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 nla = nla_reserve(skb, TCA_ACT_BPF_DIGEST, 127 sizeof(prog->filter->digest)); 128 if (nla == NULL) 129 return -EMSGSIZE; 130 131 memcpy(nla_data(nla), prog->filter->digest, nla_len(nla)); 132 133 return 0; 134 } 135 136 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act, 137 int bind, int ref) 138 { 139 unsigned char *tp = skb_tail_pointer(skb); 140 struct tcf_bpf *prog = to_bpf(act); 141 struct tc_act_bpf opt = { 142 .index = prog->tcf_index, 143 .refcnt = prog->tcf_refcnt - ref, 144 .bindcnt = prog->tcf_bindcnt - bind, 145 .action = prog->tcf_action, 146 }; 147 struct tcf_t tm; 148 int ret; 149 150 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt)) 151 goto nla_put_failure; 152 153 if (tcf_bpf_is_ebpf(prog)) 154 ret = tcf_bpf_dump_ebpf_info(prog, skb); 155 else 156 ret = tcf_bpf_dump_bpf_info(prog, skb); 157 if (ret) 158 goto nla_put_failure; 159 160 tcf_tm_dump(&tm, &prog->tcf_tm); 161 if (nla_put_64bit(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm, 162 TCA_ACT_BPF_PAD)) 163 goto nla_put_failure; 164 165 return skb->len; 166 167 nla_put_failure: 168 nlmsg_trim(skb, tp); 169 return -1; 170 } 171 172 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = { 173 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) }, 174 [TCA_ACT_BPF_FD] = { .type = NLA_U32 }, 175 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING, 176 .len = ACT_BPF_NAME_LEN }, 177 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 }, 178 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY, 179 .len = sizeof(struct sock_filter) * BPF_MAXINSNS }, 180 }; 181 182 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg) 183 { 184 struct sock_filter *bpf_ops; 185 struct sock_fprog_kern fprog_tmp; 186 struct bpf_prog *fp; 187 u16 bpf_size, bpf_num_ops; 188 int ret; 189 190 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]); 191 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0) 192 return -EINVAL; 193 194 bpf_size = bpf_num_ops * sizeof(*bpf_ops); 195 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS])) 196 return -EINVAL; 197 198 bpf_ops = kzalloc(bpf_size, GFP_KERNEL); 199 if (bpf_ops == NULL) 200 return -ENOMEM; 201 202 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size); 203 204 fprog_tmp.len = bpf_num_ops; 205 fprog_tmp.filter = bpf_ops; 206 207 ret = bpf_prog_create(&fp, &fprog_tmp); 208 if (ret < 0) { 209 kfree(bpf_ops); 210 return ret; 211 } 212 213 cfg->bpf_ops = bpf_ops; 214 cfg->bpf_num_ops = bpf_num_ops; 215 cfg->filter = fp; 216 cfg->is_ebpf = false; 217 218 return 0; 219 } 220 221 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg) 222 { 223 struct bpf_prog *fp; 224 char *name = NULL; 225 u32 bpf_fd; 226 227 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]); 228 229 fp = bpf_prog_get_type(bpf_fd, BPF_PROG_TYPE_SCHED_ACT); 230 if (IS_ERR(fp)) 231 return PTR_ERR(fp); 232 233 if (tb[TCA_ACT_BPF_NAME]) { 234 name = nla_memdup(tb[TCA_ACT_BPF_NAME], GFP_KERNEL); 235 if (!name) { 236 bpf_prog_put(fp); 237 return -ENOMEM; 238 } 239 } 240 241 cfg->bpf_name = name; 242 cfg->filter = fp; 243 cfg->is_ebpf = true; 244 245 return 0; 246 } 247 248 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg) 249 { 250 if (cfg->is_ebpf) 251 bpf_prog_put(cfg->filter); 252 else 253 bpf_prog_destroy(cfg->filter); 254 255 kfree(cfg->bpf_ops); 256 kfree(cfg->bpf_name); 257 } 258 259 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog, 260 struct tcf_bpf_cfg *cfg) 261 { 262 cfg->is_ebpf = tcf_bpf_is_ebpf(prog); 263 /* updates to prog->filter are prevented, since it's called either 264 * with rtnl lock or during final cleanup in rcu callback 265 */ 266 cfg->filter = rcu_dereference_protected(prog->filter, 1); 267 268 cfg->bpf_ops = prog->bpf_ops; 269 cfg->bpf_name = prog->bpf_name; 270 } 271 272 static int tcf_bpf_init(struct net *net, struct nlattr *nla, 273 struct nlattr *est, struct tc_action **act, 274 int