1 /* 2 * net/sched/ematch.c Extended Match API 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Thomas Graf <tgraf@suug.ch> 10 * 11 * ========================================================================== 12 * 13 * An extended match (ematch) is a small classification tool not worth 14 * writing a full classifier for. Ematches can be interconnected to form 15 * a logic expression and get attached to classifiers to extend their 16 * functionatlity. 17 * 18 * The userspace part transforms the logic expressions into an array 19 * consisting of multiple sequences of interconnected ematches separated 20 * by markers. Precedence is implemented by a special ematch kind 21 * referencing a sequence beyond the marker of the current sequence 22 * causing the current position in the sequence to be pushed onto a stack 23 * to allow the current position to be overwritten by the position referenced 24 * in the special ematch. Matching continues in the new sequence until a 25 * marker is reached causing the position to be restored from the stack. 26 * 27 * Example: 28 * A AND (B1 OR B2) AND C AND D 29 * 30 * ------->-PUSH------- 31 * -->-- / -->-- \ -->-- 32 * / \ / / \ \ / \ 33 * +-------+-------+-------+-------+-------+--------+ 34 * | A AND | B AND | C AND | D END | B1 OR | B2 END | 35 * +-------+-------+-------+-------+-------+--------+ 36 * \ / 37 * --------<-POP--------- 38 * 39 * where B is a virtual ematch referencing to sequence starting with B1. 40 * 41 * ========================================================================== 42 * 43 * How to write an ematch in 60 seconds 44 * ------------------------------------ 45 * 46 * 1) Provide a matcher function: 47 * static int my_match(struct sk_buff *skb, struct tcf_ematch *m, 48 * struct tcf_pkt_info *info) 49 * { 50 * struct mydata *d = (struct mydata *) m->data; 51 * 52 * if (...matching goes here...) 53 * return 1; 54 * else 55 * return 0; 56 * } 57 * 58 * 2) Fill out a struct tcf_ematch_ops: 59 * static struct tcf_ematch_ops my_ops = { 60 * .kind = unique id, 61 * .datalen = sizeof(struct mydata), 62 * .match = my_match, 63 * .owner = THIS_MODULE, 64 * }; 65 * 66 * 3) Register/Unregister your ematch: 67 * static int __init init_my_ematch(void) 68 * { 69 * return tcf_em_register(&my_ops); 70 * } 71 * 72 * static void __exit exit_my_ematch(void) 73 * { 74 * return tcf_em_unregister(&my_ops); 75 * } 76 * 77 * module_init(init_my_ematch); 78 * module_exit(exit_my_ematch); 79 * 80 * 4) By now you should have two more seconds left, barely enough to 81 * open up a beer to watch the compilation going. 82 */ 83 84 #include <linux/module.h> 85 #include <linux/types.h> 86 #include <linux/kernel.h> 87 #include <linux/errno.h> 88 #include <linux/rtnetlink.h> 89 #include <linux/skbuff.h> 90 #include <net/pkt_cls.h> 91 92 static LIST_HEAD(ematch_ops); 93 static DEFINE_RWLOCK(ematch_mod_lock); 94 95 static inline struct tcf_ematch_ops * tcf_em_lookup(u16 kind) 96 { 97 struct tcf_ematch_ops *e = NULL; 98 99 read_lock(&ematch_mod_lock); 100 list_for_each_entry(e, &ematch_ops, link) { 101 if (kind == e->kind) { 102 if (!try_module_get(e->owner)) 103 e = NULL; 104 read_unlock(&ematch_mod_lock); 105 return e; 106 } 107 } 108 read_unlock(&ematch_mod_lock); 109 110 return NULL; 111 } 112 113 /** 114 * tcf_em_register - register an extended match 115 * 116 * @ops: ematch operations lookup table 117 * 118 * This function must be called by ematches to announce their presence. 119 * The given @ops must have kind set to a unique identifier and the 120 * callback match() must be implemented. All other callbacks are optional 121 * and a fallback implementation is used instead. 122 * 123 * Returns -EEXISTS if an ematch of the same kind has already registered. 