * mangles the packet, it must return NF_DROP or better NF_STOLEN, | |
602 | * otherwise this must be changed to a sk_buff **. | |
603 | * NF_ACCEPT can be returned when destination is local. | |
1da177e4 LT |
604 | */ |
605 | int (*packet_xmit)(struct sk_buff *skb, struct ip_vs_conn *cp, | |
d4383f04 | 606 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); |
1da177e4 LT |
607 | |
608 | /* Note: we can group the following members into a structure, | |
07dcc686 SH |
609 | * in order to save more space, and the following members are |
610 | * only used in VS/NAT anyway | |
611 | */ | |
1da177e4 LT |
612 | struct ip_vs_app *app; /* bound ip_vs_app object */ |
613 | void *app_data; /* Application private data */ | |
614 | struct ip_vs_seq in_seq; /* incoming seq. struct */ | |
615 | struct ip_vs_seq out_seq; /* outgoing seq. struct */ | |
85999283 | 616 | |
e9e5eee8 | 617 | const struct ip_vs_pe *pe; |
85999283 SH |
618 | char *pe_data; |
619 | __u8 pe_data_len; | |
088339a5 JA |
620 | |
621 | struct rcu_head rcu_head; | |
1da177e4 LT |
622 | }; |
623 | ||
07dcc686 SH |
624 | /* Extended internal versions of struct ip_vs_service_user and ip_vs_dest_user |
625 | * for IPv6 support. | |
c860c6b1 | 626 | * |
07dcc686 SH |
627 | * We need these to conveniently pass around service and destination |
628 | * options, but unfortunately, we also need to keep the old definitions to | |
629 | * maintain userspace backwards compatibility for the setsockopt interface. | |
c860c6b1 JV |
630 | */ |
631 | struct ip_vs_service_user_kern { | |
632 | /* virtual service addresses */ | |
633 | u16 af; | |
634 | u16 protocol; | |
635 | union nf_inet_addr addr; /* virtual ip address */ | |
0a925864 | 636 | __be16 port; |
c860c6b1 JV |
637 | u32 fwmark; /* firwall mark of service */ |
638 | ||
639 | /* virtual service options */ | |
640 | char *sched_name; | |
0d1e71b0 | 641 | char *pe_name; |
95c96174 ED |
642 | unsigned int flags; /* virtual service flags */ |
643 | unsigned int timeout; /* persistent timeout in sec */ | |
0a925864 | 644 | __be32 netmask; /* persistent netmask or plen */ |
c860c6b1 JV |
645 | }; |
646 | ||
647 | ||
648 | struct ip_vs_dest_user_kern { | |
649 | /* destination server address */ | |
650 | union nf_inet_addr addr; | |
0a925864 | 651 | __be16 port; |
c860c6b1 JV |
652 | |
653 | /* real server options */ | |
95c96174 | 654 | unsigned int conn_flags; /* connection flags */ |
c860c6b1 JV |
655 | int weight; /* destination weight */ |
656 | ||
657 | /* thresholds for active connections */ | |
658 | u32 u_threshold; /* upper threshold */ | |
659 | u32 l_threshold; /* lower threshold */ | |
6cff339b AG |
660 | |
661 | /* Address family of addr */ | |
662 | u16 af; | |
c860c6b1 JV |
663 | }; |
664 | ||
665 | ||
1da177e4 | 666 | /* |
07dcc686 SH |
667 | * The information about the virtual service offered to the net and the |
668 | * forwarding entries. | |
1da177e4 LT |
669 | */ |
670 | struct ip_vs_service { | |
ceec4c38 JA |
671 | struct hlist_node s_list; /* for normal service table */ |
672 | struct hlist_node f_list; /* for fwmark-based service table */ | |
1da177e4 | 673 | atomic_t refcnt; /* reference counter */ |
1da177e4 | 674 | |
e7ade46a | 675 | u16 af; /* address family */ |
1da177e4 | 676 | __u16 protocol; /* which protocol (TCP/UDP) */ |
e7ade46a | 677 | union nf_inet_addr addr; /* IP address for virtual service */ |
014d730d | 678 | __be16 port; /* port number for the service */ |
1da177e4 | 679 | __u32 fwmark; /* firewall mark of the service */ |
95c96174 ED |
680 | unsigned int flags; /* service status flags */ |
681 | unsigned int timeout; /* persistent timeout in ticks */ | |
0a925864 | 682 | __be32 netmask; /* grouping granularity, mask/plen */ |
3109d2f2 | 683 | struct netns_ipvs *ipvs; |
1da177e4 LT |
684 | |
685 | struct list_head destinations; /* real server d-linked list */ | |
686 | __u32 num_dests; /* number of servers */ | |
687 | struct ip_vs_stats stats; /* statistics for the service */ | |
1da177e4 LT |
688 | |
689 | /* for scheduling */ | |
ceec4c38 | 690 | struct ip_vs_scheduler __rcu *scheduler; /* bound scheduler object */ |
ba3a3ce1 | 691 | spinlock_t sched_lock; /* lock sched_data */ |
1da177e4 | 692 | void *sched_data; /* scheduler application data */ |
85999283 SH |
693 | |
694 | /* alternate persistence engine */ | |
ceec4c38 JA |
695 | struct ip_vs_pe __rcu *pe; |
696 | ||
697 | struct rcu_head rcu_head; | |
1da177e4 LT |
698 | }; |
699 | ||
026ace06 JA |
700 | /* Information for cached dst */ |
701 | struct ip_vs_dest_dst { | |
702 | struct dst_entry *dst_cache; /* destination cache entry */ | |
703 | u32 dst_cookie; | |
704 | union nf_inet_addr dst_saddr; | |
705 | struct rcu_head rcu_head; | |
706 | }; | |
1da177e4 | 707 | |
07dcc686 SH |
708 | /* The real server destination forwarding entry with ip address, port number, |
709 | * and so on. | |
1da177e4 LT |
710 | */ |
711 | struct ip_vs_dest { | |
712 | struct list_head n_list; /* for the dests in the service */ | |
276472ea | 713 | struct hlist_node d_list; /* for table with all the dests */ |
1da177e4 | 714 | |
e7ade46a | 715 | u16 af; /* address family */ |
014d730d | 716 | __be16 port; /* port number of the server */ |
f2431e6e | 717 | union nf_inet_addr addr; /* IP address of the server */ |
95c96174 | 718 | volatile unsigned int flags; /* dest status flags */ |
1da177e4 LT |
719 | atomic_t conn_flags; /* flags to copy to conn */ |
720 | atomic_t weight; /* server weight */ | |
721 | ||
722 | atomic_t refcnt; /* reference counter */ | |
723 | struct ip_vs_stats stats; /* statistics */ | |
bcbde4c0 | 724 | unsigned long idle_start; /* start time, jiffies */ |
