Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * ip_vs_proto.c: transport protocol load balancing support for IPVS |
| 3 | * |
| 4 | * Version: $Id: ip_vs_proto.c,v 1.2 2003/04/18 09:03:16 wensong Exp $ |
| 5 | * |
| 6 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
| 7 | * Julian Anastasov <ja@ssi.bg> |
| 8 | * |
| 9 | * This program is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU General Public License |
| 11 | * as published by the Free Software Foundation; either version |
| 12 | * 2 of the License, or (at your option) any later version. |
| 13 | * |
| 14 | * Changes: |
| 15 | * |
| 16 | */ |
| 17 | |
| 18 | #include <linux/module.h> |
| 19 | #include <linux/kernel.h> |
| 20 | #include <linux/skbuff.h> |
| 21 | #include <linux/in.h> |
| 22 | #include <linux/ip.h> |
| 23 | #include <net/protocol.h> |
| 24 | #include <net/tcp.h> |
| 25 | #include <net/udp.h> |
| 26 | #include <asm/system.h> |
| 27 | #include <linux/stat.h> |
| 28 | #include <linux/proc_fs.h> |
| 29 | |
| 30 | #include <net/ip_vs.h> |
| 31 | |
| 32 | |
| 33 | /* |
| 34 | * IPVS protocols can only be registered/unregistered when the ipvs |
| 35 | * module is loaded/unloaded, so no lock is needed in accessing the |
| 36 | * ipvs protocol table. |
| 37 | */ |
| 38 | |
| 39 | #define IP_VS_PROTO_TAB_SIZE 32 /* must be power of 2 */ |
| 40 | #define IP_VS_PROTO_HASH(proto) ((proto) & (IP_VS_PROTO_TAB_SIZE-1)) |
| 41 | |
| 42 | static struct ip_vs_protocol *ip_vs_proto_table[IP_VS_PROTO_TAB_SIZE]; |
| 43 | |
| 44 | |
| 45 | /* |
| 46 | * register an ipvs protocol |
| 47 | */ |
Pavel Emelyanov | d535a91 | 2007-11-20 17:44:01 -0800 | [diff] [blame] | 48 | static int __used register_ip_vs_protocol(struct ip_vs_protocol *pp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | { |
| 50 | unsigned hash = IP_VS_PROTO_HASH(pp->protocol); |
| 51 | |
| 52 | pp->next = ip_vs_proto_table[hash]; |
| 53 | ip_vs_proto_table[hash] = pp; |
| 54 | |
| 55 | if (pp->init != NULL) |
| 56 | pp->init(pp); |
| 57 | |
| 58 | return 0; |
| 59 | } |
| 60 | |
| 61 | |
| 62 | /* |
| 63 | * unregister an ipvs protocol |
| 64 | */ |
| 65 | static int unregister_ip_vs_protocol(struct ip_vs_protocol *pp) |
| 66 | { |
| 67 | struct ip_vs_protocol **pp_p; |
| 68 | unsigned hash = IP_VS_PROTO_HASH(pp->protocol); |
| 69 | |
| 70 | pp_p = &ip_vs_proto_table[hash]; |
| 71 | for (; *pp_p; pp_p = &(*pp_p)->next) { |
| 72 | if (*pp_p == pp) { |
| 73 | *pp_p = pp->next; |
| 74 | if (pp->exit != NULL) |
| 75 | pp->exit(pp); |
| 76 | return 0; |
| 77 | } |
| 78 | } |
| 79 | |
| 80 | return -ESRCH; |
| 81 | } |
| 82 | |
| 83 | |
| 84 | /* |
| 85 | * get ip_vs_protocol object by its proto. |
| 86 | */ |
| 87 | struct ip_vs_protocol * ip_vs_proto_get(unsigned short proto) |
| 88 | { |
| 89 | struct ip_vs_protocol *pp; |
| 90 | unsigned hash = IP_VS_PROTO_HASH(proto); |
| 91 | |
| 92 | for (pp = ip_vs_proto_table[hash]; pp; pp = pp->next) { |
| 93 | if (pp->protocol == proto) |
| 94 | return pp; |
| 95 | } |
| 96 | |
| 97 | return NULL; |
| 98 | } |
| 99 | |
| 100 | |
| 101 | /* |
| 102 | * Propagate event for state change to all protocols |
| 103 | */ |
| 104 | void ip_vs_protocol_timeout_change(int flags) |
| 105 | { |
| 106 | struct ip_vs_protocol *pp; |
| 107 | int i; |
| 108 | |
| 109 | for (i = 0; i < IP_VS_PROTO_TAB_SIZE; i++) { |
| 110 | for (pp = ip_vs_proto_table[i]; pp; pp = pp->next) { |
| 111 | if (pp->timeout_change) |
| 112 | pp->timeout_change(pp, flags); |
| 113 | } |
| 114 | } |
| 115 | } |
| 116 | |
| 117 | |
| 118 | int * |
| 119 | ip_vs_create_timeout_table(int *table, int size) |
| 120 | { |
Arnaldo Carvalho de Melo | 8b2ed4b | 2006-11-21 01:17:18 -0200 | [diff] [blame] | 121 | return kmemdup(table, size, GFP_ATOMIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | } |
| 123 | |
| 124 | |
| 125 | /* |
| 126 | * Set timeout value for state specified by name |
| 127 | */ |
| 128 | int |
| 129 | ip_vs_set_state_timeout(int *table, int num, char **names, char *name, int to) |
| 130 | { |
| 131 | int i; |
| 132 | |
| 133 | if (!table || !name || !to) |
| 134 | return -EINVAL; |
| 135 | |
| 136 | for (i = 0; i < num; i++) { |
| 137 | if (strcmp(names[i], name)) |
| 138 | continue; |
| 139 | table[i] = to * HZ; |
| 140 | return 0; |
| 141 | } |
| 142 | return -ENOENT; |
| 143 | } |
| 144 | |
| 145 | |
| 146 | const char * ip_vs_state_name(__u16 proto, int state) |
| 147 | { |
| 148 | struct ip_vs_protocol *pp = ip_vs_proto_get(proto); |
| 149 | |
| 150 | if (pp == NULL || pp->state_name == NULL) |
Julian Anastasov | 2ad17de | 2008-04-29 03:21:23 -0700 | [diff] [blame^] | 151 | return (IPPROTO_IP == proto) ? "NONE" : "ERR!"; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | return pp->state_name(state); |
| 153 | } |
| 154 | |
| 155 | |
| 156 | void |
| 157 | ip_vs_tcpudp_debug_packet(struct ip_vs_protocol *pp, |
| 158 | const struct sk_buff *skb, |
| 159 | int offset, |
| 160 | const char *msg) |
| 161 | { |
| 162 | char buf[128]; |
| 163 | struct iphdr _iph, *ih; |
| 164 | |
| 165 | ih = skb_header_pointer(skb, offset, sizeof(_iph), &_iph); |
| 166 | if (ih == NULL) |
| 167 | sprintf(buf, "%s TRUNCATED", pp->name); |
YOSHIFUJI Hideaki | 5661df7 | 2007-12-12 03:53:11 +0900 | [diff] [blame] | 168 | else if (ih->frag_off & htons(IP_OFFSET)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 169 | sprintf(buf, "%s %u.%u.%u.%u->%u.%u.%u.%u frag", |
| 170 | pp->name, NIPQUAD(ih->saddr), |
| 171 | NIPQUAD(ih->daddr)); |
| 172 | else { |
Al Viro | 014d730 | 2006-09-28 14:29:52 -0700 | [diff] [blame] | 173 | __be16 _ports[2], *pptr |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | ; |
| 175 | pptr = skb_header_pointer(skb, offset + ih->ihl*4, |
| 176 | sizeof(_ports), _ports); |
| 177 | if (pptr == NULL) |
| 178 | sprintf(buf, "%s TRUNCATED %u.%u.%u.%u->%u.%u.%u.%u", |
| 179 | pp->name, |
| 180 | NIPQUAD(ih->saddr), |
| 181 | NIPQUAD(ih->daddr)); |
| 182 | else |
| 183 | sprintf(buf, "%s %u.%u.%u.%u:%u->%u.%u.%u.%u:%u", |
| 184 | pp->name, |
| 185 | NIPQUAD(ih->saddr), |
| 186 | ntohs(pptr[0]), |
| 187 | NIPQUAD(ih->daddr), |
| 188 | ntohs(pptr[1])); |
| 189 | } |
| 190 | |
| 191 | printk(KERN_DEBUG "IPVS: %s: %s\n", msg, buf); |
| 192 | } |
| 193 | |
| 194 | |
| 195 | int ip_vs_protocol_init(void) |
| 196 | { |
| 197 | char protocols[64]; |
| 198 | #define REGISTER_PROTOCOL(p) \ |
| 199 | do { \ |
| 200 | register_ip_vs_protocol(p); \ |
| 201 | strcat(protocols, ", "); \ |
| 202 | strcat(protocols, (p)->name); \ |
| 203 | } while (0) |
| 204 | |
| 205 | protocols[0] = '\0'; |
| 206 | protocols[2] = '\0'; |
| 207 | #ifdef CONFIG_IP_VS_PROTO_TCP |
| 208 | REGISTER_PROTOCOL(&ip_vs_protocol_tcp); |
| 209 | #endif |
| 210 | #ifdef CONFIG_IP_VS_PROTO_UDP |
| 211 | REGISTER_PROTOCOL(&ip_vs_protocol_udp); |
| 212 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | #ifdef CONFIG_IP_VS_PROTO_AH |
| 214 | REGISTER_PROTOCOL(&ip_vs_protocol_ah); |
| 215 | #endif |
| 216 | #ifdef CONFIG_IP_VS_PROTO_ESP |
| 217 | REGISTER_PROTOCOL(&ip_vs_protocol_esp); |
| 218 | #endif |
| 219 | IP_VS_INFO("Registered protocols (%s)\n", &protocols[2]); |
| 220 | |
| 221 | return 0; |
| 222 | } |
| 223 | |
| 224 | |
| 225 | void ip_vs_protocol_cleanup(void) |
| 226 | { |
| 227 | struct ip_vs_protocol *pp; |
| 228 | int i; |
| 229 | |
| 230 | /* unregister all the ipvs protocols */ |
| 231 | for (i = 0; i < IP_VS_PROTO_TAB_SIZE; i++) { |
| 232 | while ((pp = ip_vs_proto_table[i]) != NULL) |
| 233 | unregister_ip_vs_protocol(pp); |
| 234 | } |
| 235 | } |