1 /* 2 * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us> 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 as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 */ 9 10 #include <linux/module.h> 11 #include <linux/init.h> 12 #include <linux/kernel.h> 13 #include <linux/skbuff.h> 14 #include <linux/rtnetlink.h> 15 #include <linux/filter.h> 16 #include <linux/bpf.h> 17 18 #include <net/netlink.h> 19 #include <net/pkt_sched.h> 20 21 #include <linux/tc_act/tc_bpf.h> 22 #include <net/tc_act/tc_bpf.h> 23 24 #define BPF_TAB_MASK 15 25 #define ACT_BPF_NAME_LEN 256 26 27 struct tcf_bpf_cfg { 28 struct bpf_prog *filter; 29 struct sock_filter *bpf_ops; 30 const char *bpf_name; 31 u16 bpf_num_ops; 32 bool is_ebpf; 33 }; 34 35 static unsigned int bpf_net_id; 36 static struct tc_action_ops act_bpf_ops; 37 38 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act, 39 struct tcf_result *res) 40 { 41 bool at_ingress = skb_at_tc_ingress(skb); 42 struct tcf_bpf *prog = to_bpf(act); 43 struct bpf_prog *filter; 44 int action, filter_res; 45 46 tcf_lastuse_update(&prog->tcf_tm); 47 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb); 48 49 rcu_read_lock(); 50 filter = rcu_dereference(prog->filter); 51 if (at_ingress) { 52 __skb_push(skb, skb->mac_len); 53 bpf_compute_data_end(skb); 54 filter_res = BPF_PROG_RUN(filter, skb); 55 __skb_pull(skb, skb->mac_len); 56 } else { 57 bpf_compute_data_end(skb); 58 filter_res = BPF_PROG_RUN(filter, skb); 59 } 60 rcu_read_unlock(); 61 62 /* A BPF program may overwrite the default action opcode. 63 * Similarly as in cls_bpf, if filter_res == -1 we use the 64 * default action specified from tc. 65 * 66 * In case a different well-known TC_ACT opcode has been 67 * returned, it will overwrite the default one. 68 * 69 * For everything else that is unkown, TC_ACT_UNSPEC is 70 * returned. 71 */ 72 switch (filter_res) { 73 case TC_ACT_PIPE: 74 case TC_ACT_RECLASSIFY: 75 case TC_ACT_OK: 76 case TC_ACT_REDIRECT: 77 action = filter_res; 78 break; 79 case TC_ACT_SHOT: 80 action = filter_res; 81 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats)); 82 break; 83 case TC_ACT_UNSPEC: 84 action = prog->tcf_action; 85 break; 86 default: 87 action = TC_ACT_UNSPEC; 88 break; 89 } 90 91 return action; 92 } 93 94 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog) 95 { 96 return !prog->bpf_ops; 97 } 98 99 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog, 100 struct sk_buff *skb) 101 { 102 struct nlattr *nla; 103 104 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops)) 105 return -EMSGSIZE; 106 107 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops * 108 sizeof(struct sock_filter)); 109 if (nla == NULL) 110 return -EMSGSIZE; 111 112 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla)); 113 114 return 0; 115 } 116 117 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog, 118 struct sk_buff *skb) 119 { 120 struct nlattr *nla; 121 122 if (prog->bpf_name && 123 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name)) 124 return -EMSGSIZE; 125 126 if (nla_put_u32(skb, TCA_ACT_BPF_ID, prog->filter->aux->id)) 127 return -EMSGSIZE; 128 129 nla = nla_reserve(skb, TCA_ACT_BPF_TAG, sizeof(prog->filter->tag)); 130 if (nla == NULL) 131 return -EMSGSIZE; 132 133 memcpy(nla_data(nla), prog->filter->tag, nla_len(nla)); 134 135 return 0; 136 } 137 138 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act, 139 int bind, int ref) 140 { 141 unsigned char *tp = skb_tail_pointer(skb); 142 struct tcf_bpf *prog = to_bpf(act); 143 struct tc_act_bpf opt = { 144 .index = prog->tcf_index, 145 .refcnt = prog->tcf_refcnt - ref, 146 .bindcnt = prog->tcf_bindcnt - bind, 