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 | */ |
Jan Engelhardt | 8bee4ba | 2010-03-17 16:04:40 +0100 | [diff] [blame] | 8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 10 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | #include <linux/module.h> |
| 12 | #include <linux/skbuff.h> |
| 13 | #include <linux/spinlock.h> |
| 14 | #include <linux/interrupt.h> |
| 15 | |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 16 | #include <linux/netfilter/x_tables.h> |
| 17 | #include <linux/netfilter/xt_limit.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 19 | struct xt_limit_priv { |
| 20 | unsigned long prev; |
| 21 | uint32_t credit; |
| 22 | }; |
| 23 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | MODULE_LICENSE("GPL"); |
| 25 | MODULE_AUTHOR("Herve Eychenne <rv@wallfire.org>"); |
Jan Engelhardt | 2ae15b6 | 2008-01-14 23:42:28 -0800 | [diff] [blame] | 26 | MODULE_DESCRIPTION("Xtables: rate-limit match"); |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 27 | MODULE_ALIAS("ipt_limit"); |
| 28 | MODULE_ALIAS("ip6t_limit"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | |
| 30 | /* The algorithm used is the Simple Token Bucket Filter (TBF) |
| 31 | * see net/sched/sch_tbf.c in the linux source tree |
| 32 | */ |
| 33 | |
| 34 | static DEFINE_SPINLOCK(limit_lock); |
| 35 | |
| 36 | /* Rusty: This is my (non-mathematically-inclined) understanding of |
| 37 | this algorithm. The `average rate' in jiffies becomes your initial |
| 38 | amount of credit `credit' and the most credit you can ever have |
| 39 | `credit_cap'. The `peak rate' becomes the cost of passing the |
| 40 | test, `cost'. |
| 41 | |
| 42 | `prev' tracks the last packet hit: you gain one credit per jiffy. |
| 43 | If you get credit balance more than this, the extra credit is |
| 44 | discarded. Every time the match passes, you lose `cost' credits; |
| 45 | if you don't have that many, the test fails. |
| 46 | |
| 47 | See Alexey's formal explanation in net/sched/sch_tbf.c. |
| 48 | |
| 49 | To get the maxmum range, we multiply by this factor (ie. you get N |
| 50 | credits per jiffy). We want to allow a rate as low as 1 per day |
| 51 | (slowest userspace tool allows), which means |
| 52 | CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32. ie. */ |
| 53 | #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24)) |
| 54 | |
| 55 | /* Repeated shift and or gives us all 1s, final shift and add 1 gives |
| 56 | * us the power of 2 below the theoretical max, so GCC simply does a |
| 57 | * shift. */ |
| 58 | #define _POW2_BELOW2(x) ((x)|((x)>>1)) |
| 59 | #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2)) |
| 60 | #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4)) |
| 61 | #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8)) |
| 62 | #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16)) |
| 63 | #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1) |
| 64 | |
| 65 | #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ) |
| 66 | |
Jan Engelhardt | 1d93a9c | 2007-07-07 22:15:35 -0700 | [diff] [blame] | 67 | static bool |
Jan Engelhardt | 62fc805 | 2009-07-07 20:42:08 +0200 | [diff] [blame] | 68 | limit_mt(const struct sk_buff *skb, struct xt_action_param *par) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | { |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 70 | const struct xt_rateinfo *r = par->matchinfo; |
| 71 | struct xt_limit_priv *priv = r->master; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | unsigned long now = jiffies; |
| 73 | |
| 74 | spin_lock_bh(&limit_lock); |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 75 | priv->credit += (now - xchg(&priv->prev, now)) * CREDITS_PER_JIFFY; |
| 76 | if (priv->credit > r->credit_cap) |
