YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 1 | /* Cluster IP hashmark target |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | * (C) 2003-2004 by Harald Welte <laforge@netfilter.org> |
| 3 | * based on ideas of Fabio Olive Leite <olive@unixforge.org> |
| 4 | * |
| 5 | * Development of this code funded by SuSE Linux AG, http://www.suse.com/ |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify |
| 8 | * it under the terms of the GNU General Public License version 2 as |
| 9 | * published by the Free Software Foundation. |
| 10 | * |
| 11 | */ |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | #include <linux/module.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/proc_fs.h> |
| 15 | #include <linux/jhash.h> |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 16 | #include <linux/bitops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/skbuff.h> |
| 18 | #include <linux/ip.h> |
| 19 | #include <linux/tcp.h> |
| 20 | #include <linux/udp.h> |
| 21 | #include <linux/icmp.h> |
| 22 | #include <linux/if_arp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <linux/seq_file.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/netfilter_arp.h> |
Jan Engelhardt | 6709dbb | 2007-02-07 15:11:19 -0800 | [diff] [blame] | 25 | #include <linux/netfilter/x_tables.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include <linux/netfilter_ipv4/ip_tables.h> |
| 27 | #include <linux/netfilter_ipv4/ipt_CLUSTERIP.h> |
Patrick McHardy | 587aa64 | 2007-03-14 16:37:25 -0700 | [diff] [blame] | 28 | #include <net/netfilter/nf_conntrack.h> |
Eric W. Biederman | 457c4cb | 2007-09-12 12:01:34 +0200 | [diff] [blame] | 29 | #include <net/net_namespace.h> |
Patrick McHardy | 587aa64 | 2007-03-14 16:37:25 -0700 | [diff] [blame] | 30 | #include <net/checksum.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 32 | #define CLUSTERIP_VERSION "0.8" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | MODULE_LICENSE("GPL"); |
| 35 | MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>"); |
Jan Engelhardt | 2ae15b6 | 2008-01-14 23:42:28 -0800 | [diff] [blame] | 36 | MODULE_DESCRIPTION("Xtables: CLUSTERIP target"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | |
| 38 | struct clusterip_config { |
| 39 | struct list_head list; /* list of all configs */ |
| 40 | atomic_t refcount; /* reference count */ |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 41 | atomic_t entries; /* number of entries/rules |
| 42 | * referencing us */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | |
Al Viro | 6a19d61 | 2006-09-28 14:22:24 -0700 | [diff] [blame] | 44 | __be32 clusterip; /* the IP address */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | u_int8_t clustermac[ETH_ALEN]; /* the MAC address */ |
| 46 | struct net_device *dev; /* device */ |
| 47 | u_int16_t num_total_nodes; /* total number of nodes */ |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 48 | unsigned long local_nodes; /* node number array */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | |
| 50 | #ifdef CONFIG_PROC_FS |
| 51 | struct proc_dir_entry *pde; /* proc dir entry */ |
| 52 | #endif |
| 53 | enum clusterip_hashmode hash_mode; /* which hashing mode */ |
| 54 | u_int32_t hash_initval; /* hash initialization */ |
| 55 | }; |
| 56 | |
| 57 | static LIST_HEAD(clusterip_configs); |
| 58 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 59 | /* clusterip_lock protects the clusterip_configs list */ |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 60 | static DEFINE_RWLOCK(clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
| 62 | #ifdef CONFIG_PROC_FS |
Arjan van de Ven | 9a32144 | 2007-02-12 00:55:35 -0800 | [diff] [blame] | 63 | static const struct file_operations clusterip_proc_fops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | static struct proc_dir_entry *clusterip_procdir; |
| 65 | #endif |
| 66 | |
| 67 | static inline void |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 68 | clusterip_config_get(struct clusterip_config *c) |
| 69 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | atomic_inc(&c->refcount); |
| 71 | } |
| 72 | |
| 73 | static inline void |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 74 | clusterip_config_put(struct clusterip_config *c) |
| 75 | { |
| 76 | if (atomic_dec_and_test(&c->refcount)) |
| 77 | kfree(c); |
| 78 | } |
| 79 | |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 80 | /* decrease the count of entries using/referencing this config. If last |
| 81 | * entry(rule) is removed, remove the config from lists, but don't free it |
| 82 | * yet, since proc-files could still be holding references */ |
| 83 | static inline void |
| 84 | clusterip_config_entry_put(struct clusterip_config *c) |
| 85 | { |
Pavel Emelyanov | 4dee959 | 2008-04-14 00:44:52 -0700 | [diff] [blame] | 86 | write_lock_bh(&clusterip_lock); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 87 | if (atomic_dec_and_test(&c->entries)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | list_del(&c->list); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 89 | write_unlock_bh(&clusterip_lock); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 90 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | dev_mc_delete(c->dev, c->clustermac, ETH_ALEN, 0); |
