1 /* 2 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 * 25 * $FreeBSD: src/sys/netinet/ip_fw2.c,v 1.6.2.12 2003/04/08 10:42:32 maxim Exp $ 26 */ 27 28 /* 29 * Implement IP packet firewall (new version) 30 */ 31 32 #include "opt_ipfw.h" 33 #include "opt_inet.h" 34 #ifndef INET 35 #error IPFIREWALL requires INET. 36 #endif /* INET */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/malloc.h> 41 #include <sys/mbuf.h> 42 #include <sys/kernel.h> 43 #include <sys/proc.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/sysctl.h> 47 #include <sys/syslog.h> 48 #include <sys/ucred.h> 49 #include <sys/in_cksum.h> 50 #include <sys/lock.h> 51 52 #include <net/if.h> 53 #include <net/route.h> 54 #include <net/pfil.h> 55 #include <net/dummynet/ip_dummynet.h> 56 57 #include <sys/thread2.h> 58 #include <sys/mplock2.h> 59 #include <net/netmsg2.h> 60 61 #include <netinet/in.h> 62 #include <netinet/in_systm.h> 63 #include <netinet/in_var.h> 64 #include <netinet/in_pcb.h> 65 #include <netinet/ip.h> 66 #include <netinet/ip_var.h> 67 #include <netinet/ip_icmp.h> 68 #include <netinet/tcp.h> 69 #include <netinet/tcp_timer.h> 70 #include <netinet/tcp_var.h> 71 #include <netinet/tcpip.h> 72 #include <netinet/udp.h> 73 #include <netinet/udp_var.h> 74 #include <netinet/ip_divert.h> 75 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */ 76 77 #include <net/ipfw/ip_fw2.h> 78 79 #ifdef IPFIREWALL_DEBUG 80 #define DPRINTF(fmt, ...) \ 81 do { \ 82 if (fw_debug > 0) \ 83 kprintf(fmt, __VA_ARGS__); \ 84 } while (0) 85 #else 86 #define DPRINTF(fmt, ...) ((void)0) 87 #endif 88 89 /* 90 * Description about per-CPU rule duplication: 91 * 92 * Module loading/unloading and all ioctl operations are serialized 93 * by netisr0, so we don't have any ordering or locking problems. 94 * 95 * Following graph shows how operation on per-CPU rule list is 96 * performed [2 CPU case]: 97 * 98 * CPU0 CPU1 99 * 100 * netisr0 <------------------------------------+ 101 * domsg | 102 * | | 103 * | netmsg | 104 * | | 105 * V | 106 * ifnet0 | 107 * : | netmsg 108 * :(delete/add...) | 109 * : | 110 * : netmsg | 111 * forwardmsg---------->ifnet1 | 112 * : | 113 * :(delete/add...) | 114 * : | 115 * : | 116 * replymsg--------------+ 117 * 118 * 119 * 120 * 121 * Rules which will not create states (dyn rules) [2 CPU case] 122 * 123 * CPU0 CPU1 124 * layer3_chain layer3_chain 125 * | | 126 * V V 127 * +-------+ sibling +-------+ sibling 128 * | rule1 |--------->| rule1 |--------->NULL 129 * +-------+ +-------+ 130 * | | 131 * |next |next 132 * V V 133 * +-------+ sibling +-------+ sibling 134 * | rule2 |--------->| rule2 |--------->NULL 135 * +-------+ +-------+ 136 * 137 * ip_fw.sibling: 138 * 1) Ease statistics calculation during IP_FW_GET. We only need to 139 * iterate layer3_chain on CPU0; the current rule's duplication on 140 * the other CPUs could safely be read-only accessed by using 141 * ip_fw.sibling 142 * 2) Accelerate rule insertion and deletion, e.g. rule insertion: 143 * a) In netisr0 (on CPU0) rule3 is determined to be inserted between 144 * rule1 and rule2. To make this decision we need to iterate the 145 * layer3_chain on CPU0. The netmsg, which is used to insert the 146 * rule, will contain rule1 on CPU0 as prev_rule and rule2 on CPU0 147 * as next_rule 148 * b) After the insertion on CPU0 is done, we will move on to CPU1. 