1da177e4 LT |
725 | |
726 | /* connection counters and thresholds */ | |
727 | atomic_t activeconns; /* active connections */ | |
728 | atomic_t inactconns; /* inactive connections */ | |
729 | atomic_t persistconns; /* persistent connections */ | |
730 | __u32 u_threshold; /* upper threshold */ | |
731 | __u32 l_threshold; /* lower threshold */ | |
732 | ||
733 | /* for destination cache */ | |
734 | spinlock_t dst_lock; /* lock of dst_cache */ | |
026ace06 | 735 | struct ip_vs_dest_dst __rcu *dest_dst; /* cached dst info */ |
1da177e4 LT |
736 | |
737 | /* for virtual service */ | |
bcbde4c0 | 738 | struct ip_vs_service __rcu *svc; /* service it belongs to */ |
1da177e4 | 739 | __u16 protocol; /* which protocol (TCP/UDP) */ |
014d730d | 740 | __be16 vport; /* virtual port number */ |
f2431e6e | 741 | union nf_inet_addr vaddr; /* virtual IP address */ |
1da177e4 | 742 | __u32 vfwmark; /* firewall mark of service */ |
276472ea | 743 | |
578bc3ef | 744 | struct list_head t_list; /* in dest_trash */ |
276472ea | 745 | unsigned int in_rs_table:1; /* we are in rs_table */ |
1da177e4 LT |
746 | }; |
747 | ||
07dcc686 | 748 | /* The scheduler object */ |
1da177e4 LT |
749 | struct ip_vs_scheduler { |
750 | struct list_head n_list; /* d-linked list head */ | |
751 | char *name; /* scheduler name */ | |
752 | atomic_t refcnt; /* reference counter */ | |
753 | struct module *module; /* THIS_MODULE/NULL */ | |
754 | ||
755 | /* scheduler initializing service */ | |
756 | int (*init_service)(struct ip_vs_service *svc); | |
757 | /* scheduling service finish */ | |
ed3ffc4e | 758 | void (*done_service)(struct ip_vs_service *svc); |
6b6df466 JA |
759 | /* dest is linked */ |
760 | int (*add_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest); | |
761 | /* dest is unlinked */ | |
762 | int (*del_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest); | |
763 | /* dest is updated */ | |
764 | int (*upd_dest)(struct ip_vs_service *svc, struct ip_vs_dest *dest); | |
1da177e4 LT |
765 | |
766 | /* selecting a server from the given service */ | |
767 | struct ip_vs_dest* (*schedule)(struct ip_vs_service *svc, | |
bba54de5 JA |
768 | const struct sk_buff *skb, |
769 | struct ip_vs_iphdr *iph); | |
1da177e4 LT |
770 | }; |
771 | ||
85999283 SH |
772 | /* The persistence engine object */ |
773 | struct ip_vs_pe { | |
774 | struct list_head n_list; /* d-linked list head */ | |
775 | char *name; /* scheduler name */ | |
776 | atomic_t refcnt; /* reference counter */ | |
777 | struct module *module; /* THIS_MODULE/NULL */ | |
778 | ||
779 | /* get the connection template, if any */ | |
780 | int (*fill_param)(struct ip_vs_conn_param *p, struct sk_buff *skb); | |
781 | bool (*ct_match)(const struct ip_vs_conn_param *p, | |
782 | struct ip_vs_conn *ct); | |
783 | u32 (*hashkey_raw)(const struct ip_vs_conn_param *p, u32 initval, | |
784 | bool inverse); | |
a3c918ac | 785 | int (*show_pe_data)(const struct ip_vs_conn *cp, char *buf); |
85999283 | 786 | }; |
1da177e4 | 787 | |
07dcc686 | 788 | /* The application module object (a.k.a. app incarnation) */ |
fd2c3ef7 | 789 | struct ip_vs_app { |
1da177e4 LT |
790 | struct list_head a_list; /* member in app list */ |
791 | int type; /* IP_VS_APP_TYPE_xxx */ | |
792 | char *name; /* application module name */ | |
793 | __u16 protocol; | |
794 | struct module *module; /* THIS_MODULE/NULL */ | |
795 | struct list_head incs_list; /* list of incarnations */ | |
796 | ||
797 | /* members for application incarnations */ | |
798 | struct list_head p_list; /* member in proto app list */ | |
799 | struct ip_vs_app *app; /* its real application */ | |
014d730d | 800 | __be16 port; /* port number in net order */ |
1da177e4 | 801 | atomic_t usecnt; /* usage counter */ |
363c97d7 | 802 | struct rcu_head rcu_head; |
1da177e4 | 803 | |
07dcc686 | 804 | /* output hook: Process packet in inout direction, diff set for TCP. |
8b27b10f JA |
805 | * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok, |
806 | * 2=Mangled but checksum was not updated | |
807 | */ | |
1da177e4 | 808 | int (*pkt_out)(struct ip_vs_app *, struct ip_vs_conn *, |
3db05fea | 809 | struct sk_buff *, int *diff); |
1da177e4 | 810 | |
07dcc686 | 811 | /* input hook: Process packet in outin direction, diff set for TCP. |
8b27b10f JA |
812 | * Return: 0=Error, 1=Payload Not Mangled/Mangled but checksum is ok, |
813 | * 2=Mangled but checksum was not updated | |
814 | */ | |
1da177e4 | 815 | int (*pkt_in)(struct ip_vs_app *, struct ip_vs_conn *, |
3db05fea | 816 | struct sk_buff *, int *diff); |
1da177e4 LT |
817 | |
818 | /* ip_vs_app initializer */ | |
819 | int (*init_conn)(struct ip_vs_app *, struct ip_vs_conn *); | |
820 | ||
821 | /* ip_vs_app finish */ | |
822 | int (*done_conn)(struct ip_vs_app *, struct ip_vs_conn *); | |
823 | ||
824 | ||
825 | /* not used now */ | |
826 | int (*bind_conn)(struct ip_vs_app *, struct ip_vs_conn *, | |
827 | struct ip_vs_protocol *); | |
828 | ||
829 | void (*unbind_conn)(struct ip_vs_app *, struct ip_vs_conn *); | |
830 | ||
831 | int * timeout_table; | |
832 | int * timeouts; | |
833 | int timeouts_size; | |
834 | ||
835 | int (*conn_schedule)(struct sk_buff *skb, struct ip_vs_app *app, | |
836 | int *verdict, struct ip_vs_conn **cpp); | |
837 | ||
838 | struct ip_vs_conn * | |
839 | (*conn_in_get)(const struct sk_buff *skb, struct ip_vs_app *app, | |
d4383f04 | 840 | const struct iphdr *iph, int inverse); |
1da177e4 LT |
841 | |
842 | struct ip_vs_conn * | |
843 | (*conn_out_get)(const struct sk_buff *skb, struct ip_vs_app *app, | |
d4383f04 | 844 | const struct iphdr *iph, int inverse); |