| 92 | dev_put(c->dev); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 93 | |
| 94 | /* In case anyone still accesses the file, the open/close |
| 95 | * functions are also incrementing the refcount on their own, |
| 96 | * so it's safe to remove the entry even if it's in use. */ |
| 97 | #ifdef CONFIG_PROC_FS |
| 98 | remove_proc_entry(c->pde->name, c->pde->parent); |
| 99 | #endif |
Pavel Emelyanov | 4dee959 | 2008-04-14 00:44:52 -0700 | [diff] [blame] | 100 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | } |
Pavel Emelyanov | 4dee959 | 2008-04-14 00:44:52 -0700 | [diff] [blame] | 102 | write_unlock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | } |
| 104 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | static struct clusterip_config * |
Al Viro | 6a19d61 | 2006-09-28 14:22:24 -0700 | [diff] [blame] | 106 | __clusterip_config_find(__be32 clusterip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | { |
Li Zefan | 4c61097 | 2007-12-04 23:22:26 -0800 | [diff] [blame] | 108 | struct clusterip_config *c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | |
Li Zefan | 4c61097 | 2007-12-04 23:22:26 -0800 | [diff] [blame] | 110 | list_for_each_entry(c, &clusterip_configs, list) { |
Jan Engelhardt | 7c4e36b | 2007-07-07 22:19:08 -0700 | [diff] [blame] | 111 | if (c->clusterip == clusterip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | return c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | } |
| 114 | |
| 115 | return NULL; |
| 116 | } |
| 117 | |
| 118 | static inline struct clusterip_config * |
Al Viro | 6a19d61 | 2006-09-28 14:22:24 -0700 | [diff] [blame] | 119 | clusterip_config_find_get(__be32 clusterip, int entry) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | { |
| 121 | struct clusterip_config *c; |
| 122 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 123 | read_lock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | c = __clusterip_config_find(clusterip); |
| 125 | if (!c) { |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 126 | read_unlock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | return NULL; |
| 128 | } |
| 129 | atomic_inc(&c->refcount); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 130 | if (entry) |
| 131 | atomic_inc(&c->entries); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 132 | read_unlock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | |
| 134 | return c; |
| 135 | } |
| 136 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 137 | static void |
| 138 | clusterip_config_init_nodelist(struct clusterip_config *c, |
| 139 | const struct ipt_clusterip_tgt_info *i) |
| 140 | { |
| 141 | int n; |
| 142 | |
Jan Engelhardt | 7c4e36b | 2007-07-07 22:19:08 -0700 | [diff] [blame] | 143 | for (n = 0; n < i->num_local_nodes; n++) |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 144 | set_bit(i->local_nodes[n] - 1, &c->local_nodes); |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 145 | } |
| 146 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | static struct clusterip_config * |
Jan Engelhardt | 3cf93c9 | 2008-04-14 09:56:05 +0200 | [diff] [blame] | 148 | clusterip_config_init(const struct ipt_clusterip_tgt_info *i, __be32 ip, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | struct net_device *dev) |
| 150 | { |
| 151 | struct clusterip_config *c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | |
Panagiotis Issaris | 0da974f | 2006-07-21 14:51:30 -0700 | [diff] [blame] | 153 | c = kzalloc(sizeof(*c), GFP_ATOMIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | if (!c) |
| 155 | return NULL; |
| 156 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | c->dev = dev; |
| 158 | c->clusterip = ip; |
| 159 | memcpy(&c->clustermac, &i->clustermac, ETH_ALEN); |
| 160 | c->num_total_nodes = i->num_total_nodes; |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 161 | clusterip_config_init_nodelist(c, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | c->hash_mode = i->hash_mode; |
| 163 | c->hash_initval = i->hash_initval; |
| 164 | atomic_set(&c->refcount, 1); |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 165 | atomic_set(&c->entries, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | |
| 167 | #ifdef CONFIG_PROC_FS |
Patrick McHardy | 7659258 | 2006-11-29 02:35:42 +0100 | [diff] [blame] | 168 | { |
| 169 | char buffer[16]; |
| 170 | |
| 171 | /* create proc dir entry */ |
Harvey Harrison | cffee38 | 2008-10-31 00:53:08 -0700 | [diff] [blame] | 172 | sprintf(buffer, "%pI4", &ip); |
Denis V. Lunev | 6e79d85 | 2008-05-02 02:45:42 -0700 | [diff] [blame] | 173 | c->pde = proc_create_data(buffer, S_IWUSR|S_IRUSR, |
| 174 | clusterip_procdir, |
| 175 | &clusterip_proc_fops, c); |
Patrick McHardy | 7659258 | 2006-11-29 02:35:42 +0100 | [diff] [blame] | 176 | if (!c->pde) { |
| 177 | kfree(c); |
| 178 | return NULL; |
| 179 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | #endif |