149 * But instead of relocating the rule3's position on CPU1 by 150 * iterating the layer3_chain on CPU1, we set the netmsg's prev_rule 151 * to rule1->sibling and next_rule to rule2->sibling before the 152 * netmsg is forwarded to CPU1 from CPU0 153 * 154 * 155 * 156 * Rules which will create states (dyn rules) [2 CPU case] 157 * (unnecessary parts are omitted; they are same as in the previous figure) 158 * 159 * CPU0 CPU1 160 * 161 * +-------+ +-------+ 162 * | rule1 | | rule1 | 163 * +-------+ +-------+ 164 * ^ | | ^ 165 * | |stub stub| | 166 * | | | | 167 * | +----+ +----+ | 168 * | | | | 169 * | V V | 170 * | +--------------------+ | 171 * | | rule_stub | | 172 * | | (read-only shared) | | 173 * | | | | 174 * | | back pointer array | | 175 * | | (indexed by cpuid) | | 176 * | | | | 177 * +----|---------[0] | | 178 * | [1]--------|----+ 179 * | | 180 * +--------------------+ 181 * ^ ^ 182 * | | 183 * ........|............|............ 184 * : | | : 185 * : |stub |stub : 186 * : | | : 187 * : +---------+ +---------+ : 188 * : | state1a | | state1b | .... : 189 * : +---------+ +---------+ : 190 * : : 191 * : states table : 192 * : (shared) : 193 * : (protected by dyn_lock) : 194 * .................................. 195 * 196 * [state1a and state1b are states created by rule1] 197 * 198 * ip_fw_stub: 199 * This structure is introduced so that shared (locked) state table could 200 * work with per-CPU (duplicated) static rules. It mainly bridges states 201 * and static rules and serves as static rule's place holder (a read-only 202 * shared part of duplicated rules) from states point of view. 203 * 204 * IPFW_RULE_F_STATE (only for rules which create states): 205 * o During rule installation, this flag is turned on after rule's 206 * duplications reach all CPUs, to avoid at least following race: 207 * 1) rule1 is duplicated on CPU0 and is not duplicated on CPU1 yet 208 * 2) rule1 creates state1 209 * 3) state1 is located on CPU1 by check-state 210 * But rule1 is not duplicated on CPU1 yet 211 * o During rule deletion, this flag is turned off before deleting states 212 * created by the rule and before deleting the rule itself, so no 213 * more states will be created by the to-be-deleted rule even when its 214 * duplication on certain CPUs are not eliminated yet. 215 */ 216 217 #define IPFW_AUTOINC_STEP_MIN 1 218 #define IPFW_AUTOINC_STEP_MAX 1000 219 #define IPFW_AUTOINC_STEP_DEF 100 220 221 #define IPFW_DEFAULT_RULE 65535 /* rulenum for the default rule */ 222 #define IPFW_DEFAULT_SET 31 /* set number for the default rule */ 223 224 struct netmsg_ipfw { 225 struct netmsg_base base; 226 const struct ipfw_ioc_rule *ioc_rule; 227 struct ip_fw *next_rule; 228 struct ip_fw *prev_rule; 229 struct ip_fw *sibling; 230 struct ip_fw_stub *stub; 231 }; 232 233 struct netmsg_del { 234 struct netmsg_base base; 235 struct ip_fw *start_rule; 236 struct ip_fw *prev_rule; 237 uint16_t rulenum; 238 uint8_t from_set; 239 uint8_t to_set; 240 }; 241 242 struct netmsg_zent { 243 struct netmsg_base base; 244 struct ip_fw *start_rule; 245 uint16_t rulenum; 246 uint16_t log_only; 247 }; 248 249 struct ipfw_context { 250 struct ip_fw *ipfw_layer3_chain; /* list of rules for layer3 */ 251 struct ip_fw *ipfw_default_rule; /* default rule */ 252 uint64_t ipfw_norule_counter; /* counter for ipfw_log(NULL) */ 253 254 /* 255 * ipfw_set_disable contains one bit per set value (0..31). 256 * If the bit is set, all rules with the corresponding set 257 * are disabled. Set IPDW_DEFAULT_SET is reserved for the 258 * default rule and CANNOT be disabled. 