1da177e4 LT |
845 | |
846 | int (*state_transition)(struct ip_vs_conn *cp, int direction, | |
847 | const struct sk_buff *skb, | |
848 | struct ip_vs_app *app); | |
849 | ||
850 | void (*timeout_change)(struct ip_vs_app *app, int flags); | |
851 | }; | |
852 | ||
f73181c8 PNA |
853 | struct ipvs_master_sync_state { |
854 | struct list_head sync_queue; | |
855 | struct ip_vs_sync_buff *sync_buff; | |
07995674 | 856 | unsigned long sync_queue_len; |
f73181c8 PNA |
857 | unsigned int sync_queue_delay; |
858 | struct task_struct *master_thread; | |
859 | struct delayed_work master_wakeup_work; | |
860 | struct netns_ipvs *ipvs; | |
861 | }; | |
862 | ||
578bc3ef JA |
863 | /* How much time to keep dests in trash */ |
864 | #define IP_VS_DEST_TRASH_PERIOD (120 * HZ) | |
865 | ||
e4ff6751 | 866 | struct ipvs_sync_daemon_cfg { |
d3328817 | 867 | union nf_inet_addr mcast_group; |
e4ff6751 JA |
868 | int syncid; |
869 | u16 sync_maxlen; | |
d3328817 JA |
870 | u16 mcast_port; |
871 | u8 mcast_af; | |
872 | u8 mcast_ttl; | |
e4ff6751 JA |
873 | /* multicast interface name */ |
874 | char mcast_ifn[IP_VS_IFNAME_MAXLEN]; | |
875 | }; | |
876 | ||
2553d064 JA |
877 | /* IPVS in network namespace */ |
878 | struct netns_ipvs { | |
879 | int gen; /* Generation */ | |
7a4f0761 | 880 | int enable; /* enable like nf_hooks do */ |
07dcc686 | 881 | /* Hash table: for real service lookups */ |
2553d064 JA |
882 | #define IP_VS_RTAB_BITS 4 |
883 | #define IP_VS_RTAB_SIZE (1 << IP_VS_RTAB_BITS) | |
884 | #define IP_VS_RTAB_MASK (IP_VS_RTAB_SIZE - 1) | |
885 | ||
276472ea | 886 | struct hlist_head rs_table[IP_VS_RTAB_SIZE]; |
2553d064 JA |
887 | /* ip_vs_app */ |
888 | struct list_head app_list; | |
2553d064 JA |
889 | /* ip_vs_proto */ |
890 | #define IP_VS_PROTO_TAB_SIZE 32 /* must be power of 2 */ | |
891 | struct ip_vs_proto_data *proto_data_table[IP_VS_PROTO_TAB_SIZE]; | |
892 | /* ip_vs_proto_tcp */ | |
893 | #ifdef CONFIG_IP_VS_PROTO_TCP | |
894 | #define TCP_APP_TAB_BITS 4 | |
895 | #define TCP_APP_TAB_SIZE (1 << TCP_APP_TAB_BITS) | |
896 | #define TCP_APP_TAB_MASK (TCP_APP_TAB_SIZE - 1) | |
897 | struct list_head tcp_apps[TCP_APP_TAB_SIZE]; | |
2553d064 JA |
898 | #endif |
899 | /* ip_vs_proto_udp */ | |
900 | #ifdef CONFIG_IP_VS_PROTO_UDP | |
901 | #define UDP_APP_TAB_BITS 4 | |
902 | #define UDP_APP_TAB_SIZE (1 << UDP_APP_TAB_BITS) | |
903 | #define UDP_APP_TAB_MASK (UDP_APP_TAB_SIZE - 1) | |
904 | struct list_head udp_apps[UDP_APP_TAB_SIZE]; | |
2553d064 JA |
905 | #endif |
906 | /* ip_vs_proto_sctp */ | |
907 | #ifdef CONFIG_IP_VS_PROTO_SCTP | |
908 | #define SCTP_APP_TAB_BITS 4 | |
909 | #define SCTP_APP_TAB_SIZE (1 << SCTP_APP_TAB_BITS) | |
910 | #define SCTP_APP_TAB_MASK (SCTP_APP_TAB_SIZE - 1) | |
911 | /* Hash table for SCTP application incarnations */ | |
912 | struct list_head sctp_apps[SCTP_APP_TAB_SIZE]; | |
2553d064 JA |
913 | #endif |
914 | /* ip_vs_conn */ | |
07dcc686 | 915 | atomic_t conn_count; /* connection counter */ |
2553d064 JA |
916 | |
917 | /* ip_vs_ctl */ | |
2a0751af | 918 | struct ip_vs_stats tot_stats; /* Statistics & est. */ |
2553d064 JA |
919 | |
920 | int num_services; /* no of virtual services */ | |
f2247fbd | 921 | |
f2247fbd SH |
922 | /* Trash for destinations */ |
923 | struct list_head dest_trash; | |
578bc3ef JA |
924 | spinlock_t dest_trash_lock; |
925 | struct timer_list dest_trash_timer; /* expiration timer */ | |
f2247fbd SH |
926 | /* Service counters */ |
927 | atomic_t ftpsvc_counter; | |
928 | atomic_t nullsvc_counter; | |
929 | ||
930 | #ifdef CONFIG_SYSCTL | |
2553d064 JA |
931 | /* 1/rate drop and drop-entry variables */ |
932 | struct delayed_work defense_work; /* Work handler */ | |
933 | int drop_rate; | |
934 | int drop_counter; | |
935 | atomic_t dropentry; | |
936 | /* locks in ctl.c */ | |
937 | spinlock_t dropentry_lock; /* drop entry handling */ | |
938 | spinlock_t droppacket_lock; /* drop packet handling */ | |
939 | spinlock_t securetcp_lock; /* state and timeout tables */ | |
2553d064 JA |
940 | |
941 | /* sys-ctl struct */ | |
942 | struct ctl_table_header *sysctl_hdr; | |
943 | struct ctl_table *sysctl_tbl; | |
f2247fbd SH |
944 | #endif |
945 | ||
2553d064 JA |
946 | /* sysctl variables */ |
947 | int sysctl_amemthresh; | |
948 | int sysctl_am_droprate; | |
949 | int sysctl_drop_entry; | |
950 | int sysctl_drop_packet; | |
951 | int sysctl_secure_tcp; | |
952 | #ifdef CONFIG_IP_VS_NFCT | |
953 | int sysctl_conntrack; | |
954 | #endif | |
955 | int sysctl_snat_reroute; | |
956 | int sysctl_sync_ver; | |
f73181c8 | 957 | int sysctl_sync_ports; |
4d0c875d | 958 | int sysctl_sync_persist_mode; |
07995674 | 959 | unsigned long sysctl_sync_qlen_max; |
1c003b15 | 960 | int sysctl_sync_sock_size; |
2553d064 JA |
961 | int sysctl_cache_bypass; |
962 | int sysctl_expire_nodest_conn; | |
c6c96c18 AF |
963 | int sysctl_sloppy_tcp; |
964 | int sysctl_sloppy_sctp; | |
2553d064 JA |
965 | int sysctl_expire_quiescent_template; |
966 | int sysctl_sync_threshold[2]; | |
749c42b6 JA |
967 | unsigned int sysctl_sync_refresh_period; |
968 | int sysctl_sync_retries; | |
2553d064 | 969 | int sysctl_nat_icmp_send; |
3654e611 | 970 | int sysctl_pmtu_disc; |
0c12582f | 971 | int sysctl_backup_only; |
d752c364 | 972 | int sysctl_conn_reuse_mode; |
94485fed | 973 | int sysctl_schedule_icmp; |
4e478098 | 974 | int sysctl_ignore_tunneled; |
2553d064 JA |
975 | |
976 | /* ip_vs_lblc */ | |
977 | int sysctl_lblc_expiration; | |
978 | struct ctl_table_header *lblc_ctl_header; | |
979 | struct ctl_table *lblc_ctl_table; | |
980 | /* ip_vs_lblcr */ | |
981 | int sysctl_lblcr_expiration; | |
