1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * net/sched/act_sample.c - Packet sampling tc action 4 * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com> 5 */ 6 7 #include <linux/types.h> 8 #include <linux/kernel.h> 9 #include <linux/string.h> 10 #include <linux/errno.h> 11 #include <linux/skbuff.h> 12 #include <linux/rtnetlink.h> 13 #include <linux/module.h> 14 #include <linux/init.h> 15 #include <linux/gfp.h> 16 #include <net/net_namespace.h> 17 #include <net/netlink.h> 18 #include <net/pkt_sched.h> 19 #include <linux/tc_act/tc_sample.h> 20 #include <net/tc_act/tc_sample.h> 21 #include <net/psample.h> 22 #include <net/pkt_cls.h> 23 24 #include <linux/if_arp.h> 25 26 static struct tc_action_ops act_sample_ops; 27 28 static const struct nla_policy sample_policy[TCA_SAMPLE_MAX + 1] = { 29 [TCA_SAMPLE_PARMS] = { .len = sizeof(struct tc_sample) }, 30 [TCA_SAMPLE_RATE] = { .type = NLA_U32 }, 31 [TCA_SAMPLE_TRUNC_SIZE] = { .type = NLA_U32 }, 32 [TCA_SAMPLE_PSAMPLE_GROUP] = { .type = NLA_U32 }, 33 }; 34 35 static int tcf_sample_init(struct net *net, struct nlattr *nla, 36 struct nlattr *est, struct tc_action **a, 37 struct tcf_proto *tp, 38 u32 flags, struct netlink_ext_ack *extack) 39 { 40 struct tc_action_net *tn = net_generic(net, act_sample_ops.net_id); 41 bool bind = flags & TCA_ACT_FLAGS_BIND; 42 struct nlattr *tb[TCA_SAMPLE_MAX + 1]; 43 struct psample_group *psample_group; 44 u32 psample_group_num, rate, index; 45 struct tcf_chain *goto_ch = NULL; 46 struct tc_sample *parm; 47 struct tcf_sample *s; 48 bool exists = false; 49 int ret, err; 50 51 if (!nla) 52 return -EINVAL; 53 ret = nla_parse_nested_deprecated(tb, TCA_SAMPLE_MAX, nla, 54 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 index = parm->index; 63 err = tcf_idr_check_alloc(tn, &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, index, est, a, 72 &act_sample_ops, bind, true, flags); 73 if (ret) { 74 tcf_idr_cleanup(tn, index); 75 return ret; 76 } 77 ret = ACT_P_CREATED; 78 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 79 tcf_idr_release(*a, bind); 80 return -EEXIST; 81 } 82 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 83 if (err < 0) 84 goto release_idr; 85 86 rate = nla_get_u32(tb[TCA_SAMPLE_RATE]); 87 if (!rate) { 88 NL_SET_ERR_MSG(extack, "invalid sample rate"); 89 err = -EINVAL; 90 goto put_chain; 91 } 92 psample_group_num = nla_get_u32(tb[TCA_SAMPLE_PSAMPLE_GROUP]); 93 psample_group = psample_group_get(net, psample_group_num); 94 if (!psample_group) { 95 err = -ENOMEM; 96 goto put_chain; 97 } 98 99 s = to_sample(*a); 100 101 spin_lock_bh(&s->tcf_lock); 102 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 103 s->rate = rate; 104 s->psample_group_num = psample_group_num; 105 psample_group = rcu_replace_pointer(s->psample_group, psample_group, 106 lockdep_is_held(&s->tcf_lock)); 107 108 if (tb[TCA_SAMPLE_TRUNC_SIZE]) { 109 s->truncate = true; 110 s->trunc_size = nla_get_u32(tb[TCA_SAMPLE_TRUNC_SIZE]); 111 } 112 spin_unlock_bh(&s->tcf_lock); 113 114 if (psample_group) 115 psample_group_put(psample_group); 116 if (goto_ch) 117 tcf_chain_put_by_act(goto_ch); 118 119 return ret; 120 put_chain: 121 if (goto_ch) 122 tcf_chain_put_by_act(goto_ch); 123 release_idr: 124 tcf_idr_release(*a, bind); 125 return err; 126 } 127 128 static void tcf_sample_cleanup(struct tc_action *a) 129 { 130 struct tcf_sample *s = to_sample(a); 131 struct psample_group *psample_group; 132 133 /* last reference to action, no need to lock */ 134 psample_group = rcu_dereference_protected(s->psample_group, 1); 135 RCU_INIT_POINTER(s->psample_group, NULL); 136 if (psample_group) 137 psample_group_put(psample_group); 138 } 139 140 static bool tcf_sample_dev_ok_push(struct net_device *dev) 141 { 142 switch (dev->type) { 143 case ARPHRD_TUNNEL: 144 case ARPHRD_TUNNEL6: 145 case ARPHRD_SIT: 146 case ARPHRD_IPGRE: 147 case ARPHRD_IP6GRE: 148 case ARPHRD_VOID: 149 case ARPHRD_NONE: 150 return false; 151 default: 152 return true; 153 } 154 } 155 156 static int tcf_sample_act(struct sk_buff *skb, const struct tc_action *a, 157 struct tcf_result *res) 158 { 159 struct tcf_sample *s = to_sample(a); 160 struct psample_group *psample_group; 161 struct psample_metadata md = {}; 162 int retval; 163 164 tcf_lastuse_update(&s->tcf_tm); 165 bstats_update(this_cpu_ptr(s->common.cpu_bstats), skb); 166 retval = READ_ONCE(s->tcf_action); 167 168 psample_group = rcu_dereference_bh(s->psample_group); 169 170 /* randomly sample packets according to rate */ 171 if (psample_group && (prandom_u32() % s->rate == 0)) { 172 if (!skb_at_tc_ingress(skb)) { 173 md.in_ifindex = skb->skb_iif; 174 md.out_ifindex = skb->dev->ifindex; 175 } else { 176 md.in_ifindex = skb->dev->ifindex; 177 } 178 179 /* on ingress, the mac header gets popped, so push it back */ 