259 */ 260 uint32_t ipfw_set_disable; 261 uint32_t ipfw_gen; /* generation of rule list */ 262 }; 263 264 static struct ipfw_context *ipfw_ctx[MAXCPU]; 265 266 #ifdef KLD_MODULE 267 /* 268 * Module can not be unloaded, if there are references to 269 * certains rules of ipfw(4), e.g. dummynet(4) 270 */ 271 static int ipfw_refcnt; 272 #endif 273 274 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's"); 275 276 /* 277 * Following two global variables are accessed and 278 * updated only on CPU0 279 */ 280 static uint32_t static_count; /* # of static rules */ 281 static uint32_t static_ioc_len; /* bytes of static rules */ 282 283 /* 284 * If 1, then ipfw static rules are being flushed, 285 * ipfw_chk() will skip to the default rule. 286 */ 287 static int ipfw_flushing; 288 289 static int fw_verbose; 290 static int verbose_limit; 291 292 static int fw_debug; 293 static int autoinc_step = IPFW_AUTOINC_STEP_DEF; 294 295 static int ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS); 296 static int ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS); 297 static int ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS); 298 static int ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS); 299 static int ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS); 300 301 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall"); 302 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW, 303 &fw_enable, 0, ipfw_sysctl_enable, "I", "Enable ipfw"); 304 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLTYPE_INT | CTLFLAG_RW, 305 &autoinc_step, 0, ipfw_sysctl_autoinc_step, "I", 306 "Rule number autincrement step"); 307 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO,one_pass,CTLFLAG_RW, 308 &fw_one_pass, 0, 309 "Only do a single pass through ipfw when using dummynet(4)"); 310 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW, 311 &fw_debug, 0, "Enable printing of debug ip_fw statements"); 312 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_RW, 313 &fw_verbose, 0, "Log matches to ipfw rules"); 314 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW, 315 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged"); 316 317 /* 318 * Description of dynamic rules. 319 * 320 * Dynamic rules are stored in lists accessed through a hash table 321 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can 322 * be modified through the sysctl variable dyn_buckets which is 323 * updated when the table becomes empty. 324 * 325 * XXX currently there is only one list, ipfw_dyn. 326 * 327 * When a packet is received, its address fields are first masked 328 * with the mask defined for the rule, then hashed, then matched 329 * against the entries in the corresponding list. 330 * Dynamic rules can be used for different purposes: 331 * + stateful rules; 332 * + enforcing limits on the number of sessions; 333 * + in-kernel NAT (not implemented yet) 334 * 335 * The lifetime of dynamic rules is regulated by dyn_*_lifetime, 336 * measured in seconds and depending on the flags. 337 * 338 * The total number of dynamic rules is stored in dyn_count. 339 * The max number of dynamic rules is dyn_max. When we reach 340 * the maximum number of rules we do not create anymore. This is 341 * done to avoid consuming too much memory, but also too much 342 * time when searching on each packet (ideally, we should try instead 343 * to put a limit on the length of the list on each bucket...). 