| 182 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 183 | write_lock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | list_add(&c->list, &clusterip_configs); |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 185 | write_unlock_bh(&clusterip_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | |
| 187 | return c; |
| 188 | } |
| 189 | |
Patrick McHardy | 7659258 | 2006-11-29 02:35:42 +0100 | [diff] [blame] | 190 | #ifdef CONFIG_PROC_FS |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | static int |
| 192 | clusterip_add_node(struct clusterip_config *c, u_int16_t nodenum) |
| 193 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 195 | if (nodenum == 0 || |
| 196 | nodenum > c->num_total_nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 199 | /* check if we already have this number in our bitfield */ |
| 200 | if (test_and_set_bit(nodenum - 1, &c->local_nodes)) |
| 201 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | return 0; |
| 204 | } |
| 205 | |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 206 | static bool |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | clusterip_del_node(struct clusterip_config *c, u_int16_t nodenum) |
| 208 | { |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 209 | if (nodenum == 0 || |
| 210 | nodenum > c->num_total_nodes) |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 211 | return true; |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 212 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 213 | if (test_and_clear_bit(nodenum - 1, &c->local_nodes)) |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 214 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 216 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | } |
Patrick McHardy | 7659258 | 2006-11-29 02:35:42 +0100 | [diff] [blame] | 218 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | |
| 220 | static inline u_int32_t |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 221 | clusterip_hashfn(const struct sk_buff *skb, |
| 222 | const struct clusterip_config *config) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | { |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 224 | const struct iphdr *iph = ip_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | unsigned long hashval; |
| 226 | u_int16_t sport, dport; |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 227 | const u_int16_t *ports; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | |
| 229 | switch (iph->protocol) { |
| 230 | case IPPROTO_TCP: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | case IPPROTO_UDP: |
Patrick McHardy | a8d0f95 | 2007-02-07 15:07:43 -0800 | [diff] [blame] | 232 | case IPPROTO_UDPLITE: |
Patrick McHardy | 957dc80 | 2006-05-29 18:19:56 -0700 | [diff] [blame] | 233 | case IPPROTO_SCTP: |
| 234 | case IPPROTO_DCCP: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | case IPPROTO_ICMP: |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 236 | ports = (const void *)iph+iph->ihl*4; |
Patrick McHardy | 957dc80 | 2006-05-29 18:19:56 -0700 | [diff] [blame] | 237 | sport = ports[0]; |
| 238 | dport = ports[1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | break; |
| 240 | default: |
Jan Engelhardt | 7c4e36b | 2007-07-07 22:19:08 -0700 | [diff] [blame] | 241 | if (net_ratelimit()) |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 242 | pr_info("unknown protocol %u\n", iph->protocol); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 243 | sport = dport = 0; |
| 244 | } |
| 245 | |
| 246 | switch (config->hash_mode) { |
| 247 | case CLUSTERIP_HASHMODE_SIP: |
| 248 | hashval = jhash_1word(ntohl(iph->saddr), |
| 249 | config->hash_initval); |
| 250 | break; |
| 251 | case CLUSTERIP_HASHMODE_SIP_SPT: |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 252 | hashval = jhash_2words(ntohl(iph->saddr), sport, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 253 | config->hash_initval); |
| 254 | break; |
| 255 | case CLUSTERIP_HASHMODE_SIP_SPT_DPT: |
| 256 | hashval = jhash_3words(ntohl(iph->saddr), sport, dport, |
| 257 | config->hash_initval); |
| 258 | break; |
| 259 | default: |
| 260 | /* to make gcc happy */ |
| 261 | hashval = 0; |
| 262 | /* This cannot happen, unless the check function wasn't called |
| 263 | * at rule load time */ |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 264 | pr_info("unknown mode %u\n", config->hash_mode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | BUG(); |
| 266 | break; |
| 267 | } |
| 268 | |
| 269 | /* node numbers are 1..n, not 0..n */ |
Patrick McHardy | 3449882 | 2007-12-17 22:45:52 -0800 | [diff] [blame] | 270 | return (((u64)hashval * config->num_total_nodes) >> 32) + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | } |
| 272 | |
| 273 | static inline int |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 274 | clusterip_responsible(const struct clusterip_config *config, u_int32_t hash) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | { |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 276 | return test_bit(hash - 1, &config->local_nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | } |
| 278 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 279 | /*********************************************************************** |