982 | struct ctl_table_header *lblcr_ctl_header; | |
983 | struct ctl_table *lblcr_ctl_table; | |
984 | /* ip_vs_est */ | |
985 | struct list_head est_list; /* estimator list */ | |
986 | spinlock_t est_lock; | |
987 | struct timer_list est_timer; /* Estimation timer */ | |
988 | /* ip_vs_sync */ | |
2553d064 | 989 | spinlock_t sync_lock; |
f73181c8 | 990 | struct ipvs_master_sync_state *ms; |
2553d064 | 991 | spinlock_t sync_buff_lock; |
f73181c8 PNA |
992 | struct task_struct **backup_threads; |
993 | int threads_mask; | |
2553d064 | 994 | volatile int sync_state; |
ae1d48b2 | 995 | struct mutex sync_mutex; |
e4ff6751 JA |
996 | struct ipvs_sync_daemon_cfg mcfg; /* Master Configuration */ |
997 | struct ipvs_sync_daemon_cfg bcfg; /* Backup Configuration */ | |
2553d064 JA |
998 | /* net name space ptr */ |
999 | struct net *net; /* Needed by timer routines */ | |
07dcc686 SH |
1000 | /* Number of heterogeneous destinations, needed becaus heterogeneous |
1001 | * are not supported when synchronization is enabled. | |
1002 | */ | |
391f503d | 1003 | unsigned int mixed_address_family_dests; |
2553d064 | 1004 | }; |
1da177e4 | 1005 | |
59e0350e SH |
1006 | #define DEFAULT_SYNC_THRESHOLD 3 |
1007 | #define DEFAULT_SYNC_PERIOD 50 | |
7532e8d4 | 1008 | #define DEFAULT_SYNC_VER 1 |
c6c96c18 AF |
1009 | #define DEFAULT_SLOPPY_TCP 0 |
1010 | #define DEFAULT_SLOPPY_SCTP 0 | |
749c42b6 JA |
1011 | #define DEFAULT_SYNC_REFRESH_PERIOD (0U * HZ) |
1012 | #define DEFAULT_SYNC_RETRIES 0 | |
1c003b15 PNA |
1013 | #define IPVS_SYNC_WAKEUP_RATE 8 |
1014 | #define IPVS_SYNC_QLEN_MAX (IPVS_SYNC_WAKEUP_RATE * 4) | |
1015 | #define IPVS_SYNC_SEND_DELAY (HZ / 50) | |
1016 | #define IPVS_SYNC_CHECK_PERIOD HZ | |
749c42b6 | 1017 | #define IPVS_SYNC_FLUSH_TIME (HZ * 2) |
f73181c8 | 1018 | #define IPVS_SYNC_PORTS_MAX (1 << 6) |
59e0350e SH |
1019 | |
1020 | #ifdef CONFIG_SYSCTL | |
1021 | ||
1022 | static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs) | |
1023 | { | |
1024 | return ipvs->sysctl_sync_threshold[0]; | |
1025 | } | |
1026 | ||
1027 | static inline int sysctl_sync_period(struct netns_ipvs *ipvs) | |
1028 | { | |
749c42b6 JA |
1029 | return ACCESS_ONCE(ipvs->sysctl_sync_threshold[1]); |
1030 | } | |
1031 | ||
1032 | static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs) | |
1033 | { | |
1034 | return ACCESS_ONCE(ipvs->sysctl_sync_refresh_period); | |
1035 | } | |
1036 | ||
1037 | static inline int sysctl_sync_retries(struct netns_ipvs *ipvs) | |
1038 | { | |
1039 | return ipvs->sysctl_sync_retries; | |
59e0350e SH |
1040 | } |
1041 | ||
7532e8d4 SH |
1042 | static inline int sysctl_sync_ver(struct netns_ipvs *ipvs) |
1043 | { | |
1044 | return ipvs->sysctl_sync_ver; | |
1045 | } | |
1046 | ||
c6c96c18 AF |
1047 | static inline int sysctl_sloppy_tcp(struct netns_ipvs *ipvs) |
1048 | { | |
1049 | return ipvs->sysctl_sloppy_tcp; | |
1050 | } | |
1051 | ||
1052 | static inline int sysctl_sloppy_sctp(struct netns_ipvs *ipvs) | |
1053 | { | |
1054 | return ipvs->sysctl_sloppy_sctp; | |
1055 | } | |
1056 | ||
f73181c8 PNA |
1057 | static inline int sysctl_sync_ports(struct netns_ipvs *ipvs) |
1058 | { | |
1059 | return ACCESS_ONCE(ipvs->sysctl_sync_ports); | |
1060 | } | |
1061 | ||
4d0c875d JA |
1062 | static inline int sysctl_sync_persist_mode(struct netns_ipvs *ipvs) |
1063 | { | |
1064 | return ipvs->sysctl_sync_persist_mode; | |
1065 | } | |
1066 | ||
07995674 | 1067 | static inline unsigned long sysctl_sync_qlen_max(struct netns_ipvs *ipvs) |
1c003b15 PNA |
1068 | { |
1069 | return ipvs->sysctl_sync_qlen_max; | |
1070 | } | |
1071 | ||
1072 | static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs) | |
1073 | { | |
1074 | return ipvs->sysctl_sync_sock_size; | |
1075 | } | |
1076 | ||
3654e611 JA |
1077 | static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs) |
1078 | { | |
1079 | return ipvs->sysctl_pmtu_disc; | |
1080 | } | |
1081 | ||
0c12582f JA |
1082 | static inline int sysctl_backup_only(struct netns_ipvs *ipvs) |
1083 | { | |
1084 | return ipvs->sync_state & IP_VS_STATE_BACKUP && | |
1085 | ipvs->sysctl_backup_only; | |
1086 | } | |
1087 | ||
d752c364 MRL |
1088 | static inline int sysctl_conn_reuse_mode(struct netns_ipvs *ipvs) |
1089 | { | |
1090 | return ipvs->sysctl_conn_reuse_mode; | |
1091 | } | |
1092 | ||
94485fed AG |
1093 | static inline int sysctl_schedule_icmp(struct netns_ipvs *ipvs) |
1094 | { | |
1095 | return ipvs->sysctl_schedule_icmp; | |
1096 | } | |
1097 | ||
4e478098 AG |
1098 | static inline int sysctl_ignore_tunneled(struct netns_ipvs *ipvs) |
1099 | { | |
1100 | return ipvs->sysctl_ignore_tunneled; | |
1101 | } | |
1102 | ||
59e0350e SH |
1103 | #else |
1104 | ||
1105 | static inline int sysctl_sync_threshold(struct netns_ipvs *ipvs) | |
1106 | { | |
1107 | return DEFAULT_SYNC_THRESHOLD; | |
1108 | } | |
1109 | ||
1110 | static inline int sysctl_sync_period(struct netns_ipvs *ipvs) | |
1111 | { | |
1112 | return DEFAULT_SYNC_PERIOD; | |
1113 | } | |
1114 | ||
749c42b6 JA |
1115 | static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs) |
1116 | { | |
1117 | return DEFAULT_SYNC_REFRESH_PERIOD; | |
1118 | } | |
1119 | ||
1120 | static inline int sysctl_sync_retries(struct netns_ipvs *ipvs) | |
1121 | { | |
1122 | return DEFAULT_SYNC_RETRIES & 3; | |
1123 | } | |
1124 | ||
7532e8d4 SH |
1125 | static inline int sysctl_sync_ver(struct netns_ipvs *ipvs) |
1126 | { | |
1127 | return DEFAULT_SYNC_VER; | |
1128 | } | |
1129 | ||
c6c96c18 AF |
1130 | static inline int sysctl_sloppy_tcp(struct netns_ipvs *ipvs) |
1131 | { | |
1132 | return DEFAULT_SLOPPY_TCP; | |
1133 | } | |
1134 | ||