344 * 345 * Each dynamic rule holds a pointer to the parent ipfw rule so 346 * we know what action to perform. Dynamic rules are removed when 347 * the parent rule is deleted. XXX we should make them survive. 348 * 349 * There are some limitations with dynamic rules -- we do not 350 * obey the 'randomized match', and we do not do multiple 351 * passes through the firewall. XXX check the latter!!! 352 * 353 * NOTE about the SHARED LOCKMGR LOCK during dynamic rule looking up: 354 * Only TCP state transition will change dynamic rule's state and ack 355 * sequences, while all packets of one TCP connection only goes through 356 * one TCP thread, so it is safe to use shared lockmgr lock during dynamic 357 * rule looking up. The keep alive callout uses exclusive lockmgr lock 358 * when it tries to find suitable dynamic rules to send keep alive, so 359 * it will not see half updated state and ack sequences. Though the expire 360 * field updating looks racy for other protocols, the resolution (second) 361 * of expire field makes this kind of race harmless. 362 * XXX statistics' updating is _not_ MPsafe!!! 363 * XXX once UDP output path is fixed, we could use lockless dynamic rule 364 * hash table 365 */ 366 static ipfw_dyn_rule **ipfw_dyn_v = NULL; 367 static uint32_t dyn_buckets = 256; /* must be power of 2 */ 368 static uint32_t curr_dyn_buckets = 256; /* must be power of 2 */ 369 static uint32_t dyn_buckets_gen; /* generation of dyn buckets array */ 370 static struct lock dyn_lock; /* dynamic rules' hash table lock */ 371 372 static struct netmsg_base ipfw_timeout_netmsg; /* schedule ipfw timeout */ 373 static struct callout ipfw_timeout_h; 374 375 /* 376 * Timeouts for various events in handing dynamic rules. 377 */ 378 static uint32_t dyn_ack_lifetime = 300; 379 static uint32_t dyn_syn_lifetime = 20; 380 static uint32_t dyn_fin_lifetime = 1; 381 static uint32_t dyn_rst_lifetime = 1; 382 static uint32_t dyn_udp_lifetime = 10; 383 static uint32_t dyn_short_lifetime = 5; 384 385 /* 386 * Keepalives are sent if dyn_keepalive is set. They are sent every 387 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval 388 * seconds of lifetime of a rule. 389 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower 390 * than dyn_keepalive_period. 391 */ 392 393 static uint32_t dyn_keepalive_interval = 20; 394 static uint32_t dyn_keepalive_period = 5; 395 static uint32_t dyn_keepalive = 1; /* do send keepalives */ 396 397 static uint32_t dyn_count; /* # of dynamic rules */ 398 static uint32_t dyn_max = 4096; /* max # of dynamic rules */ 399 400 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLTYPE_INT | CTLFLAG_RW, 401 &dyn_buckets, 0, ipfw_sysctl_dyn_buckets, "I", "Number of dyn. buckets"); 402 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD, 403 &curr_dyn_buckets, 0, "Current Number of dyn. buckets"); 404 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD, 405 &dyn_count, 0, "Number of dyn. rules"); 406 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW, 407 &dyn_max, 0, "Max number of dyn. rules"); 408 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD, 409 &static_count, 0, "Number of static rules"); 410 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW, 411 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks"); 412 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW, 413 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn"); 414 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, 