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 u32 bpf_fd; 32 u16 bpf_num_ops; 33 bool is_ebpf; 34 }; 35 36 static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act, 37 struct tcf_result *res) 38 { 39 struct tcf_bpf *prog = act->priv; 40 struct bpf_prog *filter; 41 int action, filter_res; 42 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS; 43 44 if (unlikely(!skb_mac_header_was_set(skb))) 45 return TC_ACT_UNSPEC; 46 47 tcf_lastuse_update(&prog->tcf_tm); 48 bstats_cpu_update(this_cpu_ptr(prog->common.cpu_bstats), skb); 49 50 rcu_read_lock(); 51 filter = rcu_dereference(prog->filter); 52 if (at_ingress) { 53 __skb_push(skb, skb->mac_len); 54 filter_res = BPF_PROG_RUN(filter, skb); 55 __skb_pull(skb, skb->mac_len); 56 } else { 57 filter_res = BPF_PROG_RUN(filter, skb); 58 } 59 rcu_read_unlock(); 60 61 /* A BPF program may overwrite the default action opcode. 62 * Similarly as in cls_bpf, if filter_res == -1 we use the 63 * default action specified from tc. 64 * 65 * In case a different well-known TC_ACT opcode has been 66 * returned, it will overwrite the default one. 67 * 68 * For everything else that is unkown, TC_ACT_UNSPEC is 69 * returned. 70 */ 71 switch (filter_res) { 72 case TC_ACT_PIPE: 73 case TC_ACT_RECLASSIFY: 74 case TC_ACT_OK: 75 action = filter_res; 76 break; 77 case TC_ACT_SHOT: 78 action = filter_res; 79 qstats_drop_inc(this_cpu_ptr(prog->common.cpu_qstats)); 80 break; 81 case TC_ACT_UNSPEC: 82 action = prog->tcf_action; 83 break; 84 default: 85 action = TC_ACT_UNSPEC; 86 break; 87 } 88 89 return action; 90 } 91 92 static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog) 93 { 94 return !prog->bpf_ops; 95 } 96 97 static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog, 98 struct sk_buff *skb) 99 { 100 struct nlattr *nla; 101 102 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops)) 103 return -EMSGSIZE; 104 105 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops * 106 sizeof(struct sock_filter)); 107 if (nla == NULL) 108 return -EMSGSIZE; 109 110 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla)); 111 112 return 0; 113 } 114 115 static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog, 116 struct sk_buff *skb) 117 { 118 if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd)) 119 return -EMSGSIZE; 120 121 if (prog->bpf_name && 122 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name)) 123 return -EMSGSIZE; 124 125 return 0; 126 } 127 128 static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act, 129 int bind, int ref) 130 { 131 unsigned char *tp = skb_tail_pointer(skb); 132 struct tcf_bpf *prog = act->priv; 133 struct tc_act_bpf opt = { 134 .index = prog->tcf_index, 135 .refcnt = prog->tcf_refcnt - ref, 136 .bindcnt = prog->tcf_bindcnt - bind, 137 .action = prog->tcf_action, 138 }; 139 struct tcf_t tm; 140 int ret; 141 142 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt)) 143 goto nla_put_failure; 144 145 if (tcf_bpf_is_ebpf(prog)) 146 ret = tcf_bpf_dump_ebpf_info(prog, skb); 147 else 148 ret = tcf_bpf_dump_bpf_info(prog, skb); 149 if (ret) 150 goto nla_put_failure; 151 152 tm.install = jiffies_to_clock_t(jiffies - prog->tcf_tm.install); 153 tm.lastuse = jiffies_to_clock_t(jiffies - prog->tcf_tm.lastuse); 154 tm.expires = jiffies_to_clock_t(prog->tcf_tm.expires); 155 156 if (nla_put(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm)) 157 goto nla_put_failure; 158 159 return skb->len; 160 161 nla_put_failure: 162 nlmsg_trim(skb, tp); 163 return -1; 164 } 165 166 static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = { 167 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) }, 168 [TCA_ACT_BPF_FD] = { .type = NLA_U32 }, 169 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING, .len = ACT_BPF_NAME_LEN }, 170 