1135 | static inline int sysctl_sloppy_sctp(struct netns_ipvs *ipvs) | |
1136 | { | |
1137 | return DEFAULT_SLOPPY_SCTP; | |
1138 | } | |
1139 | ||
f73181c8 PNA |
1140 | static inline int sysctl_sync_ports(struct netns_ipvs *ipvs) |
1141 | { | |
1142 | return 1; | |
1143 | } | |
1144 | ||
4d0c875d JA |
1145 | static inline int sysctl_sync_persist_mode(struct netns_ipvs *ipvs) |
1146 | { | |
1147 | return 0; | |
1148 | } | |
1149 | ||
07995674 | 1150 | static inline unsigned long sysctl_sync_qlen_max(struct netns_ipvs *ipvs) |
1c003b15 PNA |
1151 | { |
1152 | return IPVS_SYNC_QLEN_MAX; | |
1153 | } | |
1154 | ||
1155 | static inline int sysctl_sync_sock_size(struct netns_ipvs *ipvs) | |
1156 | { | |
1157 | return 0; | |
1158 | } | |
1159 | ||
3654e611 JA |
1160 | static inline int sysctl_pmtu_disc(struct netns_ipvs *ipvs) |
1161 | { | |
1162 | return 1; | |
1163 | } | |
1164 | ||
0c12582f JA |
1165 | static inline int sysctl_backup_only(struct netns_ipvs *ipvs) |
1166 | { | |
1167 | return 0; | |
1168 | } | |
1169 | ||
d752c364 MRL |
1170 | static inline int sysctl_conn_reuse_mode(struct netns_ipvs *ipvs) |
1171 | { | |
1172 | return 1; | |
1173 | } | |
1174 | ||
94485fed AG |
1175 | static inline int sysctl_schedule_icmp(struct netns_ipvs *ipvs) |
1176 | { | |
1177 | return 0; | |
1178 | } | |
1179 | ||
4e478098 AG |
1180 | static inline int sysctl_ignore_tunneled(struct netns_ipvs *ipvs) |
1181 | { | |
1182 | return 0; | |
1183 | } | |
1184 | ||
59e0350e SH |
1185 | #endif |
1186 | ||
07dcc686 SH |
1187 | /* IPVS core functions |
1188 | * (from ip_vs_core.c) | |
1da177e4 | 1189 | */ |
5c3a0fd7 JP |
1190 | const char *ip_vs_proto_name(unsigned int proto); |
1191 | void ip_vs_init_hash_table(struct list_head *table, int rows); | |
afdd6140 | 1192 | #define IP_VS_INIT_HASH_TABLE(t) ip_vs_init_hash_table((t), ARRAY_SIZE((t))) |
1da177e4 | 1193 | |
1da177e4 LT |
1194 | #define IP_VS_APP_TYPE_FTP 1 |
1195 | ||
07dcc686 SH |
1196 | /* ip_vs_conn handling functions |
1197 | * (from ip_vs_conn.c) | |
1da177e4 | 1198 | */ |
1da177e4 LT |
1199 | enum { |
1200 | IP_VS_DIR_INPUT = 0, | |
1201 | IP_VS_DIR_OUTPUT, | |
1202 | IP_VS_DIR_INPUT_ONLY, | |
1203 | IP_VS_DIR_LAST, | |
1204 | }; | |
1205 | ||
19913dec | 1206 | static inline void ip_vs_conn_fill_param(struct netns_ipvs *ipvs, int af, int protocol, |
f11017ec SH |
1207 | const union nf_inet_addr *caddr, |
1208 | __be16 cport, | |
1209 | const union nf_inet_addr *vaddr, | |
1210 | __be16 vport, | |
1211 | struct ip_vs_conn_param *p) | |
1212 | { | |
19913dec | 1213 | p->ipvs = ipvs; |
f11017ec SH |
1214 | p->af = af; |
1215 | p->protocol = protocol; | |
1216 | p->caddr = caddr; | |
1217 | p->cport = cport; | |
1218 | p->vaddr = vaddr; | |
1219 | p->vport = vport; | |
85999283 SH |
1220 | p->pe = NULL; |
1221 | p->pe_data = NULL; | |
f11017ec | 1222 | } |
28364a59 | 1223 | |
f11017ec SH |
1224 | struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p); |
1225 | struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p); | |
28364a59 | 1226 | |
ab161976 EB |
1227 | struct ip_vs_conn * ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af, |
1228 | const struct sk_buff *skb, | |
802c41ad | 1229 | const struct ip_vs_iphdr *iph); |
5c0d2374 | 1230 | |
f11017ec | 1231 | struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p); |
1da177e4 | 1232 | |
0cf705c8 EB |
1233 | struct ip_vs_conn * ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af, |
1234 | const struct sk_buff *skb, | |
802c41ad | 1235 | const struct ip_vs_iphdr *iph); |
5c0d2374 | 1236 | |
088339a5 JA |
1237 | /* Get reference to gain full access to conn. |
1238 | * By default, RCU read-side critical sections have access only to | |
1239 | * conn fields and its PE data, see ip_vs_conn_rcu_free() for reference. | |
1240 | */ | |
1241 | static inline bool __ip_vs_conn_get(struct ip_vs_conn *cp) | |
1242 | { | |
1243 | return atomic_inc_not_zero(&cp->refcnt); | |
1244 | } | |
1245 | ||
1da177e4 LT |
1246 | /* put back the conn without restarting its timer */ |
1247 | static inline void __ip_vs_conn_put(struct ip_vs_conn *cp) | |
1248 | { | |
4e857c58 | 1249 | smp_mb__before_atomic(); |
1da177e4 LT |
1250 | atomic_dec(&cp->refcnt); |
1251 | } | |
5c3a0fd7 JP |
1252 | void ip_vs_conn_put(struct ip_vs_conn *cp); |
1253 | void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport); | |
1da177e4 | 1254 | |
ba38528a | 1255 | struct ip_vs_conn *ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af, |
f11017ec | 1256 | const union nf_inet_addr *daddr, |
95c96174 | 1257 | __be16 dport, unsigned int flags, |
0e051e68 | 1258 | struct ip_vs_dest *dest, __u32 fwmark); |
5c3a0fd7 | 1259 | void ip_vs_conn_expire_now(struct ip_vs_conn *cp); |
1da177e4 | 1260 | |
5c3a0fd7 | 1261 | const char *ip_vs_state_name(__u16 proto, int state); |
1da177e4 | 1262 | |
5c3a0fd7 JP |
1263 | void ip_vs_tcp_conn_listen(struct net *net, struct ip_vs_conn *cp); |
1264 | int ip_vs_check_template(struct ip_vs_conn *ct); | |
423b5595 | 1265 | void ip_vs_random_dropentry(struct netns_ipvs *ipvs); |
5c3a0fd7 JP |
1266 | int ip_vs_conn_init(void); |
1267 | void ip_vs_conn_cleanup(void); | |
1da177e4 LT |
1268 | |
1269 | static inline void ip_vs_control_del(struct ip_vs_conn *cp) | |
1270 | { | |
1271 | struct ip_vs_conn *ctl_cp = cp->control; | |
1272 | if (!ctl_cp) { | |
cfc78c5a JV |
1273 | IP_VS_ERR_BUF("request control DEL for uncontrolled: " |
1274 | "%s:%d to %s:%d\n", | |
1275 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1276 | ntohs(cp->cport), | |
1277 | IP_VS_DBG_ADDR(cp->af, &cp->vaddr), | |
1278 | ntohs(cp->vport)); | |
1279 | ||
1da177e4 LT |
1280 | return; |