147 .action = prog->tcf_action, 148 }; 149 struct tcf_t tm; 150 int ret; 151 152 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt)) 153 goto nla_put_failure; 154 155 if (tcf_bpf_is_ebpf(prog)) 156 ret = tcf_bpf_dump_ebpf_info(prog, skb); 157 else 158 ret = tcf_bpf_dump_bpf_info(prog, skb); 159 if (ret) 160 goto nla_put_failure; 161 162 tcf_tm_dump(&tm, &prog->tcf_tm); 163 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm, 164 TCA_ACT_BPF_PAD)) 165 goto nla_put_failure; 166 167 return skb->len; 168 169 nla_put_failure: 170 nlmsg_trim(skb, tp); 171 return -1; 172 } 173 174 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = { 175 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) }, 176 [TCA_ACT_BPF_FD] = { .type = NLA_U32 }, 177 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING, 178 .len = ACT_BPF_NAME_LEN }, 179 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 }, 180 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY, 181 .len = sizeof(struct sock_filter) * BPF_MAXINSNS }, 182 }; 183 184 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg) 185 { 186 struct sock_filter *bpf_ops; 187 struct sock_fprog_kern fprog_tmp; 188 struct bpf_prog *fp; 189 u16 bpf_size, bpf_num_ops; 190 int ret; 191 192 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]); 193 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0) 194 return -EINVAL; 195 196 bpf_size = bpf_num_ops * sizeof(*bpf_ops); 197 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS])) 198 return -EINVAL; 199 200 bpf_ops = kzalloc(bpf_size, GFP_KERNEL); 201 if (bpf_ops == NULL) 202 return -ENOMEM; 203 204 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size); 205 206 fprog_tmp.len = bpf_num_ops; 207 fprog_tmp.filter = bpf_ops; 208 209 ret = bpf_prog_create(&fp, &fprog_tmp); 210 if (ret < 0) { 211 kfree(bpf_ops); 212 return ret; 213 } 214 215 cfg->bpf_ops = bpf_ops; 216 cfg->bpf_num_ops = bpf_num_ops; 217 cfg->filter = fp; 218 cfg->is_ebpf = false; 219 220 return 0; 221 } 222 223 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg) 224 { 225 struct bpf_prog *fp; 226 char *name = NULL; 227 u32 bpf_fd; 228 229 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]); 230 231 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT); 232 if (IS_ERR(fp)) 233 return PTR_ERR(fp); 234 235 if (tb[TCA_ACT_BPF_NAME]) { 236 name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL); 237 if (!name) { 238 bpf_prog_put(fp); 239 return -ENOMEM; 240 } 241 } 242 243 cfg->bpf_name = name; 244 cfg->filter = fp; 245 cfg->is_ebpf = true; 246 247 return 0; 248 } 249 250 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg) 251 { 252 if (cfg->is_ebpf) 253 bpf_prog_put(cfg->filter); 254 else 255 bpf_prog_destroy(cfg->filter); 256 257 kfree(cfg->bpf_ops); 258 kfree(cfg->bpf_name); 259 } 260 261 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog, 262 struct tcf_bpf_cfg *cfg) 263 { 264 cfg->is_ebpf = tcf_bpf_is_ebpf(prog); 265 /* updates to prog->filter are prevented, since it's called either 266 * with rtnl lock or during final cleanup in rcu callback 267 */ 268 cfg->filter = rcu_dereference_protected(prog->filter, 1); 269 270 cfg->bpf_ops = prog->bpf_ops; 271 cfg->bpf_name = prog->bpf_name; 272 } 273 274 static int tcf_bpf_init(struct net *net, struct nlattr *nla, 275 struct nlattr *est, struct tc_action **act, 276 int replace, int bind) 277 { 278 struct tc_action_net *tn = net_generic(net, bpf_net_id); 279 struct nlattr *tb[TCA_ACT_BPF_MAX + 1]; 280 struct tcf_bpf_cfg cfg, old; 281 struct tc_act_bpf *parm; 282 struct tcf_bpf *prog; 283 bool is_bpf, is_ebpf; 284 int ret, res = 0; 285 286 if (!nla) 287 return -EINVAL; 