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 }, 171 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY, 172 .len = sizeof(struct sock_filter) * BPF_MAXINSNS }, 173 }; 174 175 static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg) 176 { 177 struct sock_filter *bpf_ops; 178 struct sock_fprog_kern fprog_tmp; 179 struct bpf_prog *fp; 180 u16 bpf_size, bpf_num_ops; 181 int ret; 182 183 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]); 184 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0) 185 return -EINVAL; 186 187 bpf_size = bpf_num_ops * sizeof(*bpf_ops); 188 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS])) 189 return -EINVAL; 190 191 bpf_ops = kzalloc(bpf_size, GFP_KERNEL); 192 if (bpf_ops == NULL) 193 return -ENOMEM; 194 195 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size); 196 197 fprog_tmp.len = bpf_num_ops; 198 fprog_tmp.filter = bpf_ops; 199 200 ret = bpf_prog_create(&fp, &fprog_tmp); 201 if (ret < 0) { 202 kfree(bpf_ops); 203 return ret; 204 } 205 206 cfg->bpf_ops = bpf_ops; 207 cfg->bpf_num_ops = bpf_num_ops; 208 cfg->filter = fp; 209 cfg->is_ebpf = false; 210 211 return 0; 212 } 213 214 static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg) 215 { 216 struct bpf_prog *fp; 217 char *name = NULL; 218 u32 bpf_fd; 219 220 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]); 221 222 fp = bpf_prog_get(bpf_fd); 223 if (IS_ERR(fp)) 224 return PTR_ERR(fp); 225 226 if (fp->type != BPF_PROG_TYPE_SCHED_ACT) { 227 bpf_prog_put(fp); 228 return -EINVAL; 229 } 230 231 if (tb[TCA_ACT_BPF_NAME]) { 232 name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]), 233 nla_len(tb[TCA_ACT_BPF_NAME]), 234 GFP_KERNEL); 235 if (!name) { 236 bpf_prog_put(fp); 237 return -ENOMEM; 238 } 239 } 240 241 cfg->bpf_fd = bpf_fd; 242 cfg->bpf_name = name; 243 cfg->filter = fp; 244 cfg->is_ebpf = true; 245 246 return 0; 247 } 248 249 static void tcf_bpf_cfg_cleanup(const struct tcf_bpf_cfg *cfg) 250 { 251 if (cfg->is_ebpf) 252 bpf_prog_put(cfg->filter); 253 else 254 bpf_prog_destroy(cfg->filter); 255 256 kfree(cfg->bpf_ops); 257 kfree(cfg->bpf_name); 258 } 259 260 static void tcf_bpf_prog_fill_cfg(const struct tcf_bpf *prog, 261 struct tcf_bpf_cfg *cfg) 262 { 263 cfg->is_ebpf = tcf_bpf_is_ebpf(prog); 264 /* updates to prog->filter are prevented, since it's called either 265 * with rtnl lock or during final cleanup in rcu callback 266 */ 267 cfg->filter = rcu_dereference_protected(prog->filter, 1); 268 269 cfg->bpf_ops = prog->bpf_ops; 270 cfg->bpf_name = prog->bpf_name; 271 } 272 273 static int tcf_bpf_init(struct net *net, struct nlattr *nla, 274 struct nlattr *est, struct tc_action *act, 275 int replace, int bind) 276 { 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(parm->index, act, bind)) { 297 ret = tcf_hash_create(parm->index, est, act, 298 sizeof(*prog), 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 if (cfg.bpf_fd) 340 prog->bpf_fd = cfg.bpf_fd; 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(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(act->priv, &tmp); 366 tcf_bpf_cfg_cleanup(&tmp); 367 } 368 369 static struct tc_action_ops act_bpf_ops __read_mostly = { 370 .kind = "bpf", 371 .type = TCA_ACT_BPF, 372 .owner = THIS_MODULE, 373 .act = tcf_bpf, 374 .dump = tcf_bpf_dump, 375 .cleanup = tcf_bpf_cleanup, 376 .init = tcf_bpf_init, 377 }; 378 379 static int __init bpf_init_module(void) 380 { 381 return tcf_register_action(&act_bpf_ops, BPF_TAB_MASK); 382 } 383 384 static void __exit bpf_cleanup_module(void) 385 { 386 tcf_unregister_action(&act_bpf_ops); 387 } 388 389 module_init(bpf_init_module); 390 module_exit(bpf_cleanup_module); 391 392 MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>"); 393 MODULE_DESCRIPTION("TC BPF based action"); 394 MODULE_LICENSE("GPL v2"); 395