1281 | } | |
1282 | ||
cfc78c5a JV |
1283 | IP_VS_DBG_BUF(7, "DELeting control for: " |
1284 | "cp.dst=%s:%d ctl_cp.dst=%s:%d\n", | |
1285 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1286 | ntohs(cp->cport), | |
1287 | IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr), | |
1288 | ntohs(ctl_cp->cport)); | |
1da177e4 LT |
1289 | |
1290 | cp->control = NULL; | |
1291 | if (atomic_read(&ctl_cp->n_control) == 0) { | |
cfc78c5a JV |
1292 | IP_VS_ERR_BUF("BUG control DEL with n=0 : " |
1293 | "%s:%d to %s:%d\n", | |
1294 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1295 | ntohs(cp->cport), | |
1296 | IP_VS_DBG_ADDR(cp->af, &cp->vaddr), | |
1297 | ntohs(cp->vport)); | |
1298 | ||
1da177e4 LT |
1299 | return; |
1300 | } | |
1301 | atomic_dec(&ctl_cp->n_control); | |
1302 | } | |
1303 | ||
1304 | static inline void | |
1305 | ip_vs_control_add(struct ip_vs_conn *cp, struct ip_vs_conn *ctl_cp) | |
1306 | { | |
1307 | if (cp->control) { | |
cfc78c5a JV |
1308 | IP_VS_ERR_BUF("request control ADD for already controlled: " |
1309 | "%s:%d to %s:%d\n", | |
1310 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1311 | ntohs(cp->cport), | |
1312 | IP_VS_DBG_ADDR(cp->af, &cp->vaddr), | |
1313 | ntohs(cp->vport)); | |
1314 | ||
1da177e4 LT |
1315 | ip_vs_control_del(cp); |
1316 | } | |
1317 | ||
cfc78c5a JV |
1318 | IP_VS_DBG_BUF(7, "ADDing control for: " |
1319 | "cp.dst=%s:%d ctl_cp.dst=%s:%d\n", | |
1320 | IP_VS_DBG_ADDR(cp->af, &cp->caddr), | |
1321 | ntohs(cp->cport), | |
1322 | IP_VS_DBG_ADDR(cp->af, &ctl_cp->caddr), | |
1323 | ntohs(ctl_cp->cport)); | |
1da177e4 LT |
1324 | |
1325 | cp->control = ctl_cp; | |
1326 | atomic_inc(&ctl_cp->n_control); | |
1327 | } | |
1328 | ||
07dcc686 | 1329 | /* IPVS netns init & cleanup functions */ |
a4dd0360 | 1330 | int ip_vs_estimator_net_init(struct netns_ipvs *ipvs); |
3d993766 | 1331 | int ip_vs_control_net_init(struct netns_ipvs *ipvs); |
5c3a0fd7 | 1332 | int ip_vs_protocol_net_init(struct net *net); |
b5dd212c | 1333 | int ip_vs_app_net_init(struct netns_ipvs *ipvs); |
5c3a0fd7 | 1334 | int ip_vs_conn_net_init(struct net *net); |
802cb437 | 1335 | int ip_vs_sync_net_init(struct netns_ipvs *ipvs); |
5c3a0fd7 | 1336 | void ip_vs_conn_net_cleanup(struct net *net); |
b5dd212c | 1337 | void ip_vs_app_net_cleanup(struct netns_ipvs *ipvs); |
5c3a0fd7 | 1338 | void ip_vs_protocol_net_cleanup(struct net *net); |
3d993766 | 1339 | void ip_vs_control_net_cleanup(struct netns_ipvs *ipvs); |
a4dd0360 | 1340 | void ip_vs_estimator_net_cleanup(struct netns_ipvs *ipvs); |
ebea1f7c | 1341 | void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs); |
56d2169b | 1342 | void ip_vs_service_net_cleanup(struct netns_ipvs *ipvs); |
1da177e4 | 1343 | |
07dcc686 SH |
1344 | /* IPVS application functions |
1345 | * (from ip_vs_app.c) | |
1da177e4 LT |
1346 | */ |
1347 | #define IP_VS_APP_MAX_PORTS 8 | |
9f8128a5 EB |
1348 | struct ip_vs_app *register_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app); |
1349 | void unregister_ip_vs_app(struct netns_ipvs *ipvs, struct ip_vs_app *app); | |
5c3a0fd7 JP |
1350 | int ip_vs_bind_app(struct ip_vs_conn *cp, struct ip_vs_protocol *pp); |
1351 | void ip_vs_unbind_app(struct ip_vs_conn *cp); | |
3250dc9c | 1352 | int register_ip_vs_app_inc(struct netns_ipvs *ipvs, struct ip_vs_app *app, __u16 proto, |
5c3a0fd7 JP |
1353 | __u16 port); |
1354 | int ip_vs_app_inc_get(struct ip_vs_app *inc); | |
1355 | void ip_vs_app_inc_put(struct ip_vs_app *inc); | |
1356 | ||
1357 | int ip_vs_app_pkt_out(struct ip_vs_conn *, struct sk_buff *skb); | |
1358 | int ip_vs_app_pkt_in(struct ip_vs_conn *, struct sk_buff *skb); | |
1da177e4 | 1359 | |
8be67a66 SH |
1360 | int register_ip_vs_pe(struct ip_vs_pe *pe); |
1361 | int unregister_ip_vs_pe(struct ip_vs_pe *pe); | |
e9e5eee8 | 1362 | struct ip_vs_pe *ip_vs_pe_getbyname(const char *name); |
fe5e7a1e | 1363 | struct ip_vs_pe *__ip_vs_pe_getbyname(const char *pe_name); |
e9e5eee8 | 1364 | |
07dcc686 | 1365 | /* Use a #define to avoid all of module.h just for these trivial ops */ |
69e7dae4 PG |
1366 | #define ip_vs_pe_get(pe) \ |
1367 | if (pe && pe->module) \ | |
e9e5eee8 | 1368 | __module_get(pe->module); |
e9e5eee8 | 1369 | |
69e7dae4 PG |
1370 | #define ip_vs_pe_put(pe) \ |
1371 | if (pe && pe->module) \ | |
e9e5eee8 | 1372 | module_put(pe->module); |
1da177e4 | 1373 | |
07dcc686 | 1374 | /* IPVS protocol functions (from ip_vs_proto.c) */ |
5c3a0fd7 JP |
1375 | int ip_vs_protocol_init(void); |
1376 | void ip_vs_protocol_cleanup(void); | |
1377 | void ip_vs_protocol_timeout_change(struct netns_ipvs *ipvs, int flags); | |
1378 | int *ip_vs_create_timeout_table(int *table, int size); | |
1379 | int ip_vs_set_state_timeout(int *table, int num, const char *const *names, | |
1380 | const char *name, int to); | |
1381 | void ip_vs_tcpudp_debug_packet(int af, struct ip_vs_protocol *pp, | |
1382 | const struct sk_buff *skb, int offset, | |
1383 | const char *msg); | |
1da177e4 LT |
1384 | |
1385 | extern struct ip_vs_protocol ip_vs_protocol_tcp; | |
1386 | extern struct ip_vs_protocol ip_vs_protocol_udp; | |
1387 | extern struct ip_vs_protocol ip_vs_protocol_icmp; | |
1388 | extern struct ip_vs_protocol ip_vs_protocol_esp; | |
1389 | extern struct ip_vs_protocol ip_vs_protocol_ah; | |
2906f66a | 1390 | extern struct ip_vs_protocol ip_vs_protocol_sctp; |
1da177e4 | 1391 | |
07dcc686 SH |
1392 | /* Registering/unregistering scheduler functions |
1393 | * (from ip_vs_sched.c) | |
1da177e4 | 1394 | */ |
5c3a0fd7 JP |
1395 | int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler); |
1396 | int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler); | |