| 77 | priv->credit = r->credit_cap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 79 | if (priv->credit >= r->cost) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | /* We're not limited. */ |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 81 | priv->credit -= r->cost; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | spin_unlock_bh(&limit_lock); |
Jan Engelhardt | 1d93a9c | 2007-07-07 22:15:35 -0700 | [diff] [blame] | 83 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | } |
| 85 | |
YOSHIFUJI Hideaki | 601e68e | 2007-02-12 11:15:49 -0800 | [diff] [blame] | 86 | spin_unlock_bh(&limit_lock); |
Jan Engelhardt | 1d93a9c | 2007-07-07 22:15:35 -0700 | [diff] [blame] | 87 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | } |
| 89 | |
| 90 | /* Precision saver. */ |
Florian Westphal | 7a909ac | 2012-05-07 10:51:43 +0000 | [diff] [blame] | 91 | static u32 user2credits(u32 user) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | { |
| 93 | /* If multiplying would overflow... */ |
| 94 | if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY)) |
| 95 | /* Divide first. */ |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 96 | return (user / XT_LIMIT_SCALE) * HZ * CREDITS_PER_JIFFY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | |
Harald Welte | 2e4e6a1 | 2006-01-12 13:30:04 -0800 | [diff] [blame] | 98 | return (user * HZ * CREDITS_PER_JIFFY) / XT_LIMIT_SCALE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | } |
| 100 | |
Jan Engelhardt | b0f3845 | 2010-03-19 17:16:42 +0100 | [diff] [blame] | 101 | static int limit_mt_check(const struct xt_mtchk_param *par) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | { |
Jan Engelhardt | 9b4fce7 | 2008-10-08 11:35:18 +0200 | [diff] [blame] | 103 | struct xt_rateinfo *r = par->matchinfo; |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 104 | struct xt_limit_priv *priv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | /* Check for overflow. */ |
| 107 | if (r->burst == 0 |
| 108 | || user2credits(r->avg * r->burst) < user2credits(r->avg)) { |
Jan Engelhardt | 8bee4ba | 2010-03-17 16:04:40 +0100 | [diff] [blame] | 109 | pr_info("Overflow, try lower: %u/%u\n", |
| 110 | r->avg, r->burst); |
Jan Engelhardt | 4a5a5c7 | 2010-03-19 17:32:59 +0100 | [diff] [blame] | 111 | return -ERANGE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | } |
| 113 | |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 114 | priv = kmalloc(sizeof(*priv), GFP_KERNEL); |
| 115 | if (priv == NULL) |
Jan Engelhardt | 4a5a5c7 | 2010-03-19 17:32:59 +0100 | [diff] [blame] | 116 | return -ENOMEM; |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 117 | |
| 118 | /* For SMP, we only want to use one set of state. */ |
| 119 | r->master = priv; |
Jan Engelhardt | 82e6bfe | 2012-09-21 22:26:52 +0000 | [diff] [blame] | 120 | /* User avg in seconds * XT_LIMIT_SCALE: convert to jiffies * |
| 121 | 128. */ |
| 122 | priv->prev = jiffies; |
| 123 | priv->credit = user2credits(r->avg * r->burst); /* Credits full. */ |
Patrick McHardy | 57dab5d | 2006-09-20 11:59:25 -0700 | [diff] [blame] | 124 | if (r->cost == 0) { |
Florian Westphal | 7a909ac | 2012-05-07 10:51:43 +0000 | [diff] [blame] | 125 | r->credit_cap = priv->credit; /* Credits full. */ |
Patrick McHardy | 57dab5d | 2006-09-20 11:59:25 -0700 | [diff] [blame] | 126 | r->cost = user2credits(r->avg); |
| 127 | } |
Jan Engelhardt | bd414ee | 2010-03-23 16:35:56 +0100 | [diff] [blame] | 128 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 129 | } |
| 130 | |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 131 | static void limit_mt_destroy(const struct xt_mtdtor_param *par) |
| 132 | { |
| 133 | const struct xt_rateinfo *info = par->matchinfo; |
| 134 | |
| 135 | kfree(info->master); |
| 136 | } |