415 CTLTYPE_INT | CTLFLAG_RW, &dyn_fin_lifetime, 0, ipfw_sysctl_dyn_fin, "I", 416 "Lifetime of dyn. rules for fin"); 417 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, 418 CTLTYPE_INT | CTLFLAG_RW, &dyn_rst_lifetime, 0, ipfw_sysctl_dyn_rst, "I", 419 "Lifetime of dyn. rules for rst"); 420 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW, 421 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP"); 422 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW, 423 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations"); 424 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW, 425 &dyn_keepalive, 0, "Enable keepalives for dyn. rules"); 426 427 static ip_fw_chk_t ipfw_chk; 428 static void ipfw_tick(void *); 429 430 static __inline int 431 ipfw_free_rule(struct ip_fw *rule) 432 { 433 KASSERT(rule->cpuid == mycpuid, ("rule freed on cpu%d", mycpuid)); 434 KASSERT(rule->refcnt > 0, ("invalid refcnt %u", rule->refcnt)); 435 rule->refcnt--; 436 if (rule->refcnt == 0) { 437 kfree(rule, M_IPFW); 438 return 1; 439 } 440 return 0; 441 } 442 443 static void 444 ipfw_unref_rule(void *priv) 445 { 446 ipfw_free_rule(priv); 447 #ifdef KLD_MODULE 448 atomic_subtract_int(&ipfw_refcnt, 1); 449 #endif 450 } 451 452 static __inline void 453 ipfw_ref_rule(struct ip_fw *rule) 454 { 455 KASSERT(rule->cpuid == mycpuid, ("rule used on cpu%d", mycpuid)); 456 #ifdef KLD_MODULE 457 atomic_add_int(&ipfw_refcnt, 1); 458 #endif 459 rule->refcnt++; 460 } 461 462 /* 463 * This macro maps an ip pointer into a layer3 header pointer of type T 464 */ 465 #define L3HDR(T, ip) ((T *)((uint32_t *)(ip) + (ip)->ip_hl)) 466 467 static __inline int 468 icmptype_match(struct ip *ip, ipfw_insn_u32 *cmd) 469 { 470 int type = L3HDR(struct icmp,ip)->icmp_type; 471 472 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1 << type))); 473 } 474 475 #define TT ((1 << ICMP_ECHO) | \ 476 (1 << ICMP_ROUTERSOLICIT) | \ 477 (1 << ICMP_TSTAMP) | \ 478 (1 << ICMP_IREQ) | \ 479 (1 << ICMP_MASKREQ)) 480 481 static int 482 is_icmp_query(struct ip *ip) 483 { 484 int type = L3HDR(struct icmp, ip)->icmp_type; 485 486 return (type <= ICMP_MAXTYPE && (TT & (1 << type))); 487 } 488 489 #undef TT 490 491 /* 492 * The following checks use two arrays of 8 or 16 bits to store the 493 * bits that we want set or clear, respectively. They are in the 494 * low and high half of cmd->arg1 or cmd->d[0]. 495 * 496 * We scan options and store the bits we find set. We succeed if 497 * 498 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear 499 * 500 * The code is sometimes optimized not to store additional variables. 501 */ 502 503 static int 504 flags_match(ipfw_insn *cmd, uint8_t bits) 505 { 506 u_char want_clear; 507 bits = ~bits; 508 509 if (((cmd->arg1 & 0xff) & bits) != 0) 510 return 0; /* some bits we want set were clear */ 511 512 want_clear = (cmd->arg1 >> 8) & 0xff; 513 if ((want_clear & bits) != want_clear) 514 return 0; /* some bits we want clear were set */ 515 return 1; 516 } 517 518 static int 519 ipopts_match(struct ip *ip, ipfw_insn *cmd) 520 { 521 int optlen, bits = 0; 522 u_char *cp = (u_char *)(ip + 1); 523 int x = (ip->ip_hl << 2) - sizeof(struct ip); 524 525 for (; x > 0; x -= optlen, cp += optlen) { 526 int opt = cp[IPOPT_OPTVAL]; 527 528 if (opt == IPOPT_EOL) 529 break; 530 531 if (opt == IPOPT_NOP) { 532 optlen = 1; 533 } else { 534 optlen = cp[IPOPT_OLEN]; 535 if (optlen <= 0 || optlen > x) 536 return 