1397 | int ip_vs_bind_scheduler(struct ip_vs_service *svc, | |
1398 | struct ip_vs_scheduler *scheduler); | |
1399 | void ip_vs_unbind_scheduler(struct ip_vs_service *svc, | |
1400 | struct ip_vs_scheduler *sched); | |
1401 | struct ip_vs_scheduler *ip_vs_scheduler_get(const char *sched_name); | |
1402 | void ip_vs_scheduler_put(struct ip_vs_scheduler *scheduler); | |
1403 | struct ip_vs_conn * | |
190ecd27 | 1404 | ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb, |
d4383f04 JDB |
1405 | struct ip_vs_proto_data *pd, int *ignored, |
1406 | struct ip_vs_iphdr *iph); | |
5c3a0fd7 JP |
1407 | int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb, |
1408 | struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph); | |
1da177e4 | 1409 | |
5c3a0fd7 | 1410 | void ip_vs_scheduler_err(struct ip_vs_service *svc, const char *msg); |
41ac51ee | 1411 | |
07dcc686 | 1412 | /* IPVS control data and functions (from ip_vs_ctl.c) */ |
1da177e4 | 1413 | extern struct ip_vs_stats ip_vs_stats; |
b880c1f0 | 1414 | extern int sysctl_ip_vs_sync_ver; |
1da177e4 | 1415 | |
5c3a0fd7 | 1416 | struct ip_vs_service * |
0a4fd6ce | 1417 | ip_vs_service_find(struct netns_ipvs *ipvs, int af, __u32 fwmark, __u16 protocol, |
3c2e0505 | 1418 | const union nf_inet_addr *vaddr, __be16 vport); |
1da177e4 | 1419 | |
48aed1b0 | 1420 | bool ip_vs_has_real_service(struct netns_ipvs *ipvs, int af, __u16 protocol, |
5c3a0fd7 JP |
1421 | const union nf_inet_addr *daddr, __be16 dport); |
1422 | ||
1423 | int ip_vs_use_count_inc(void); | |
1424 | void ip_vs_use_count_dec(void); | |
1425 | int ip_vs_register_nl_ioctl(void); | |
1426 | void ip_vs_unregister_nl_ioctl(void); | |
1427 | int ip_vs_control_init(void); | |
1428 | void ip_vs_control_cleanup(void); | |
1429 | struct ip_vs_dest * | |
dc2add6f | 1430 | ip_vs_find_dest(struct netns_ipvs *ipvs, int svc_af, int dest_af, |
655eef10 AG |
1431 | const union nf_inet_addr *daddr, __be16 dport, |
1432 | const union nf_inet_addr *vaddr, __be16 vport, | |
52793dbe | 1433 | __u16 protocol, __u32 fwmark, __u32 flags); |
5c3a0fd7 | 1434 | void ip_vs_try_bind_dest(struct ip_vs_conn *cp); |
1da177e4 | 1435 | |
fca9c20a JA |
1436 | static inline void ip_vs_dest_hold(struct ip_vs_dest *dest) |
1437 | { | |
1438 | atomic_inc(&dest->refcnt); | |
1439 | } | |
1440 | ||
1441 | static inline void ip_vs_dest_put(struct ip_vs_dest *dest) | |
1442 | { | |
4e857c58 | 1443 | smp_mb__before_atomic(); |
fca9c20a JA |
1444 | atomic_dec(&dest->refcnt); |
1445 | } | |
1da177e4 | 1446 | |
9e4e948a JA |
1447 | static inline void ip_vs_dest_put_and_free(struct ip_vs_dest *dest) |
1448 | { | |
1449 | if (atomic_dec_return(&dest->refcnt) < 0) | |
1450 | kfree(dest); | |
1451 | } | |
1452 | ||
07dcc686 SH |
1453 | /* IPVS sync daemon data and function prototypes |
1454 | * (from ip_vs_sync.c) | |
1da177e4 | 1455 | */ |
6ac121d7 | 1456 | int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *cfg, |
e4ff6751 | 1457 | int state); |
b3cf3cbf | 1458 | int stop_sync_thread(struct netns_ipvs *ipvs, int state); |
b61a8c1a | 1459 | void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts); |
1da177e4 | 1460 | |
07dcc686 | 1461 | /* IPVS rate estimator prototypes (from ip_vs_est.c) */ |
0f34d54b EB |
1462 | void ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats); |
1463 | void ip_vs_stop_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats); | |
5c3a0fd7 | 1464 | void ip_vs_zero_estimator(struct ip_vs_stats *stats); |
cd67cd5e | 1465 | void ip_vs_read_estimator(struct ip_vs_kstats *dst, struct ip_vs_stats *stats); |
1da177e4 | 1466 | |
07dcc686 | 1467 | /* Various IPVS packet transmitters (from ip_vs_xmit.c) */ |
5c3a0fd7 JP |
1468 | int ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, |
1469 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1470 | int ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1471 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1472 | int ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1473 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1474 | int ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1475 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1476 | int ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1477 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1478 | int ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1479 | struct ip_vs_protocol *pp, int offset, | |
1480 | unsigned int hooknum, struct ip_vs_iphdr *iph); | |
1481 | void ip_vs_dest_dst_rcu_free(struct rcu_head *head); | |
1da177e4 | 1482 | |
b3cdd2a7 | 1483 | #ifdef CONFIG_IP_VS_IPV6 |
5c3a0fd7 JP |
1484 | int ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, |
1485 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1486 | int ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1487 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1488 | int ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1489 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1490 | int ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1491 | struct ip_vs_protocol *pp, struct ip_vs_iphdr *iph); | |
1492 | int ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp, | |
1493 | struct ip_vs_protocol *pp, int offset, | |
1494 | unsigned int hooknum, struct ip_vs_iphdr *iph); | |
b3cdd2a7 | 1495 | #endif |
1da177e4 | 1496 | |
3a1bbf18 | 1497 | #ifdef CONFIG_SYSCTL |
07dcc686 SH |
1498 | /* This is a simple mechanism to ignore packets when |
1499 | * we are loaded. Just set ip_vs_drop_rate to 'n' and | |
1500 | * we start to drop 1/rate of the packets | |
1da177e4 | 1501 | */ |