| 280 | * IPTABLES TARGET |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | ***********************************************************************/ |
| 282 | |
| 283 | static unsigned int |
Jan Engelhardt | 7eb3558 | 2008-10-08 11:35:19 +0200 | [diff] [blame] | 284 | clusterip_tg(struct sk_buff *skb, const struct xt_target_param *par) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | { |
Jan Engelhardt | 7eb3558 | 2008-10-08 11:35:19 +0200 | [diff] [blame] | 286 | const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
Patrick McHardy | 587aa64 | 2007-03-14 16:37:25 -0700 | [diff] [blame] | 287 | struct nf_conn *ct; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | enum ip_conntrack_info ctinfo; |
Patrick McHardy | 587aa64 | 2007-03-14 16:37:25 -0700 | [diff] [blame] | 289 | u_int32_t hash; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | |
| 291 | /* don't need to clusterip_config_get() here, since refcount |
| 292 | * is only decremented by destroy() - and ip_tables guarantees |
| 293 | * that the ->target() function isn't called after ->destroy() */ |
| 294 | |
Herbert Xu | 3db05fe | 2007-10-15 00:53:15 -0700 | [diff] [blame] | 295 | ct = nf_ct_get(skb, &ctinfo); |
Patrick McHardy | 587aa64 | 2007-03-14 16:37:25 -0700 | [diff] [blame] | 296 | if (ct == NULL) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 297 | pr_info("no conntrack!\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | /* FIXME: need to drop invalid ones, since replies |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 299 | * to outgoing connections of other nodes will be |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 300 | * marked as INVALID */ |
| 301 | return NF_DROP; |
| 302 | } |
| 303 | |
| 304 | /* special case: ICMP error handling. conntrack distinguishes between |
| 305 | * error messages (RELATED) and information requests (see below) */ |
Joe Perches | 3666ed1 | 2009-11-23 23:17:06 +0100 | [diff] [blame] | 306 | if (ip_hdr(skb)->protocol == IPPROTO_ICMP && |
| 307 | (ctinfo == IP_CT_RELATED || |
| 308 | ctinfo == IP_CT_RELATED + IP_CT_IS_REPLY)) |
Jan Engelhardt | 6709dbb | 2007-02-07 15:11:19 -0800 | [diff] [blame] | 309 | return XT_CONTINUE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 311 | /* ip_conntrack_icmp guarantees us that we only have ICMP_ECHO, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | * TIMESTAMP, INFO_REQUEST or ADDRESS type icmp packets from here |
| 313 | * on, which all have an ID field [relevant for hashing]. */ |
| 314 | |
Herbert Xu | 3db05fe | 2007-10-15 00:53:15 -0700 | [diff] [blame] | 315 | hash = clusterip_hashfn(skb, cipinfo->config); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | |
| 317 | switch (ctinfo) { |
| 318 | case IP_CT_NEW: |
Patrick McHardy | 587aa64 | 2007-03-14 16:37:25 -0700 | [diff] [blame] | 319 | ct->mark = hash; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | break; |
| 321 | case IP_CT_RELATED: |
| 322 | case IP_CT_RELATED+IP_CT_IS_REPLY: |
| 323 | /* FIXME: we don't handle expectations at the |
| 324 | * moment. they can arrive on a different node than |
| 325 | * the master connection (e.g. FTP passive mode) */ |
| 326 | case IP_CT_ESTABLISHED: |
| 327 | case IP_CT_ESTABLISHED+IP_CT_IS_REPLY: |
| 328 | break; |
| 329 | default: |
| 330 | break; |
| 331 | } |
| 332 | |
Patrick McHardy | 0d53778 | 2007-07-07 22:39:38 -0700 | [diff] [blame] | 333 | #ifdef DEBUG |
Jan Engelhardt | 3c9fba6 | 2008-04-14 11:15:54 +0200 | [diff] [blame] | 334 | nf_ct_dump_tuple_ip(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | #endif |
Patrick McHardy | 0d53778 | 2007-07-07 22:39:38 -0700 | [diff] [blame] | 336 | pr_debug("hash=%u ct_hash=%u ", hash, ct->mark); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | if (!clusterip_responsible(cipinfo->config, hash)) { |
Patrick McHardy | 0d53778 | 2007-07-07 22:39:38 -0700 | [diff] [blame] | 338 | pr_debug("not responsible\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | return NF_DROP; |
| 340 | } |
Patrick McHardy | 0d53778 | 2007-07-07 22:39:38 -0700 | [diff] [blame] | 341 | pr_debug("responsible\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 342 | |
| 343 | /* despite being received via linklayer multicast, this is |
| 344 | * actually a unicast IP packet. TCP doesn't like PACKET_MULTICAST */ |
Herbert Xu | 3db05fe | 2007-10-15 00:53:15 -0700 | [diff] [blame] | 345 | skb->pkt_type = PACKET_HOST; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | |
Jan Engelhardt | 6709dbb | 2007-02-07 15:11:19 -0800 | [diff] [blame] | 347 | return XT_CONTINUE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | } |
| 349 | |
Jan Engelhardt | 135367b | 2010-03-19 17:16:42 +0100 | [diff] [blame^] | 350 | static int clusterip_tg_check(const struct xt_tgchk_param *par) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | { |
Jan Engelhardt | af5d6dc | 2008-10-08 11:35:19 +0200 | [diff] [blame] | 352 | struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
| 353 | const struct ipt_entry *e = par->entryinfo; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | |
| 355 | struct clusterip_config *config; |
| 356 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | if (cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP && |
| 358 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT && |
| 359 | cipinfo->hash_mode != CLUSTERIP_HASHMODE_SIP_SPT_DPT) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 360 | pr_info("unknown mode %u\n", cipinfo->hash_mode); |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 361 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | |
| 363 | } |
Joe Perches | 3666ed1 | 2009-11-23 23:17:06 +0100 | [diff] [blame] | 364 | if (e->ip.dmsk.s_addr != htonl(0xffffffff) || |
| 365 | e->ip.dst.s_addr == 0) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 366 | pr_info("Please specify destination IP\n"); |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 367 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | } |
| 369 | |
| 370 | /* FIXME: further sanity checks */ |
| 371 | |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 372 | config = clusterip_config_find_get(e->ip.dst.s_addr, 1); |
Patrick McHardy | d3c3f42 | 2007-07-07 22:38:30 -0700 | [diff] [blame] | 373 | if (!config) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | if (!(cipinfo->flags & CLUSTERIP_FLAG_NEW)) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 375 | pr_info("no config found for %pI4, need 'new'\n", |
| 376 | &e->ip.dst.s_addr); |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 377 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | } else { |
| 379 | struct net_device *dev; |
| 380 | |
| 381 | if (e->ip.iniface[0] == '\0') { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 382 | pr_info("Please specify an interface name\n"); |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 383 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | } |
| 385 | |
Eric W. Biederman | 881d966 | 2007-09-17 11:56:21 -0700 | [diff] [blame] | 386 | dev = dev_get_by_name(&init_net, e->ip.iniface); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 387 | if (!dev) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 388 | pr_info("no such interface %s\n", |
| 389 | e->ip.iniface); |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 390 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | } |
| 392 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 393 | config = clusterip_config_init(cipinfo, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | e->ip.dst.s_addr, dev); |
| 395 | if (!config) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 396 | pr_info("cannot allocate config\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | dev_put(dev); |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 398 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | } |
| 400 | dev_mc_add(config->dev,config->clustermac, ETH_ALEN, 0); |
| 401 | } |
| 402 | } |
Patrick McHardy | d3c3f42 | 2007-07-07 22:38:30 -0700 | [diff] [blame] | 403 | cipinfo->config = config; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | |
Jan Engelhardt | 0d34545 | 2010-03-19 18:47:51 +0100 | [diff] [blame] | 405 | if (nf_ct_l3proto_try_module_get(par->family) < 0) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 406 | pr_info("cannot load conntrack support for proto=%u\n", |
| 407 | par->family); |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 408 | return false; |
Yasuyuki Kozakai | 11078c3 | 2006-12-12 00:29:02 -0800 | [diff] [blame] | 409 | } |
| 410 | |
Jan Engelhardt | e1931b7 | 2007-07-07 22:16:26 -0700 | [diff] [blame] | 411 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | } |
| 413 | |
| 414 | /* drop reference count of cluster config when rule is deleted */ |
Jan Engelhardt | a2df164 | 2008-10-08 11:35:19 +0200 | [diff] [blame] | 415 | static void clusterip_tg_destroy(const struct xt_tgdtor_param *par) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | { |
Jan Engelhardt | a2df164 | 2008-10-08 11:35:19 +0200 | [diff] [blame] | 417 | const struct ipt_clusterip_tgt_info *cipinfo = par->targinfo; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 419 | /* if no more entries are referencing the config, remove it |
| 420 | * from the list and destroy the proc entry */ |
| 421 | clusterip_config_entry_put(cipinfo->config); |
| 422 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | clusterip_config_put(cipinfo->config); |
Yasuyuki Kozakai | 11078c3 | 2006-12-12 00:29:02 -0800 | [diff] [blame] | 424 | |
Jan Engelhardt | 0d34545 | 2010-03-19 18:47:51 +0100 | [diff] [blame] | 425 | nf_ct_l3proto_module_put(par->family); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | } |
| 427 | |
Patrick McHardy | d3c3f42 | 2007-07-07 22:38:30 -0700 | [diff] [blame] | 428 | #ifdef CONFIG_COMPAT |
| 429 | struct compat_ipt_clusterip_tgt_info |
| 430 | { |
| 431 | u_int32_t flags; |
| 432 | u_int8_t clustermac[6]; |
| 433 | u_int16_t num_total_nodes; |
| 434 | u_int16_t num_local_nodes; |
| 435 | u_int16_t local_nodes[CLUSTERIP_MAX_NODES]; |
| 436 | u_int32_t hash_mode; |
| 437 | u_int32_t hash_initval; |
| 438 | compat_uptr_t config; |
| 439 | }; |
| 440 | #endif /* CONFIG_COMPAT */ |
| 441 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 442 | static struct xt_target clusterip_tg_reg __read_mostly = { |
