Marcin Garski | db955170 | 2007-10-19 23:22:11 +0200 | [diff] [blame] | 1 | /* (C) 1999 Jérôme de Vivie <devivie@info.enserb.u-bordeaux.fr> |
| 2 | * (C) 1999 Hervé Eychenne <eychenne@info.enserb.u-bordeaux.fr> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
| 4 | * This program is free software; you can redistribute it and/or modify |
| 5 | * it under the terms of the GNU General Public License version 2 as |
| 6 | * published by the Free Software Foundation. |
| 7 | */ |
| 8 | |
| 9 | #include <linux/module.h> |
| 10 | #include <linux/skbuff.h> |
| 11 | #include <linux/spinlock.h> |
| 12 | #include <linux/interrupt.h> |
| 13 | |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 14 | #include <linux/netfilter/x_tables.h> |
| 15 | #include <linux/netfilter/xt_limit.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | |
| 17 | MODULE_LICENSE("GPL"); |
| 18 | MODULE_AUTHOR("Herve Eychenne <rv@wallfire.org>"); |
Jan Engelhardt | 2ae15b6 | 2008-01-14 23:42:28 -0800 | [diff] [blame^] | 19 | MODULE_DESCRIPTION("Xtables: rate-limit match"); |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 20 | MODULE_ALIAS("ipt_limit"); |
| 21 | MODULE_ALIAS("ip6t_limit"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | |
| 23 | /* The algorithm used is the Simple Token Bucket Filter (TBF) |
| 24 | * see net/sched/sch_tbf.c in the linux source tree |
| 25 | */ |
| 26 | |
| 27 | static DEFINE_SPINLOCK(limit_lock); |
| 28 | |
| 29 | /* Rusty: This is my (non-mathematically-inclined) understanding of |
| 30 | this algorithm. The `average rate' in jiffies becomes your initial |
| 31 | amount of credit `credit' and the most credit you can ever have |
| 32 | `credit_cap'. The `peak rate' becomes the cost of passing the |
| 33 | test, `cost'. |
| 34 | |
| 35 | `prev' tracks the last packet hit: you gain one credit per jiffy. |
| 36 | If you get credit balance more than this, the extra credit is |
| 37 | discarded. Every time the match passes, you lose `cost' credits; |
| 38 | if you don't have that many, the test fails. |
| 39 | |
| 40 | See Alexey's formal explanation in net/sched/sch_tbf.c. |
| 41 | |
| 42 | To get the maxmum range, we multiply by this factor (ie. you get N |
| 43 | credits per jiffy). We want to allow a rate as low as 1 per day |
| 44 | (slowest userspace tool allows), which means |
| 45 | CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32. ie. */ |
| 46 | #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24)) |
| 47 | |
| 48 | /* Repeated shift and or gives us all 1s, final shift and add 1 gives |
| 49 | * us the power of 2 below the theoretical max, so GCC simply does a |
| 50 | * shift. */ |
| 51 | #define _POW2_BELOW2(x) ((x)|((x)>>1)) |
| 52 | #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2)) |
| 53 | #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4)) |
| 54 | #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8)) |
| 55 | #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16)) |
| 56 | #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1) |
| 57 | |
| 58 | #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ) |
| 59 | |
Jan Engelhardt | 1d93a9c | 2007-07-07 22:15:35 -0700 | [diff] [blame] | 60 | static bool |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 61 | limit_mt(const struct sk_buff *skb, const struct net_device *in, |
| 62 | const struct net_device *out, const struct xt_match *match, |
| 63 | const void *matchinfo, int offset, unsigned int protoff, |
| 64 | bool *hotdrop) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | { |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 66 | struct xt_rateinfo *r = |
| 67 | ((const struct xt_rateinfo *)matchinfo)->master; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | unsigned long now = jiffies; |
| 69 | |
| 70 | spin_lock_bh(&limit_lock); |
| 71 | r->credit += (now - xchg(&r->prev, now)) * CREDITS_PER_JIFFY; |
| 72 | if (r->credit > r->credit_cap) |
| 73 | r->credit = r->credit_cap; |
| 74 | |
| 75 | if (r->credit >= r->cost) { |
| 76 | /* We're not limited. */ |
| 77 | r->credit -= r->cost; |
| 78 | spin_unlock_bh(&limit_lock); |
Jan Engelhardt | 1d93a9c | 2007-07-07 22:15:35 -0700 | [diff] [blame] | 79 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | } |
| 81 | |
YOSHIFUJI Hideaki | 601e68e | 2007-02-12 11:15:49 -0800 | [diff] [blame] | 82 | spin_unlock_bh(&limit_lock); |
Jan Engelhardt | 1d93a9c | 2007-07-07 22:15:35 -0700 | [diff] [blame] | 83 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | } |
| 85 | |
| 86 | /* Precision saver. */ |
| 87 | static u_int32_t |
| 88 | user2credits(u_int32_t user) |
| 89 | { |
| 90 | /* If multiplying would overflow... */ |
| 91 | if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY)) |
| 92 | /* Divide first. */ |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 93 | return (user / XT_LIMIT_SCALE) * HZ * CREDITS_PER_JIFFY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 95 | return (user * HZ * CREDITS_PER_JIFFY) / XT_LIMIT_SCALE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | } |
| 97 | |
Jan Engelhardt | ccb79bd | 2007-07-07 22:16:00 -0700 | [diff] [blame] | 98 | static bool |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 99 | limit_mt_check(const char *tablename, const void *inf, |
| 100 | const struct xt_match *match, void *matchinfo, |
| 101 | unsigned int hook_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | { |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 103 | struct xt_rateinfo *r = matchinfo; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | /* Check for overflow. */ |
