1 /* 2 * net/sched/act_sample.c - Packet sampling tc action 3 * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com> 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License version 2 as 7 * published by the Free Software Foundation. 8 */ 9 10 #include <linux/types.h> 11 #include <linux/kernel.h> 12 #include <linux/string.h> 13 #include <linux/errno.h> 14 #include <linux/skbuff.h> 15 #include <linux/rtnetlink.h> 16 #include <linux/module.h> 17 #include <linux/init.h> 18 #include <linux/gfp.h> 19 #include <net/net_namespace.h> 20 #include <net/netlink.h> 21 #include <net/pkt_sched.h> 22 #include <linux/tc_act/tc_sample.h> 23 #include <net/tc_act/tc_sample.h> 24 #include <net/psample.h> 25 26 #include <linux/if_arp.h> 27 28 static unsigned int sample_net_id; 29 static struct tc_action_ops act_sample_ops; 30 31 static const struct nla_policy sample_policy[TCA_SAMPLE_MAX + 1] = { 32 [TCA_SAMPLE_PARMS] = { .len = sizeof(struct tc_sample) }, 33 [TCA_SAMPLE_RATE] = { .type = NLA_U32 }, 34 [TCA_SAMPLE_TRUNC_SIZE] = { .type = NLA_U32 }, 35 [TCA_SAMPLE_PSAMPLE_GROUP] = { .type = NLA_U32 }, 36 }; 37 38 static int tcf_sample_init(struct net *net, struct nlattr *nla, 39 struct nlattr *est, struct tc_action **a, int ovr, 40 int bind, bool rtnl_held, 41 struct netlink_ext_ack *extack) 42 { 43 struct tc_action_net *tn = net_generic(net, sample_net_id); 44 struct nlattr *tb[TCA_SAMPLE_MAX + 1]; 45 struct psample_group *psample_group; 46 struct tc_sample *parm; 47 u32 psample_group_num; 48 struct tcf_sample *s; 49 bool exists = false; 50 int ret, err; 51 52 if (!nla) 53 return -EINVAL; 54 ret = nla_parse_nested(tb, TCA_SAMPLE_MAX, nla, sample_policy, NULL); 55 if (ret < 0) 56 return ret; 57 if (!tb[TCA_SAMPLE_PARMS] || !tb[TCA_SAMPLE_RATE] || 58 !tb[TCA_SAMPLE_PSAMPLE_GROUP]) 59 return -EINVAL; 60 61 parm = nla_data(tb[TCA_SAMPLE_PARMS]); 62 63 err = tcf_idr_check_alloc(tn, &parm->index, a, bind); 64 if (err < 0) 65 return err; 66 exists = err; 67 if (exists && bind) 68 return 0; 69 70 if (!exists) { 71 ret = tcf_idr_create(tn, parm->index, est, a, 72 &act_sample_ops, bind, true); 73 if (ret) { 74 tcf_idr_cleanup(tn, parm->index); 75 return ret; 76 } 77 ret = ACT_P_CREATED; 78 } else if (!ovr) { 79 tcf_idr_release(*a, bind); 80 return -EEXIST; 81 } 82 83 psample_group_num = nla_get_u32(tb[TCA_SAMPLE_PSAMPLE_GROUP]); 84 psample_group = psample_group_get(net, psample_group_num); 85 if (!psample_group) { 86 tcf_idr_release(*a, bind); 87 return -ENOMEM; 88 } 89 90 s = to_sample(*a); 91 92 spin_lock_bh(&s->tcf_lock); 93 s->tcf_action = parm->action; 94 s->rate = nla_get_u32(tb[TCA_SAMPLE_RATE]); 95 s->psample_group_num = psample_group_num; 96 RCU_INIT_POINTER(s->psample_group, psample_group); 97 98 if (tb[TCA_SAMPLE_TRUNC_SIZE]) { 99 s->truncate = true; 100 s->trunc_size = nla_get_u32(tb[TCA_SAMPLE_TRUNC_SIZE]); 101 } 102 spin_unlock_bh(&s->tcf_lock); 103 104 if (ret == ACT_P_CREATED) 105 tcf_idr_insert(tn, *a); 106 return ret; 107 } 108 109 static void tcf_sample_cleanup(struct tc_action *a) 110 { 111 struct tcf_sample *s = to_sample(a); 112 struct psample_group *psample_group; 113 114 /* last reference to action, no need to lock */ 115 psample_group = rcu_dereference_protected(s->psample_group, 1); 116 RCU_INIT_POINTER(s->psample_group, NULL); 117 if (psample_group) 118 psample_group_put(psample_group); 119 } 120 121 static bool tcf_sample_dev_ok_push(struct net_device *dev) 122 { 123 switch (dev->type) { 124 case ARPHRD_TUNNEL: 125 case ARPHRD_TUNNEL6: 126 case ARPHRD_SIT: 127 case ARPHRD_IPGRE: 128 case ARPHRD_VOID: 129 case ARPHRD_NONE: 130 return false; 131 default: 132 return true; 133 } 134 } 135 136 static int tcf_sample_act(struct sk_buff *skb, const struct tc_action *a, 137 struct tcf_result *res) 138 { 139 struct tcf_sample *s = to_sample(a); 140 struct psample_group *psample_group; 141 int retval; 142 int size; 143 int iif; 144 int oif; 145 146 