Patrick McHardy | 1d5cd90 | 2006-03-20 18:01:14 -0800 | [diff] [blame] | 443 | .name = "CLUSTERIP", |
Jan Engelhardt | ee999d8 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 444 | .family = NFPROTO_IPV4, |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 445 | .target = clusterip_tg, |
| 446 | .checkentry = clusterip_tg_check, |
| 447 | .destroy = clusterip_tg_destroy, |
Patrick McHardy | d3c3f42 | 2007-07-07 22:38:30 -0700 | [diff] [blame] | 448 | .targetsize = sizeof(struct ipt_clusterip_tgt_info), |
| 449 | #ifdef CONFIG_COMPAT |
| 450 | .compatsize = sizeof(struct compat_ipt_clusterip_tgt_info), |
| 451 | #endif /* CONFIG_COMPAT */ |
Patrick McHardy | 1d5cd90 | 2006-03-20 18:01:14 -0800 | [diff] [blame] | 452 | .me = THIS_MODULE |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | }; |
| 454 | |
| 455 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 456 | /*********************************************************************** |
| 457 | * ARP MANGLING CODE |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | ***********************************************************************/ |
| 459 | |
| 460 | /* hardcoded for 48bit ethernet and 32bit ipv4 addresses */ |
| 461 | struct arp_payload { |
| 462 | u_int8_t src_hw[ETH_ALEN]; |
Al Viro | 6a19d61 | 2006-09-28 14:22:24 -0700 | [diff] [blame] | 463 | __be32 src_ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | u_int8_t dst_hw[ETH_ALEN]; |
Al Viro | 6a19d61 | 2006-09-28 14:22:24 -0700 | [diff] [blame] | 465 | __be32 dst_ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | } __attribute__ ((packed)); |
| 467 | |
Patrick McHardy | 0d53778 | 2007-07-07 22:39:38 -0700 | [diff] [blame] | 468 | #ifdef DEBUG |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 469 | static void arp_print(struct arp_payload *payload) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | { |
| 471 | #define HBUFFERLEN 30 |
| 472 | char hbuffer[HBUFFERLEN]; |
| 473 | int j,k; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | |
| 475 | for (k=0, j=0; k < HBUFFERLEN-3 && j < ETH_ALEN; j++) { |
Harvey Harrison | 6a8341b | 2008-07-30 16:30:15 -0700 | [diff] [blame] | 476 | hbuffer[k++] = hex_asc_hi(payload->src_hw[j]); |
| 477 | hbuffer[k++] = hex_asc_lo(payload->src_hw[j]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | hbuffer[k++]=':'; |
| 479 | } |
| 480 | hbuffer[--k]='\0'; |
| 481 | |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 482 | pr_debug("src %pI4@%s, dst %pI4\n", |
| 483 | &payload->src_ip, hbuffer, &payload->dst_ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | } |
| 485 | #endif |
| 486 | |
| 487 | static unsigned int |
| 488 | arp_mangle(unsigned int hook, |
Herbert Xu | 3db05fe | 2007-10-15 00:53:15 -0700 | [diff] [blame] | 489 | struct sk_buff *skb, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | const struct net_device *in, |
| 491 | const struct net_device *out, |
| 492 | int (*okfn)(struct sk_buff *)) |
| 493 | { |
Herbert Xu | 3db05fe | 2007-10-15 00:53:15 -0700 | [diff] [blame] | 494 | struct arphdr *arp = arp_hdr(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | struct arp_payload *payload; |
| 496 | struct clusterip_config *c; |
| 497 | |
| 498 | /* we don't care about non-ethernet and non-ipv4 ARP */ |
Joe Perches | 3666ed1 | 2009-11-23 23:17:06 +0100 | [diff] [blame] | 499 | if (arp->ar_hrd != htons(ARPHRD_ETHER) || |
| 500 | arp->ar_pro != htons(ETH_P_IP) || |
| 501 | arp->ar_pln != 4 || arp->ar_hln != ETH_ALEN) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 502 | return NF_ACCEPT; |
| 503 | |
Harald Welte | 4095ebf | 2005-06-28 12:49:30 -0700 | [diff] [blame] | 504 | /* we only want to mangle arp requests and replies */ |
Joe Perches | 3666ed1 | 2009-11-23 23:17:06 +0100 | [diff] [blame] | 505 | if (arp->ar_op != htons(ARPOP_REPLY) && |
| 506 | arp->ar_op != htons(ARPOP_REQUEST)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | return NF_ACCEPT; |
| 508 | |
| 509 | payload = (void *)(arp+1); |
| 510 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 511 | /* if there is no clusterip configuration for the arp reply's |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | * source ip, we don't want to mangle it */ |
KOVACS Krisztian | 4451362 | 2005-09-16 16:59:46 -0700 | [diff] [blame] | 513 | c = clusterip_config_find_get(payload->src_ip, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | if (!c) |
| 515 | return NF_ACCEPT; |
| 516 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 517 | /* normally the linux kernel always replies to arp queries of |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | * addresses on different interfacs. However, in the CLUSTERIP case |
| 519 | * this wouldn't work, since we didn't subscribe the mcast group on |
| 520 | * other interfaces */ |
| 521 | if (c->dev != out) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 522 | pr_debug("not mangling arp reply on different " |
Patrick McHardy | 0d53778 | 2007-07-07 22:39:38 -0700 | [diff] [blame] | 523 | "interface: cip'%s'-skb'%s'\n", |
| 524 | c->dev->name, out->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 525 | clusterip_config_put(c); |
| 526 | return NF_ACCEPT; |
| 527 | } |
| 528 | |