288 289 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy, NULL); 290 if (ret < 0) 291 return ret; 292 293 if (!tb[TCA_ACT_BPF_PARMS]) 294 return -EINVAL; 295 296 parm = nla_data(tb[TCA_ACT_BPF_PARMS]); 297 298 if (!tcf_hash_check(tn, parm->index, act, bind)) { 299 ret = tcf_hash_create(tn, parm->index, est, act, 300 &act_bpf_ops, bind, true); 301 if (ret < 0) 302 return ret; 303 304 res = ACT_P_CREATED; 305 } else { 306 /* Don't override defaults. */ 307 if (bind) 308 return 0; 309 310 tcf_hash_release(*act, bind); 311 if (!replace) 312 return -EEXIST; 313 } 314 315 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS]; 316 is_ebpf = tb[TCA_ACT_BPF_FD]; 317 318 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) { 319 ret = -EINVAL; 320 goto out; 321 } 322 323 memset(&cfg, 0, sizeof(cfg)); 324 325 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) : 326 tcf_bpf_init_from_efd(tb, &cfg); 327 if (ret < 0) 328 goto out; 329 330 prog = to_bpf(*act); 331 ASSERT_RTNL(); 332 333 if (res != ACT_P_CREATED) 334 tcf_bpf_prog_fill_cfg(prog, &old); 335 336 prog->bpf_ops = cfg.bpf_ops; 337 prog->bpf_name = cfg.bpf_name; 338 339 if (cfg.bpf_num_ops) 340 prog->bpf_num_ops = cfg.bpf_num_ops; 341 342 prog->tcf_action = parm->action; 343 rcu_assign_pointer(prog->filter, cfg.filter); 344 345 if (res == ACT_P_CREATED) { 346 tcf_hash_insert(tn, *act); 347 } else { 348 /* make sure the program being replaced is no longer executing */ 349 synchronize_rcu(); 350 tcf_bpf_cfg_cleanup(&old); 351 } 352 353 return res; 354 out: 355 if (res == ACT_P_CREATED) 356 tcf_hash_cleanup(*act, est); 357 358 return ret; 359 } 360 361 static void tcf_bpf_cleanup(struct tc_action *act, int bind) 362 { 363 struct tcf_bpf_cfg tmp; 364 365 tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp); 366 tcf_bpf_cfg_cleanup(&tmp); 367 } 368 369 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb, 370 struct netlink_callback *cb, int type, 371 const struct tc_action_ops *ops) 372 { 373 struct tc_action_net *tn = net_generic(net, bpf_net_id); 374 375 return tcf_generic_walker(tn, skb, cb, type, ops); 376 } 377 378 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index) 379 { 380 struct tc_action_net *tn = net_generic(net, bpf_net_id); 381 382 return tcf_hash_search(tn, a, index); 383 } 384 385 static struct tc_action_ops act_bpf_ops __read_mostly = { 386 .kind = "bpf", 387 .type = TCA_ACT_BPF, 388 .owner = THIS_MODULE, 389 .act = tcf_bpf, 390 .dump = tcf_bpf_dump, 391 .cleanup = tcf_bpf_cleanup, 392 .init = tcf_bpf_init, 393 .walk = tcf_bpf_walker, 394 .lookup = tcf_bpf_search, 395 .size = sizeof(struct tcf_bpf), 396 }; 397 398 static __net_init int bpf_init_net(struct net *net) 399 { 400 struct tc_action_net *tn = net_generic(net, bpf_net_id); 401 402 return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK); 403 } 404 405 static void __net_exit bpf_exit_net(struct net *net) 406 { 407 struct tc_action_net *tn = net_generic(net, bpf_net_id); 408 409 tc_action_net_exit(tn); 410 } 411 412 static struct pernet_operations bpf_net_ops = { 413 .init = bpf_init_net, 414 .exit = bpf_exit_net, 415 .id = &bpf_net_id, 416 .size = sizeof(struct tc_action_net), 417 }; 418 419 static int __init bpf_init_module(void) 420 { 421 return tcf_register_action(&act_bpf_ops, &bpf_net_ops); 422 } 423 424 static void __exit bpf_cleanup_module(void) 425 { 426 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops); 427 } 428 429 module_init(bpf_init_module); 430 module_exit(bpf_cleanup_module); 431 432 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>"); 433 MODULE_DESCRIPTION("TC BPF based action"); 434 MODULE_LICENSE("GPL v2"); 435