124 */ 125 int tcf_em_register(struct tcf_ematch_ops *ops) 126 { 127 int err = -EEXIST; 128 struct tcf_ematch_ops *e; 129 130 if (ops->match == NULL) 131 return -EINVAL; 132 133 write_lock(&ematch_mod_lock); 134 list_for_each_entry(e, &ematch_ops, link) 135 if (ops->kind == e->kind) 136 goto errout; 137 138 list_add_tail(&ops->link, &ematch_ops); 139 err = 0; 140 errout: 141 write_unlock(&ematch_mod_lock); 142 return err; 143 } 144 145 /** 146 * tcf_em_unregister - unregster and extended match 147 * 148 * @ops: ematch operations lookup table 149 * 150 * This function must be called by ematches to announce their disappearance 151 * for examples when the module gets unloaded. The @ops parameter must be 152 * the same as the one used for registration. 153 * 154 * Returns -ENOENT if no matching ematch was found. 155 */ 156 int tcf_em_unregister(struct tcf_ematch_ops *ops) 157 { 158 int err = 0; 159 struct tcf_ematch_ops *e; 160 161 write_lock(&ematch_mod_lock); 162 list_for_each_entry(e, &ematch_ops, link) { 163 if (e == ops) { 164 list_del(&e->link); 165 goto out; 166 } 167 } 168 169 err = -ENOENT; 170 out: 171 write_unlock(&ematch_mod_lock); 172 return err; 173 } 174 175 static inline struct tcf_ematch * tcf_em_get_match(struct tcf_ematch_tree *tree, 176 int index) 177 { 178 return &tree->matches[index]; 179 } 180 181 182 static int tcf_em_validate(struct tcf_proto *tp, 183 struct tcf_ematch_tree_hdr *tree_hdr, 184 struct tcf_ematch *em, struct rtattr *rta, int idx) 185 { 186 int err = -EINVAL; 187 struct tcf_ematch_hdr *em_hdr = RTA_DATA(rta); 188 int data_len = RTA_PAYLOAD(rta) - sizeof(*em_hdr); 189 void *data = (void *) em_hdr + sizeof(*em_hdr); 190 191 if (!TCF_EM_REL_VALID(em_hdr->flags)) 192 goto errout; 193 194 if (em_hdr->kind == TCF_EM_CONTAINER) { 195 /* Special ematch called "container", carries an index 196 * referencing an external ematch sequence. */ 197 u32 ref; 198 199 if (data_len < sizeof(ref)) 200 goto errout; 201 ref = *(u32 *) data; 202 203 if (ref >= tree_hdr->nmatches) 204 goto errout; 205 206 /* We do not allow backward jumps to avoid loops and jumps 207 * to our own position are of course illegal. */ 208 if (ref <= idx) 209 goto errout; 210 211 212 em->data = ref; 213 } else { 214 /* Note: This lookup will increase the module refcnt 215 * of the ematch module referenced. In case of a failure, 216 * a destroy function is called by the underlying layer 217 * which automatically releases the reference again, therefore 218 * the module MUST not be given back under any circumstances 219 * here. Be aware, the destroy function assumes that the 220 * module is held if the ops field is non zero. */ 221 em->ops = tcf_em_lookup(em_hdr->kind); 222 223 if (em->ops == NULL) { 224 err = -ENOENT; 225 goto errout; 226 } 227 228 /* ematch module provides expected length of data, so we 229 * can do a basic sanity check. */ 230 if (em->ops->datalen && data_len < em->ops->datalen) 231 goto errout; 232 233 if (em->ops->change) { 234 err = em->ops->change(tp, data, data_len, em); 235 if (err < 0) 236 goto errout; 237 } else if (data_len > 0) { 238 /* ematch module doesn't provide an own change 239 * procedure and expects us to allocate and copy 240 * the ematch data. 241 * 242 * TCF_EM_SIMPLE may be specified stating that the 243 * data only consists of a u32 integer and the module 244 * does not expected a memory reference but rather 245 * the value carried. */ 246 if (em_hdr->flags & TCF_EM_SIMPLE) { 247 if (data_len < sizeof(u32)) 248 goto errout; 249 em->data = *(u32 *) data; 250 } else { 251 void *v = kmemdup(data, data_len, GFP_KERNEL); 252 if (v == NULL) { 253 err = -ENOBUFS; 254 goto errout; 255 } 256 em->data = (unsigned long) v; 257 } 258 } 259 } 260 261 em->matchid = em_hdr->matchid; 262 em->flags = em_hdr->flags; 263 em->datalen = data_len; 264 265 err = 