| 529 | /* mangle reply hardware address */ |
| 530 | memcpy(payload->src_hw, c->clustermac, arp->ar_hln); |
| 531 | |
Patrick McHardy | 0d53778 | 2007-07-07 22:39:38 -0700 | [diff] [blame] | 532 | #ifdef DEBUG |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 533 | pr_debug("mangled arp reply: "); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | arp_print(payload); |
| 535 | #endif |
| 536 | |
| 537 | clusterip_config_put(c); |
| 538 | |
| 539 | return NF_ACCEPT; |
| 540 | } |
| 541 | |
Patrick McHardy | 1999414 | 2007-12-05 01:23:00 -0800 | [diff] [blame] | 542 | static struct nf_hook_ops cip_arp_ops __read_mostly = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | .hook = arp_mangle, |
Jan Engelhardt | ee999d8 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 544 | .pf = NFPROTO_ARP, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | .hooknum = NF_ARP_OUT, |
| 546 | .priority = -1 |
| 547 | }; |
| 548 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 549 | /*********************************************************************** |
| 550 | * PROC DIR HANDLING |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 551 | ***********************************************************************/ |
| 552 | |
| 553 | #ifdef CONFIG_PROC_FS |
| 554 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 555 | struct clusterip_seq_position { |
| 556 | unsigned int pos; /* position */ |
| 557 | unsigned int weight; /* number of bits set == size */ |
| 558 | unsigned int bit; /* current bit */ |
| 559 | unsigned long val; /* current value */ |
| 560 | }; |
| 561 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | static void *clusterip_seq_start(struct seq_file *s, loff_t *pos) |
| 563 | { |
Alexey Dobriyan | 4777814 | 2010-01-22 22:21:18 +0100 | [diff] [blame] | 564 | struct clusterip_config *c = s->private; |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 565 | unsigned int weight; |
| 566 | u_int32_t local_nodes; |
| 567 | struct clusterip_seq_position *idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 569 | /* FIXME: possible race */ |
| 570 | local_nodes = c->local_nodes; |
| 571 | weight = hweight32(local_nodes); |
| 572 | if (*pos >= weight) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | return NULL; |
| 574 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 575 | idx = kmalloc(sizeof(struct clusterip_seq_position), GFP_KERNEL); |
| 576 | if (!idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | return ERR_PTR(-ENOMEM); |
| 578 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 579 | idx->pos = *pos; |
| 580 | idx->weight = weight; |
| 581 | idx->bit = ffs(local_nodes); |
| 582 | idx->val = local_nodes; |
| 583 | clear_bit(idx->bit - 1, &idx->val); |
| 584 | |
| 585 | return idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 586 | } |
| 587 | |
| 588 | static void *clusterip_seq_next(struct seq_file *s, void *v, loff_t *pos) |
| 589 | { |
Jan Engelhardt | 3cf93c9 | 2008-04-14 09:56:05 +0200 | [diff] [blame] | 590 | struct clusterip_seq_position *idx = v; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 592 | *pos = ++idx->pos; |
| 593 | if (*pos >= idx->weight) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 594 | kfree(v); |
| 595 | return NULL; |
| 596 | } |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 597 | idx->bit = ffs(idx->val); |
| 598 | clear_bit(idx->bit - 1, &idx->val); |
| 599 | return idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | } |
| 601 | |
| 602 | static void clusterip_seq_stop(struct seq_file *s, void *v) |
| 603 | { |
| 604 | kfree(v); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 605 | } |
| 606 | |
| 607 | static int clusterip_seq_show(struct seq_file *s, void *v) |
| 608 | { |
Jan Engelhardt | 3cf93c9 | 2008-04-14 09:56:05 +0200 | [diff] [blame] | 609 | struct clusterip_seq_position *idx = v; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 611 | if (idx->pos != 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 612 | seq_putc(s, ','); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | |
KOVACS Krisztian | 136e92b | 2005-09-16 17:00:04 -0700 | [diff] [blame] | 614 | seq_printf(s, "%u", idx->bit); |
| 615 | |
| 616 | if (idx->pos == idx->weight - 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | seq_putc(s, '\n'); |
| 618 | |
| 619 | return 0; |
| 620 | } |
| 621 | |
Philippe De Muyter | 56b3d97 | 2007-07-10 23:07:31 -0700 | [diff] [blame] | 622 | static const struct seq_operations clusterip_seq_ops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | .start = clusterip_seq_start, |
| 624 | .next = clusterip_seq_next, |
| 625 | .stop = clusterip_seq_stop, |
| 626 | .show = clusterip_seq_show, |
| 627 | }; |
| 628 | |
| 629 | static int clusterip_proc_open(struct inode *inode, struct file *file) |
| 630 | { |
| 631 | int ret = seq_open(file, &clusterip_seq_ops); |
| 632 | |
| 633 | if (!ret) { |
| 634 | struct seq_file *sf = file->private_data; |
Alexey Dobriyan | 4777814 | 2010-01-22 22:21:18 +0100 | [diff] [blame] | 635 | struct clusterip_config *c = PDE(inode)->data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | |
Alexey Dobriyan | 4777814 | 2010-01-22 22:21:18 +0100 | [diff] [blame] | 637 | sf->private = c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | |
| 639 | clusterip_config_get(c); |
| 640 | } |
| 641 | |
| 642 | return ret; |
| 643 | } |
| 644 | |
| 645 | static int clusterip_proc_release(struct inode *inode, struct file *file) |
| 646 | { |
Alexey Dobriyan | 4777814 | 2010-01-22 22:21:18 +0100 | [diff] [blame] | 647 | struct clusterip_config *c = PDE(inode)->data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | int ret; |
| 649 | |
| 650 | ret = seq_release(inode, file); |
| 651 | |
| 652 | if (!ret) |
| 653 | clusterip_config_put(c); |
| 654 | |
| 655 | return ret; |
| 656 | } |
| 657 | |
| 658 | static ssize_t clusterip_proc_write(struct file *file, const char __user *input, |
| 659 | size_t size, loff_t *ofs) |
| 660 | { |
Alexey Dobriyan | 4777814 | 2010-01-22 22:21:18 +0100 | [diff] [blame] | 661 | struct clusterip_config *c = PDE(file->f_path.dentry->d_inode)->data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 662 | #define PROC_WRITELEN 10 |
| 663 | char buffer[PROC_WRITELEN+1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | unsigned long nodenum; |
| 665 | |
| 666 | if (copy_from_user(buffer, input, PROC_WRITELEN)) |
| 667 | return -EFAULT; |
| 668 | |
| 669 | if (*buffer == '+') { |
| 670 | nodenum = simple_strtoul(buffer+1, NULL, 10); |
| 671 | if (clusterip_add_node(c, nodenum)) |
| 672 | return -ENOMEM; |
| 673 | } else if (*buffer == '-') { |
| 674 | nodenum = simple_strtoul(buffer+1, NULL,10); |
| 675 | if (clusterip_del_node(c, nodenum)) |
| 676 | return -ENOENT; |
| 677 | } else |
| 678 | return -EIO; |
| 679 | |
| 680 | return size; |
| 681 | } |
| 682 | |
Arjan van de Ven | 9a32144 | 2007-02-12 00:55:35 -0800 | [diff] [blame] | 683 | static const struct file_operations clusterip_proc_fops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | .owner = THIS_MODULE, |
| 685 | .open = clusterip_proc_open, |
| 686 | .read = seq_read, |
| 687 | .write = clusterip_proc_write, |
| 688 | .llseek = seq_lseek, |
| 689 | .release = clusterip_proc_release, |
| 690 | }; |
| 691 | |
| 692 | #endif /* CONFIG_PROC_FS */ |
| 693 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 694 | static int __init clusterip_tg_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | { |
| 696 | int ret; |
| 697 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 698 | ret = xt_register_target(&clusterip_tg_reg); |
Patrick McHardy | 32292a7 | 2006-04-06 14:11:30 -0700 | [diff] [blame] | 699 | if (ret < 0) |
| 700 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 701 | |
Patrick McHardy | 32292a7 | 2006-04-06 14:11:30 -0700 | [diff] [blame] | 702 | ret = nf_register_hook(&cip_arp_ops); |
| 703 | if (ret < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | goto cleanup_target; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 705 | |
| 706 | #ifdef CONFIG_PROC_FS |
Eric W. Biederman | 457c4cb | 2007-09-12 12:01:34 +0200 | [diff] [blame] | 707 | clusterip_procdir = proc_mkdir("ipt_CLUSTERIP", init_net.proc_net); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | if (!clusterip_procdir) { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 709 | pr_err("Unable to proc dir entry\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 710 | ret = -ENOMEM; |
| 711 | goto cleanup_hook; |
| 712 | } |
| 713 | #endif /* CONFIG_PROC_FS */ |
| 714 | |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 715 | pr_info("ClusterIP Version %s loaded successfully\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | CLUSTERIP_VERSION); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | return 0; |
| 718 | |
Patrick McHardy | 7659258 | 2006-11-29 02:35:42 +0100 | [diff] [blame] | 719 | #ifdef CONFIG_PROC_FS |
Patrick McHardy | 32292a7 | 2006-04-06 14:11:30 -0700 | [diff] [blame] | 720 | cleanup_hook: |
| 721 | nf_unregister_hook(&cip_arp_ops); |
Patrick McHardy | 7659258 | 2006-11-29 02:35:42 +0100 | [diff] [blame] | 722 | #endif /* CONFIG_PROC_FS */ |
Patrick McHardy | 32292a7 | 2006-04-06 14:11:30 -0700 | [diff] [blame] | 723 | cleanup_target: |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 724 | xt_unregister_target(&clusterip_tg_reg); |
Patrick McHardy | 32292a7 | 2006-04-06 14:11:30 -0700 | [diff] [blame] | 725 | return ret; |
| 726 | } |
| 727 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 728 | static void __exit clusterip_tg_exit(void) |
Patrick McHardy | 32292a7 | 2006-04-06 14:11:30 -0700 | [diff] [blame] | 729 | { |
Jan Engelhardt | ff67e4e | 2010-03-19 21:08:16 +0100 | [diff] [blame] | 730 | pr_info("ClusterIP Version %s unloading\n", CLUSTERIP_VERSION); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | #ifdef CONFIG_PROC_FS |
| 732 | remove_proc_entry(clusterip_procdir->name, clusterip_procdir->parent); |
| 733 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | nf_unregister_hook(&cip_arp_ops); |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 735 | xt_unregister_target(&clusterip_tg_reg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 736 | } |
| 737 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 738 | module_init(clusterip_tg_init); |
| 739 | module_exit(clusterip_tg_exit); |