tcf_lastuse_update(&s->tcf_tm); 147 bstats_cpu_update(this_cpu_ptr(s->common.cpu_bstats), skb); 148 retval = READ_ONCE(s->tcf_action); 149 150 psample_group = rcu_dereference_bh(s->psample_group); 151 152 /* randomly sample packets according to rate */ 153 if (psample_group && (prandom_u32() % s->rate == 0)) { 154 if (!skb_at_tc_ingress(skb)) { 155 iif = skb->skb_iif; 156 oif = skb->dev->ifindex; 157 } else { 158 iif = skb->dev->ifindex; 159 oif = 0; 160 } 161 162 /* on ingress, the mac header gets popped, so push it back */ 163 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev)) 164 skb_push(skb, skb->mac_len); 165 166 size = s->truncate ? s->trunc_size : skb->len; 167 psample_sample_packet(psample_group, skb, size, iif, oif, 168 s->rate); 169 170 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev)) 171 skb_pull(skb, skb->mac_len); 172 } 173 174 return retval; 175 } 176 177 static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a, 178 int bind, int ref) 179 { 180 unsigned char *b = skb_tail_pointer(skb); 181 struct tcf_sample *s = to_sample(a); 182 struct tc_sample opt = { 183 .index = s->tcf_index, 184 .refcnt = refcount_read(&s->tcf_refcnt) - ref, 185 .bindcnt = atomic_read(&s->tcf_bindcnt) - bind, 186 }; 187 struct tcf_t t; 188 189 spin_lock_bh(&s->tcf_lock); 190 opt.action = s->tcf_action; 191 if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt)) 192 goto nla_put_failure; 193 194 tcf_tm_dump(&t, &s->tcf_tm); 195 if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD)) 196 goto nla_put_failure; 197 198 if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate)) 199 goto nla_put_failure; 200 201 if (s->truncate) 202 if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size)) 203 goto nla_put_failure; 204 205 if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num)) 206 goto nla_put_failure; 207 spin_unlock_bh(&s->tcf_lock); 208 209 return skb->len; 210 211 nla_put_failure: 212 spin_unlock_bh(&s->tcf_lock); 213 nlmsg_trim(skb, b); 214 return -1; 215 } 216 217 static int tcf_sample_walker(struct net *net, struct sk_buff *skb, 218 struct netlink_callback *cb, int type, 219 const struct tc_action_ops *ops, 220 struct netlink_ext_ack *extack) 221 { 222 struct tc_action_net *tn = net_generic(net, sample_net_id); 223 224 return tcf_generic_walker(tn, skb, cb, type, ops, extack); 225 } 226 227 static int tcf_sample_search(struct net *net, struct tc_action **a, u32 index) 228 { 229 struct tc_action_net *tn = net_generic(net, sample_net_id); 230 231 return tcf_idr_search(tn, a, index); 232 } 233 234 static struct tc_action_ops act_sample_ops = { 235 .kind = "sample", 236 .type = TCA_ACT_SAMPLE, 237 .owner = THIS_MODULE, 238 .act = tcf_sample_act, 239 .dump = tcf_sample_dump, 240 .init = tcf_sample_init, 241 .cleanup = tcf_sample_cleanup, 242 .walk = tcf_sample_walker, 243 .lookup = tcf_sample_search, 244 .size = sizeof(struct tcf_sample), 245 }; 246 247 static __net_init int sample_init_net(struct net *net) 248 { 249 struct tc_action_net *tn = net_generic(net, sample_net_id); 250 251 return tc_action_net_init(tn, &act_sample_ops); 252 } 253 254 static void __net_exit sample_exit_net(struct list_head *net_list) 255 { 256 tc_action_net_exit(net_list, sample_net_id); 257 } 258 259 static struct pernet_operations sample_net_ops = { 260 .init = sample_init_net, 261 .exit_batch = sample_exit_net, 262 .id = &sample_net_id, 263 .size = sizeof(struct tc_action_net), 264 }; 265 266 static int __init sample_init_module(void) 267 { 268 return tcf_register_action(&act_sample_ops, &sample_net_ops); 269 } 270 271 static void __exit sample_cleanup_module(void) 272 { 273 tcf_unregister_action(&act_sample_ops, &sample_net_ops); 274 } 275 276 module_init(sample_init_module); 277 module_exit(sample_cleanup_module); 278 279 MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>"); 280 MODULE_DESCRIPTION("Packet sampling action"); 281 MODULE_LICENSE("GPL v2"); 282