180 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev)) 181 skb_push(skb, skb->mac_len); 182 183 md.trunc_size = s->truncate ? s->trunc_size : skb->len; 184 psample_sample_packet(psample_group, skb, s->rate, &md); 185 186 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev)) 187 skb_pull(skb, skb->mac_len); 188 } 189 190 return retval; 191 } 192 193 static void tcf_sample_stats_update(struct tc_action *a, u64 bytes, u64 packets, 194 u64 drops, u64 lastuse, bool hw) 195 { 196 struct tcf_sample *s = to_sample(a); 197 struct tcf_t *tm = &s->tcf_tm; 198 199 tcf_action_update_stats(a, bytes, packets, drops, hw); 200 tm->lastuse = max_t(u64, tm->lastuse, lastuse); 201 } 202 203 static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a, 204 int bind, int ref) 205 { 206 unsigned char *b = skb_tail_pointer(skb); 207 struct tcf_sample *s = to_sample(a); 208 struct tc_sample opt = { 209 .index = s->tcf_index, 210 .refcnt = refcount_read(&s->tcf_refcnt) - ref, 211 .bindcnt = atomic_read(&s->tcf_bindcnt) - bind, 212 }; 213 struct tcf_t t; 214 215 spin_lock_bh(&s->tcf_lock); 216 opt.action = s->tcf_action; 217 if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt)) 218 goto nla_put_failure; 219 220 tcf_tm_dump(&t, &s->tcf_tm); 221 if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD)) 222 goto nla_put_failure; 223 224 if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate)) 225 goto nla_put_failure; 226 227 if (s->truncate) 228 if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size)) 229 goto nla_put_failure; 230 231 if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num)) 232 goto nla_put_failure; 233 spin_unlock_bh(&s->tcf_lock); 234 235 return skb->len; 236 237 nla_put_failure: 238 spin_unlock_bh(&s->tcf_lock); 239 nlmsg_trim(skb, b); 240 return -1; 241 } 242 243 static void tcf_psample_group_put(void *priv) 244 { 245 struct psample_group *group = priv; 246 247 psample_group_put(group); 248 } 249 250 static struct psample_group * 251 tcf_sample_get_group(const struct tc_action *a, 252 tc_action_priv_destructor *destructor) 253 { 254 struct tcf_sample *s = to_sample(a); 255 struct psample_group *group; 256 257 group = rcu_dereference_protected(s->psample_group, 258 lockdep_is_held(&s->tcf_lock)); 259 if (group) { 260 psample_group_take(group); 261 *destructor = tcf_psample_group_put; 262 } 263 264 return group; 265 } 266 267 static void tcf_offload_sample_get_group(struct flow_action_entry *entry, 268 const struct tc_action *act) 269 { 270 entry->sample.psample_group = 271 act->ops->get_psample_group(act, &entry->destructor); 272 entry->destructor_priv = entry->sample.psample_group; 273 } 274 275 static int tcf_sample_offload_act_setup(struct tc_action *act, void *entry_data, 276 u32 *index_inc, bool bind, 277 struct netlink_ext_ack *extack) 278 { 279 if (bind) { 280 struct flow_action_entry *entry = entry_data; 281 282 entry->id = FLOW_ACTION_SAMPLE; 283 entry->sample.trunc_size = tcf_sample_trunc_size(act); 284 entry->sample.truncate = tcf_sample_truncate(act); 285 entry->sample.rate = tcf_sample_rate(act); 286 tcf_offload_sample_get_group(entry, act); 287 *index_inc = 1; 288 } else { 289 struct flow_offload_action *fl_action = entry_data; 290 291 fl_action->id = FLOW_ACTION_SAMPLE; 292 } 293 294 return 0; 295 } 296 297 static struct tc_action_ops act_sample_ops = { 298 .kind = "sample", 299 .id = TCA_ID_SAMPLE, 300 .owner = THIS_MODULE, 301 .act = tcf_sample_act, 302 .stats_update = tcf_sample_stats_update, 303 .dump = tcf_sample_dump, 304 .init = tcf_sample_init, 305 .cleanup = tcf_sample_cleanup, 306 .get_psample_group = tcf_sample_get_group, 307 .offload_act_setup = tcf_sample_offload_act_setup, 308 .size = sizeof(struct tcf_sample), 309 }; 310 311 static __net_init int sample_init_net(struct net *net) 312 { 313 struct tc_action_net *tn = net_generic(net, act_sample_ops.net_id); 314 315 return tc_action_net_init(net, tn, &act_sample_ops); 316 } 317 318 static void __net_exit sample_exit_net(struct list_head *net_list) 319 { 320 tc_action_net_exit(net_list, act_sample_ops.net_id); 321 } 322 323 static struct pernet_operations sample_net_ops = { 324 .init = sample_init_net, 325 .exit_batch = sample_exit_net, 326 .id = &act_sample_ops.net_id, 327 .size = sizeof(struct tc_action_net), 328 }; 329 330 static int __init sample_init_module(void) 331 { 332 return tcf_register_action(&act_sample_ops, &sample_net_ops); 333 } 334 335 static void __exit sample_cleanup_module(void) 336 { 337 tcf_unregister_action(&act_sample_ops, &sample_net_ops); 338 } 339 340 module_init(sample_init_module); 341 module_exit(sample_cleanup_module); 342 343 MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>"); 344 MODULE_DESCRIPTION("Packet sampling action"); 345 MODULE_LICENSE("GPL v2"); 346