a0840e2e | 1502 | static inline int ip_vs_todrop(struct netns_ipvs *ipvs) |
1da177e4 | 1503 | { |
a0840e2e HS |
1504 | if (!ipvs->drop_rate) |
1505 | return 0; | |
1506 | if (--ipvs->drop_counter > 0) | |
1507 | return 0; | |
1508 | ipvs->drop_counter = ipvs->drop_rate; | |
1da177e4 LT |
1509 | return 1; |
1510 | } | |
3a1bbf18 SH |
1511 | #else |
1512 | static inline int ip_vs_todrop(struct netns_ipvs *ipvs) { return 0; } | |
1513 | #endif | |
1da177e4 | 1514 | |
07dcc686 | 1515 | /* ip_vs_fwd_tag returns the forwarding tag of the connection */ |
1da177e4 LT |
1516 | #define IP_VS_FWD_METHOD(cp) (cp->flags & IP_VS_CONN_F_FWD_MASK) |
1517 | ||
732db659 | 1518 | static inline char ip_vs_fwd_tag(struct ip_vs_conn *cp) |
1da177e4 LT |
1519 | { |
1520 | char fwd; | |
1521 | ||
1522 | switch (IP_VS_FWD_METHOD(cp)) { | |
1523 | case IP_VS_CONN_F_MASQ: | |
1524 | fwd = 'M'; break; | |
1525 | case IP_VS_CONN_F_LOCALNODE: | |
1526 | fwd = 'L'; break; | |
1527 | case IP_VS_CONN_F_TUNNEL: | |
1528 | fwd = 'T'; break; | |
1529 | case IP_VS_CONN_F_DROUTE: | |
1530 | fwd = 'R'; break; | |
1531 | case IP_VS_CONN_F_BYPASS: | |
1532 | fwd = 'B'; break; | |
1533 | default: | |
1534 | fwd = '?'; break; | |
1535 | } | |
1536 | return fwd; | |
1537 | } | |
1538 | ||
5c3a0fd7 JP |
1539 | void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp, |
1540 | struct ip_vs_conn *cp, int dir); | |
b3cdd2a7 JV |
1541 | |
1542 | #ifdef CONFIG_IP_VS_IPV6 | |
5c3a0fd7 JP |
1543 | void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp, |
1544 | struct ip_vs_conn *cp, int dir); | |
b3cdd2a7 | 1545 | #endif |
1da177e4 | 1546 | |
5c3a0fd7 | 1547 | __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset); |
1da177e4 | 1548 | |
f9214b26 | 1549 | static inline __wsum ip_vs_check_diff4(__be32 old, __be32 new, __wsum oldsum) |
1da177e4 | 1550 | { |
f9214b26 | 1551 | __be32 diff[2] = { ~old, new }; |
1da177e4 | 1552 | |
07f0757a | 1553 | return csum_partial(diff, sizeof(diff), oldsum); |
f9214b26 AV |
1554 | } |
1555 | ||
0bbdd42b JV |
1556 | #ifdef CONFIG_IP_VS_IPV6 |
1557 | static inline __wsum ip_vs_check_diff16(const __be32 *old, const __be32 *new, | |
1558 | __wsum oldsum) | |
1559 | { | |
1560 | __be32 diff[8] = { ~old[3], ~old[2], ~old[1], ~old[0], | |
1561 | new[3], new[2], new[1], new[0] }; | |
1562 | ||
07f0757a | 1563 | return csum_partial(diff, sizeof(diff), oldsum); |
0bbdd42b JV |
1564 | } |
1565 | #endif | |
1566 | ||
f9214b26 AV |
1567 | static inline __wsum ip_vs_check_diff2(__be16 old, __be16 new, __wsum oldsum) |
1568 | { | |
1569 | __be16 diff[2] = { ~old, new }; | |
1570 | ||
07f0757a | 1571 | return csum_partial(diff, sizeof(diff), oldsum); |
1da177e4 LT |
1572 | } |
1573 | ||
07dcc686 | 1574 | /* Forget current conntrack (unconfirmed) and attach notrack entry */ |
cf356d69 JA |
1575 | static inline void ip_vs_notrack(struct sk_buff *skb) |
1576 | { | |
1577 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) | |
1578 | enum ip_conntrack_info ctinfo; | |
06b69390 | 1579 | struct nf_conn *ct = nf_ct_get(skb, &ctinfo); |
cf356d69 JA |
1580 | |
1581 | if (!ct || !nf_ct_is_untracked(ct)) { | |
9e33ce45 | 1582 | nf_conntrack_put(skb->nfct); |
cf356d69 JA |
1583 | skb->nfct = &nf_ct_untracked_get()->ct_general; |
1584 | skb->nfctinfo = IP_CT_NEW; | |
1585 | nf_conntrack_get(skb->nfct); | |
1586 | } | |
1587 | #endif | |
1588 | } | |
1589 | ||
f4bc17cd | 1590 | #ifdef CONFIG_IP_VS_NFCT |
07dcc686 SH |
1591 | /* Netfilter connection tracking |
1592 | * (from ip_vs_nfct.c) | |
f4bc17cd | 1593 | */ |
a0840e2e | 1594 | static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs) |
f4bc17cd | 1595 | { |
a4e2f5a7 | 1596 | #ifdef CONFIG_SYSCTL |
a0840e2e | 1597 | return ipvs->sysctl_conntrack; |
a4e2f5a7 SH |
1598 | #else |
1599 | return 0; | |
1600 | #endif | |
f4bc17cd JA |
1601 | } |
1602 | ||
5c3a0fd7 JP |
1603 | void ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp, |
1604 | int outin); | |
1605 | int ip_vs_confirm_conntrack(struct sk_buff *skb); | |
1606 | void ip_vs_nfct_expect_related(struct sk_buff *skb, struct nf_conn *ct, | |
1607 | struct ip_vs_conn *cp, u_int8_t proto, | |
1608 | const __be16 port, int from_rs); | |
1609 | void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp); | |
f4bc17cd JA |
1610 | |
1611 | #else | |
1612 | ||
a0840e2e | 1613 | static inline int ip_vs_conntrack_enabled(struct netns_ipvs *ipvs) |
f4bc17cd JA |
1614 | { |
1615 | return 0; | |
1616 | } | |
1617 | ||
1618 | static inline void ip_vs_update_conntrack(struct sk_buff *skb, | |
1619 | struct ip_vs_conn *cp, int outin) | |
1620 | { | |
1621 | } | |
1622 | ||
e23ebf0f | 1623 | static inline int ip_vs_confirm_conntrack(struct sk_buff *skb) |
f4bc17cd JA |
1624 | { |
1625 | return NF_ACCEPT; | |
1626 | } | |
1627 | ||
1628 | static inline void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp) | |
1629 | { | |
1630 | } | |
07dcc686 | 1631 | #endif /* CONFIG_IP_VS_NFCT */ |
6523ce15 | 1632 | |
c16526a7 | 1633 | static inline int |
b552f7e3 CG |
1634 | ip_vs_dest_conn_overhead(struct ip_vs_dest *dest) |
1635 | { | |
07dcc686 | 1636 | /* We think the overhead of processing active connections is 256 |
b552f7e3 CG |
1637 | * times higher than that of inactive connections in average. (This |
1638 | * 256 times might not be accurate, we will change it later) We | |
1639 | * use the following formula to estimate the overhead now: | |
1640 | * dest->activeconns*256 + dest->inactconns | |
1641 | */ | |
1642 | return (atomic_read(&dest->activeconns) << 8) + | |
1643 | atomic_read(&dest->inactconns); | |
1644 | } | |
1645 | ||
bc4768eb | 1646 | #endif /* _NET_IP_VS_H */ |