0; 266 errout: 267 return err; 268 } 269 270 /** 271 * tcf_em_tree_validate - validate ematch config TLV and build ematch tree 272 * 273 * @tp: classifier kind handle 274 * @rta: ematch tree configuration TLV 275 * @tree: destination ematch tree variable to store the resulting 276 * ematch tree. 277 * 278 * This function validates the given configuration TLV @rta and builds an 279 * ematch tree in @tree. The resulting tree must later be copied into 280 * the private classifier data using tcf_em_tree_change(). You MUST NOT 281 * provide the ematch tree variable of the private classifier data directly, 282 * the changes would not be locked properly. 283 * 284 * Returns a negative error code if the configuration TLV contains errors. 285 */ 286 int tcf_em_tree_validate(struct tcf_proto *tp, struct rtattr *rta, 287 struct tcf_ematch_tree *tree) 288 { 289 int idx, list_len, matches_len, err = -EINVAL; 290 struct rtattr *tb[TCA_EMATCH_TREE_MAX]; 291 struct rtattr *rt_match, *rt_hdr, *rt_list; 292 struct tcf_ematch_tree_hdr *tree_hdr; 293 struct tcf_ematch *em; 294 295 if (!rta) { 296 memset(tree, 0, sizeof(*tree)); 297 return 0; 298 } 299 300 if (rtattr_parse_nested(tb, TCA_EMATCH_TREE_MAX, rta) < 0) 301 goto errout; 302 303 rt_hdr = tb[TCA_EMATCH_TREE_HDR-1]; 304 rt_list = tb[TCA_EMATCH_TREE_LIST-1]; 305 306 if (rt_hdr == NULL || rt_list == NULL) 307 goto errout; 308 309 if (RTA_PAYLOAD(rt_hdr) < sizeof(*tree_hdr) || 310 RTA_PAYLOAD(rt_list) < sizeof(*rt_match)) 311 goto errout; 312 313 tree_hdr = RTA_DATA(rt_hdr); 314 memcpy(&tree->hdr, tree_hdr, sizeof(*tree_hdr)); 315 316 rt_match = RTA_DATA(rt_list); 317 list_len = RTA_PAYLOAD(rt_list); 318 matches_len = tree_hdr->nmatches * sizeof(*em); 319 320 tree->matches = kzalloc(matches_len, GFP_KERNEL); 321 if (tree->matches == NULL) 322 goto errout; 323 324 /* We do not use rtattr_parse_nested here because the maximum 325 * number of attributes is unknown. This saves us the allocation 326 * for a tb buffer which would serve no purpose at all. 327 * 328 * The array of rt attributes is parsed in the order as they are 329 * provided, their type must be incremental from 1 to n. Even 330 * if it does not serve any real purpose, a failure of sticking 331 * to this policy will result in parsing failure. */ 332 for (idx = 0; RTA_OK(rt_match, list_len); idx++) { 333 err = -EINVAL; 334 335 if (rt_match->rta_type != (idx + 1)) 336 goto errout_abort; 337 338 if (idx >= tree_hdr->nmatches) 339 goto errout_abort; 340 341 if (RTA_PAYLOAD(rt_match) < sizeof(struct tcf_ematch_hdr)) 342 goto errout_abort; 343 344 em = tcf_em_get_match(tree, idx); 345 346 err = tcf_em_validate(tp, tree_hdr, em, rt_match, idx); 347 if (err < 0) 348 goto errout_abort; 349 350 rt_match = RTA_NEXT(rt_match, list_len); 351 } 352 353 /* Check if the number of matches provided by userspace actually 354 * complies with the array of matches. The number was used for 355 * the validation of references and a mismatch could lead to 356 * undefined references during the matching process. */ 357 if (idx != tree_hdr->nmatches) { 358 err = -EINVAL; 359 goto errout_abort; 360 } 361 362 err = 0; 363 errout: 364 return err; 365 366 errout_abort: 367 tcf_em_tree_destroy(tp, tree); 368 return err; 369 } 370 371 /** 372 * tcf_em_tree_destroy - destroy an ematch tree 373 * 374 * @tp: classifier kind handle 375 * @tree: ematch tree to be deleted 376 * 377 * This functions destroys an ematch tree previously created by 378 * tcf_em_tree_validate()/tcf_em_tree_change(). You must ensure that 379 * the ematch tree is not in use before calling this function. 