0; /* invalid or truncated */ 537 } 538 539 switch (opt) { 540 case IPOPT_LSRR: 541 bits |= IP_FW_IPOPT_LSRR; 542 break; 543 544 case IPOPT_SSRR: 545 bits |= IP_FW_IPOPT_SSRR; 546 break; 547 548 case IPOPT_RR: 549 bits |= IP_FW_IPOPT_RR; 550 break; 551 552 case IPOPT_TS: 553 bits |= IP_FW_IPOPT_TS; 554 break; 555 556 default: 557 break; 558 } 559 } 560 return (flags_match(cmd, bits)); 561 } 562 563 static int 564 tcpopts_match(struct ip *ip, ipfw_insn *cmd) 565 { 566 int optlen, bits = 0; 567 struct tcphdr *tcp = L3HDR(struct tcphdr,ip); 568 u_char *cp = (u_char *)(tcp + 1); 569 int x = (tcp->th_off << 2) - sizeof(struct tcphdr); 570 571 for (; x > 0; x -= optlen, cp += optlen) { 572 int opt = cp[0]; 573 574 if (opt == TCPOPT_EOL) 575 break; 576 577 if (opt == TCPOPT_NOP) { 578 optlen = 1; 579 } else { 580 optlen = cp[1]; 581 if (optlen <= 0) 582 break; 583 } 584 585 switch (opt) { 586 case TCPOPT_MAXSEG: 587 bits |= IP_FW_TCPOPT_MSS; 588 break; 589 590 case TCPOPT_WINDOW: 591 bits |= IP_FW_TCPOPT_WINDOW; 592 break; 593 594 case TCPOPT_SACK_PERMITTED: 595 case TCPOPT_SACK: 596 bits |= IP_FW_TCPOPT_SACK; 597 break; 598 599 case TCPOPT_TIMESTAMP: 600 bits |= IP_FW_TCPOPT_TS; 601 break; 602 603 case TCPOPT_CC: 604 case TCPOPT_CCNEW: 605 case TCPOPT_CCECHO: 606 bits |= IP_FW_TCPOPT_CC; 607 break; 608 609 default: 610 break; 611 } 612 } 613 return (flags_match(cmd, bits)); 614 } 615 616 static int 617 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd) 618 { 619 if (ifp == NULL) /* no iface with this packet, match fails */ 620 return 0; 621 622 /* Check by name or by IP address */ 623 if (cmd->name[0] != '\0') { /* match by name */ 624 /* Check name */ 625 if (cmd->p.glob) { 626 if (kfnmatch(cmd->name, ifp->if_xname, 0) == 0) 627 return(1); 628 } else { 629 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0) 630 return(1); 631 } 632 } else { 633 struct ifaddr_container *ifac; 634 635 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 636 struct ifaddr *ia = ifac->ifa; 637 638 if (ia->ifa_addr == NULL) 639 continue; 640 if (ia->ifa_addr->sa_family != AF_INET) 641 continue; 642 if (cmd->p.ip.s_addr == ((struct sockaddr_in *) 643 (ia->ifa_addr))->sin_addr.s_addr) 644 return(1); /* match */ 645 } 646 } 647 return(0); /* no match, fail ... */ 648 } 649 650 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0 651 652 /* 653 * We enter here when we have a rule with O_LOG. 654 * XXX this function alone takes about 2Kbytes of code! 655 */ 656 static void 657 ipfw_log(struct ip_fw *f, u_int hlen, struct ether_header *eh, 658 struct mbuf *m, struct ifnet *oif) 659 { 660 char *action; 661 int limit_reached = 0; 662 char action2[40], proto[48], fragment[28]; 663 664 fragment[0] = '\0'; 665 proto[0] = '\0'; 666 667 if (f == NULL) { /* bogus pkt */ 668 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 669 670 if (verbose_limit != 0 && 671 ctx->ipfw_norule_counter >= verbose_limit) 672 return; 673 ctx->ipfw_norule_counter++; 674 if (ctx->ipfw_norule_counter == verbose_limit) 675 limit_reached = verbose_limit; 676 action = "Refuse"; 677 } else { /* O_LOG is the first action, find the real one */ 678 ipfw_insn *cmd = ACTION_PTR(f); 679 ipfw_insn_log *l = (ipfw_insn_log *)cmd; 680 681 if (l->max_log != 0 && l->log_left == 0) 682 return; 683 l->log_left--; 684 if (l->log_left == 0) 685 limit_reached = l->max_log; 686 cmd += F_LEN(cmd); /* point to first action */ 687 if (cmd->opcode == O_PROB) 688 