| 137 | |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 138 | #ifdef CONFIG_COMPAT |
| 139 | struct compat_xt_rateinfo { |
| 140 | u_int32_t avg; |
| 141 | u_int32_t burst; |
| 142 | |
| 143 | compat_ulong_t prev; |
| 144 | u_int32_t credit; |
| 145 | u_int32_t credit_cap, cost; |
| 146 | |
| 147 | u_int32_t master; |
| 148 | }; |
| 149 | |
| 150 | /* To keep the full "prev" timestamp, the upper 32 bits are stored in the |
| 151 | * master pointer, which does not need to be preserved. */ |
Jan Engelhardt | 739674f | 2009-06-26 08:23:19 +0200 | [diff] [blame] | 152 | static void limit_mt_compat_from_user(void *dst, const void *src) |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 153 | { |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 154 | const struct compat_xt_rateinfo *cm = src; |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 155 | struct xt_rateinfo m = { |
| 156 | .avg = cm->avg, |
| 157 | .burst = cm->burst, |
| 158 | .prev = cm->prev | (unsigned long)cm->master << 32, |
| 159 | .credit = cm->credit, |
| 160 | .credit_cap = cm->credit_cap, |
| 161 | .cost = cm->cost, |
| 162 | }; |
| 163 | memcpy(dst, &m, sizeof(m)); |
| 164 | } |
| 165 | |
Jan Engelhardt | 739674f | 2009-06-26 08:23:19 +0200 | [diff] [blame] | 166 | static int limit_mt_compat_to_user(void __user *dst, const void *src) |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 167 | { |
Jan Engelhardt | a47362a | 2007-07-07 22:16:55 -0700 | [diff] [blame] | 168 | const struct xt_rateinfo *m = src; |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 169 | struct compat_xt_rateinfo cm = { |
| 170 | .avg = m->avg, |
| 171 | .burst = m->burst, |
| 172 | .prev = m->prev, |
| 173 | .credit = m->credit, |
| 174 | .credit_cap = m->credit_cap, |
| 175 | .cost = m->cost, |
| 176 | .master = m->prev >> 32, |
| 177 | }; |
| 178 | return copy_to_user(dst, &cm, sizeof(cm)) ? -EFAULT : 0; |
| 179 | } |
| 180 | #endif /* CONFIG_COMPAT */ |
| 181 | |
Jan Engelhardt | 55b69e9 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 182 | static struct xt_match limit_mt_reg __read_mostly = { |
| 183 | .name = "limit", |
| 184 | .revision = 0, |
| 185 | .family = NFPROTO_UNSPEC, |
| 186 | .match = limit_mt, |
| 187 | .checkentry = limit_mt_check, |
Jan Engelhardt | acc738f | 2009-03-16 15:35:29 +0100 | [diff] [blame] | 188 | .destroy = limit_mt_destroy, |
Jan Engelhardt | 55b69e9 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 189 | .matchsize = sizeof(struct xt_rateinfo), |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 190 | #ifdef CONFIG_COMPAT |
Jan Engelhardt | 55b69e9 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 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, |
Patrick McHardy | 02c63cf | 2006-09-20 12:07:06 -0700 | [diff] [blame] | 194 | #endif |
Jan Engelhardt | 55b69e9 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 195 | .me = THIS_MODULE, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | }; |
| 197 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 198 | static int __init limit_mt_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | { |
Jan Engelhardt | 55b69e9 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 200 | return xt_register_match(&limit_mt_reg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | } |
| 202 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 203 | static void __exit limit_mt_exit(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | { |
Jan Engelhardt | 55b69e9 | 2008-10-08 11:35:01 +0200 | [diff] [blame] | 205 | xt_unregister_match(&limit_mt_reg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | } |
| 207 | |
Jan Engelhardt | d3c5ee6 | 2007-12-04 23:24:03 -0800 | [diff] [blame] | 208 | module_init(limit_mt_init); |
| 209 | module_exit(limit_mt_exit); |