| 106 | if (r->burst == 0 |
| 107 | || user2credits(r->avg * r->burst) < user2credits(r->avg)) { |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 108 | printk("Overflow in xt_limit, try lower: %u/%u\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | r->avg, r->burst); |
Jan Engelhardt | ccb79bd | 2007-07-07 22:16:00 -0700 | [diff] [blame] | 110 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | } |
| 112 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | /* For SMP, we only want to use one set of counters. */ |
| 114 | r->master = r; |
Patrick McHardy | 57dab5d | 2006-09-20 11:59:25 -0700 | [diff] [blame] | 115 | if (r->cost == 0) { |
| 116 | /* User avg in seconds * XT_LIMIT_SCALE: convert to jiffies * |
| 117 | 128. */ |
| 118 | r->prev = jiffies; |
| 119 | r->credit = user2credits(r->avg * r->burst); /* Credits full. */ |
| 120 | r->credit_cap = user2credits(r->avg * r->burst); /* Credits full. */ |
| 121 | r->cost = user2credits(r->avg); |
| 122 | } |
Jan Engelhardt | ccb79bd | 2007-07-07 22:16:00 -0700 | [diff] [blame] | 123 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | } |
| 125 | |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 126 | #ifdef CONFIG_COMPAT |
| 127 | struct compat_xt_rateinfo { |
| 128 | u_int32_t avg; |
| 129 | u_int32_t burst; |
| 130 | |
| 131 | compat_ulong_t prev; |
| 132 | u_int32_t credit; |
| 133 | u_int32_t credit_cap, cost; |
| 134 | |
| 135 | u_int32_t master; |
| 136 | }; |
| 137 | |
| 138 | /* To keep the full "prev" timestamp, the upper 32 bits are stored in the |
| 139 | * master pointer, which does not need to be preserved. */ |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 140 | static void limit_mt_compat_from_user(void *dst, void *src) |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 141 | { |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 142 | const struct compat_xt_rateinfo *cm = src; |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 143 | struct xt_rateinfo m = { |
| 144 | .avg = cm->avg, |
| 145 | .burst = cm->burst, |
| 146 | .prev = cm->prev | (unsigned long)cm->master << 32, |
| 147 | .credit = cm->credit, |
| 148 | .credit_cap = cm->credit_cap, |
| 149 | .cost = cm->cost, |
| 150 | }; |
| 151 | memcpy(dst, &m, sizeof(m)); |
| 152 | } |
| 153 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 154 | static int limit_mt_compat_to_user(void __user *dst, void *src) |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 155 | { |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 156 | const struct xt_rateinfo *m = src; |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 157 | struct compat_xt_rateinfo cm = { |
| 158 | .avg = m->avg, |
| 159 | .burst = m->burst, |
| 160 | .prev = m->prev, |
| 161 | .credit = m->credit, |
| 162 | .credit_cap = m->credit_cap, |
| 163 | .cost = m->cost, |
| 164 | .master = m->prev >> 32, |
| 165 | }; |
| 166 | return copy_to_user(dst, &cm, sizeof(cm)) ? -EFAULT : 0; |
| 167 | } |
| 168 | #endif /* CONFIG_COMPAT */ |
| 169 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 170 | static struct xt_match limit_mt_reg[] __read_mostly = { |
Patrick McHardy | 4470bbc | 2006-08-22 00:34:04 -0700 | [diff] [blame] | 171 | { |
| 172 | .name = "limit", |
| 173 | .family = AF_INET, |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 174 | .checkentry = limit_mt_check, |
| 175 | .match = limit_mt, |
Patrick McHardy | 4470bbc | 2006-08-22 00:34:04 -0700 | [diff] [blame] | 176 | .matchsize = sizeof(struct xt_rateinfo), |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 177 | #ifdef CONFIG_COMPAT |
| 178 | .compatsize = sizeof(struct compat_xt_rateinfo), |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 179 | .compat_from_user = limit_mt_compat_from_user, |
| 180 | .compat_to_user = limit_mt_compat_to_user, |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 181 | #endif |
Patrick McHardy | 4470bbc | 2006-08-22 00:34:04 -0700 | [diff] [blame] | 182 | .me = THIS_MODULE, |
| 183 | }, |
| 184 | { |
| 185 | .name = "limit", |
| 186 | .family = AF_INET6, |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 187 | .checkentry = limit_mt_check, |
| 188 | .match = limit_mt, |
Patrick McHardy | 4470bbc | 2006-08-22 00:34:04 -0700 | [diff] [blame] | 189 | .matchsize = sizeof(struct xt_rateinfo), |
Patrick McHardy | 34f4c42 | 2007-12-17 21:50:53 -0800 | [diff] [blame] | 190 | #ifdef CONFIG_COMPAT |
| 191 | .compatsize = sizeof(struct compat_xt_rateinfo), |
| 192 | .compat_from_user = limit_mt_compat_from_user, |
| 193 | .compat_to_user = limit_mt_compat_to_user, |
| 194 | #endif |
Patrick McHardy | 4470bbc | 2006-08-22 00:34:04 -0700 | [diff] [blame] | 195 | .me = THIS_MODULE, |
| 196 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | }; |
| 198 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 199 | static int __init limit_mt_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | { |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 201 | return xt_register_matches(limit_mt_reg, ARRAY_SIZE(limit_mt_reg)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | } |
| 203 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 204 | static void __exit limit_mt_exit(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | { |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 206 | xt_unregister_matches(limit_mt_reg, ARRAY_SIZE(limit_mt_reg)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | } |
| 208 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 209 | module_init(limit_mt_init); |
| 210 | module_exit(limit_mt_exit); |