380 */ 381 void tcf_em_tree_destroy(struct tcf_proto *tp, struct tcf_ematch_tree *tree) 382 { 383 int i; 384 385 if (tree->matches == NULL) 386 return; 387 388 for (i = 0; i < tree->hdr.nmatches; i++) { 389 struct tcf_ematch *em = tcf_em_get_match(tree, i); 390 391 if (em->ops) { 392 if (em->ops->destroy) 393 em->ops->destroy(tp, em); 394 else if (!tcf_em_is_simple(em) && em->data) 395 kfree((void *) em->data); 396 module_put(em->ops->owner); 397 } 398 } 399 400 tree->hdr.nmatches = 0; 401 kfree(tree->matches); 402 } 403 404 /** 405 * tcf_em_tree_dump - dump ematch tree into a rtnl message 406 * 407 * @skb: skb holding the rtnl message 408 * @t: ematch tree to be dumped 409 * @tlv: TLV type to be used to encapsulate the tree 410 * 411 * This function dumps a ematch tree into a rtnl message. It is valid to 412 * call this function while the ematch tree is in use. 413 * 414 * Returns -1 if the skb tailroom is insufficient. 415 */ 416 int tcf_em_tree_dump(struct sk_buff *skb, struct tcf_ematch_tree *tree, int tlv) 417 { 418 int i; 419 u8 *tail; 420 struct rtattr *top_start = (struct rtattr *)skb_tail_pointer(skb); 421 struct rtattr *list_start; 422 423 RTA_PUT(skb, tlv, 0, NULL); 424 RTA_PUT(skb, TCA_EMATCH_TREE_HDR, sizeof(tree->hdr), &tree->hdr); 425 426 list_start = (struct rtattr *)skb_tail_pointer(skb); 427 RTA_PUT(skb, TCA_EMATCH_TREE_LIST, 0, NULL); 428 429 tail = skb_tail_pointer(skb); 430 for (i = 0; i < tree->hdr.nmatches; i++) { 431 struct rtattr *match_start = (struct rtattr *)tail; 432 struct tcf_ematch *em = tcf_em_get_match(tree, i); 433 struct tcf_ematch_hdr em_hdr = { 434 .kind = em->ops ? em->ops->kind : TCF_EM_CONTAINER, 435 .matchid = em->matchid, 436 .flags = em->flags 437 }; 438 439 RTA_PUT(skb, i+1, sizeof(em_hdr), &em_hdr); 440 441 if (em->ops && em->ops->dump) { 442 if (em->ops->dump(skb, em) < 0) 443 goto rtattr_failure; 444 } else if (tcf_em_is_container(em) || tcf_em_is_simple(em)) { 445 u32 u = em->data; 446 RTA_PUT_NOHDR(skb, sizeof(u), &u); 447 } else if (em->datalen > 0) 448 RTA_PUT_NOHDR(skb, em->datalen, (void *) em->data); 449 450 tail = skb_tail_pointer(skb); 451 match_start->rta_len = tail - (u8 *)match_start; 452 } 453 454 list_start->rta_len = tail - (u8 *)list_start; 455 top_start->rta_len = tail - (u8 *)top_start; 456 457 return 0; 458 459 rtattr_failure: 460 return -1; 461 } 462 463 static inline int tcf_em_match(struct sk_buff *skb, struct tcf_ematch *em, 464 struct tcf_pkt_info *info) 465 { 466 int r = em->ops->match(skb, em, info); 467 return tcf_em_is_inverted(em) ? !r : r; 468 } 469 470 /* Do not use this function directly, use tcf_em_tree_match instead */ 471 int __tcf_em_tree_match(struct sk_buff *skb, struct tcf_ematch_tree *tree, 472 struct tcf_pkt_info *info) 473 { 474 int stackp = 0, match_idx = 0, res = 0; 475 struct tcf_ematch *cur_match; 476 int stack[CONFIG_NET_EMATCH_STACK]; 477 478 proceed: 479 while (match_idx < tree->hdr.nmatches) { 480 cur_match = tcf_em_get_match(tree, match_idx); 481 482 if (tcf_em_is_container(cur_match)) { 483 if (unlikely(stackp >= CONFIG_NET_EMATCH_STACK)) 484 goto stack_overflow; 485 486 stack[stackp++] = match_idx; 487 match_idx = cur_match->data; 488 goto proceed; 489 } 490 491 res = tcf_em_match(skb, cur_match, info); 492 493 if (tcf_em_early_end(cur_match, res)) 494 break; 495 496 match_idx++; 497 } 498 499 pop_stack: 500 if (stackp > 0) { 501 match_idx = stack[--stackp]; 502 cur_match = tcf_em_get_match(tree, match_idx); 503 504 if (tcf_em_early_end(cur_match, res)) 505 goto pop_stack; 506 else { 507 match_idx++; 508 goto proceed; 509 } 510 } 511 512 return res; 513 514 stack_overflow: 515 if (net_ratelimit()) 516 printk("Local stack overflow, increase NET_EMATCH_STACK\n"); 517 return -1; 518 } 519 520 EXPORT_SYMBOL(tcf_em_register); 521 EXPORT_SYMBOL(tcf_em_unregister); 522 EXPORT_SYMBOL(tcf_em_tree_validate); 523 EXPORT_SYMBOL(tcf_em_tree_destroy); 524 EXPORT_SYMBOL(tcf_em_tree_dump); 525 EXPORT_SYMBOL(__tcf_em_tree_match); 526