replace, int bind) 275 { 276 struct tc_action_net *tn = net_generic(net, bpf_net_id); 277 struct nlattr *tb[TCA_ACT_BPF_MAX + 1]; 278 struct tcf_bpf_cfg cfg, old; 279 struct tc_act_bpf *parm; 280 struct tcf_bpf *prog; 281 bool is_bpf, is_ebpf; 282 int ret, res = 0; 283 284 if (!nla) 285 return -EINVAL; 286 287 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy); 288 if (ret < 0) 289 return ret; 290 291 if (!tb[TCA_ACT_BPF_PARMS]) 292 return -EINVAL; 293 294 parm = nla_data(tb[TCA_ACT_BPF_PARMS]); 295 296 if (!tcf_hash_check(tn, parm->index, act, bind)) { 297 ret = tcf_hash_create(tn, parm->index, est, act, 298 &act_bpf_ops, bind, true); 299 if (ret < 0) 300 return ret; 301 302 res = ACT_P_CREATED; 303 } else { 304 /* Don't override defaults. */ 305 if (bind) 306 return 0; 307 308 tcf_hash_release(*act, bind); 309 if (!replace) 310 return -EEXIST; 311 } 312 313 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS]; 314 is_ebpf = tb[TCA_ACT_BPF_FD]; 315 316 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf)) { 317 ret = -EINVAL; 318 goto out; 319 } 320 321 memset(&cfg, 0, sizeof(cfg)); 322 323 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) : 324 tcf_bpf_init_from_efd(tb, &cfg); 325 if (ret < 0) 326 goto out; 327 328 prog = to_bpf(*act); 329 ASSERT_RTNL(); 330 331 if (res != ACT_P_CREATED) 332 tcf_bpf_prog_fill_cfg(prog, &old); 333 334 prog->bpf_ops = cfg.bpf_ops; 335 prog->bpf_name = cfg.bpf_name; 336 337 if (cfg.bpf_num_ops) 338 prog->bpf_num_ops = cfg.bpf_num_ops; 339 340 prog->tcf_action = parm->action; 341 rcu_assign_pointer(prog->filter, cfg.filter); 342 343 if (res == ACT_P_CREATED) { 344 tcf_hash_insert(tn, *act); 345 } else { 346 /* make sure the program being replaced is no longer executing */ 347 synchronize_rcu(); 348 tcf_bpf_cfg_cleanup(&old); 349 } 350 351 return res; 352 out: 353 if (res == ACT_P_CREATED) 354 tcf_hash_cleanup(*act, est); 355 356 return ret; 357 } 358 359 static void tcf_bpf_cleanup(struct tc_action *act, int bind) 360 { 361 struct tcf_bpf_cfg tmp; 362 363 tcf_bpf_prog_fill_cfg(to_bpf(act), &tmp); 364 tcf_bpf_cfg_cleanup(&tmp); 365 } 366 367 static int tcf_bpf_walker(struct net *net, struct sk_buff *skb, 368 struct netlink_callback *cb, int type, 369 const struct tc_action_ops *ops) 370 { 371 struct tc_action_net *tn = net_generic(net, bpf_net_id); 372 373 return tcf_generic_walker(tn, skb, cb, type, ops); 374 } 375 376 static int tcf_bpf_search(struct net *net, struct tc_action **a, u32 index) 377 { 378 struct tc_action_net *tn = net_generic(net, bpf_net_id); 379 380 return tcf_hash_search(tn, a, index); 381 } 382 383 static struct tc_action_ops act_bpf_ops __read_mostly = { 384 .kind = "bpf", 385 .type = TCA_ACT_BPF, 386 .owner = THIS_MODULE, 387 .act = tcf_bpf, 388 .dump = tcf_bpf_dump, 389 .cleanup = tcf_bpf_cleanup, 390 .init = tcf_bpf_init, 391 .walk = tcf_bpf_walker, 392 .lookup = tcf_bpf_search, 393 .size = sizeof(struct tcf_bpf), 394 }; 395 396 static __net_init int bpf_init_net(struct net *net) 397 { 398 struct tc_action_net *tn = net_generic(net, bpf_net_id); 399 400 return tc_action_net_init(tn, &act_bpf_ops, BPF_TAB_MASK); 401 } 402 403 static void __net_exit bpf_exit_net(struct net *net) 404 { 405 struct tc_action_net *tn = net_generic(net, bpf_net_id); 406 407 tc_action_net_exit(tn); 408 } 409 410 static struct pernet_operations bpf_net_ops = { 411 .init = bpf_init_net, 412 .exit = bpf_exit_net, 413 .id = &bpf_net_id, 414 .size = sizeof(struct tc_action_net), 415 }; 416 417 static int __init bpf_init_module(void) 418 { 419 return tcf_register_action(&act_bpf_ops, &bpf_net_ops); 420 } 421 422 static void __exit bpf_cleanup_module(void) 423 { 424 tcf_unregister_action(&act_bpf_ops, &bpf_net_ops); 425 } 426 427 module_init(bpf_init_module); 428 module_exit(bpf_cleanup_module); 429 430 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>"); 431 MODULE_DESCRIPTION("TC BPF based action"); 432 MODULE_LICENSE("GPL v2"); 433