cmd += F_LEN(cmd); 689 690 action = action2; 691 switch (cmd->opcode) { 692 case O_DENY: 693 action = "Deny"; 694 break; 695 696 case O_REJECT: 697 if (cmd->arg1==ICMP_REJECT_RST) { 698 action = "Reset"; 699 } else if (cmd->arg1==ICMP_UNREACH_HOST) { 700 action = "Reject"; 701 } else { 702 ksnprintf(SNPARGS(action2, 0), "Unreach %d", 703 cmd->arg1); 704 } 705 break; 706 707 case O_ACCEPT: 708 action = "Accept"; 709 break; 710 711 case O_COUNT: 712 action = "Count"; 713 break; 714 715 case O_DIVERT: 716 ksnprintf(SNPARGS(action2, 0), "Divert %d", cmd->arg1); 717 break; 718 719 case O_TEE: 720 ksnprintf(SNPARGS(action2, 0), "Tee %d", cmd->arg1); 721 break; 722 723 case O_SKIPTO: 724 ksnprintf(SNPARGS(action2, 0), "SkipTo %d", cmd->arg1); 725 break; 726 727 case O_PIPE: 728 ksnprintf(SNPARGS(action2, 0), "Pipe %d", cmd->arg1); 729 break; 730 731 case O_QUEUE: 732 ksnprintf(SNPARGS(action2, 0), "Queue %d", cmd->arg1); 733 break; 734 735 case O_FORWARD_IP: 736 { 737 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd; 738 int len; 739 740 len = ksnprintf(SNPARGS(action2, 0), 741 "Forward to %s", 742 inet_ntoa(sa->sa.sin_addr)); 743 if (sa->sa.sin_port) { 744 ksnprintf(SNPARGS(action2, len), ":%d", 745 sa->sa.sin_port); 746 } 747 } 748 break; 749 750 default: 751 action = "UNKNOWN"; 752 break; 753 } 754 } 755 756 if (hlen == 0) { /* non-ip */ 757 ksnprintf(SNPARGS(proto, 0), "MAC"); 758 } else { 759 struct ip *ip = mtod(m, struct ip *); 760 /* these three are all aliases to the same thing */ 761 struct icmp *const icmp = L3HDR(struct icmp, ip); 762 struct tcphdr *const tcp = (struct tcphdr *)icmp; 763 struct udphdr *const udp = (struct udphdr *)icmp; 764 765 int ip_off, offset, ip_len; 766 int len; 767 768 if (eh != NULL) { /* layer 2 packets are as on the wire */ 769 ip_off = ntohs(ip->ip_off); 770 ip_len = ntohs(ip->ip_len); 771 } else { 772 ip_off = ip->ip_off; 773 ip_len = ip->ip_len; 774 } 775 offset = ip_off & IP_OFFMASK; 776 switch (ip->ip_p) { 777 case IPPROTO_TCP: 778 len = ksnprintf(SNPARGS(proto, 0), "TCP %s", 779 inet_ntoa(ip->ip_src)); 780 if (offset == 0) { 781 ksnprintf(SNPARGS(proto, len), ":%d %s:%d", 782 ntohs(tcp->th_sport), 783 inet_ntoa(ip->ip_dst), 784 ntohs(tcp->th_dport)); 785 } else { 786 ksnprintf(SNPARGS(proto, len), " %s", 787 inet_ntoa(ip->ip_dst)); 788 } 789 break; 790 791 case IPPROTO_UDP: 792 len = ksnprintf(SNPARGS(proto, 0), "UDP %s", 793 inet_ntoa(ip->ip_src)); 794 if (offset == 0) { 795 ksnprintf(SNPARGS(proto, len), ":%d %s:%d", 796 ntohs(udp->uh_sport), 797 inet_ntoa(ip->ip_dst), 798 ntohs(udp->uh_dport)); 799 } else { 800 ksnprintf(SNPARGS(proto, len), " %s", 801 inet_ntoa(ip->ip_dst)); 802 } 803 break; 804 805 case IPPROTO_ICMP: 806 if (offset == 0) { 807 len = ksnprintf(SNPARGS(proto, 0), 808 "ICMP:%u.%u ", 809 icmp->icmp_type, 810 icmp->icmp_code); 811 } else { 812 len = ksnprintf(SNPARGS(proto, 0), "ICMP "); 813 } 814 len += ksnprintf(SNPARGS(proto, len), "%s", 815 inet_ntoa(ip->ip_src)); 816 ksnprintf(SNPARGS(proto, len), " %s", 817 inet_ntoa(ip->ip_dst)); 818 break; 819 820 default: 821 len = ksnprintf(SNPARGS(proto, 0), "P:%d %s", ip->ip_p, 822 inet_ntoa(ip->ip_src)); 823 ksnprintf(SNPARGS(proto, len), " %s", 824 inet_ntoa(ip->ip_dst)); 825 break; 826 } 827 828 if (ip_off & (IP_MF | IP_OFFMASK)) { 829 ksnprintf(SNPARGS(fragment, 0), " (frag %d:%d@%d%s)", 830 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2), 831 offset << 3, (ip_off & IP_MF) ? "+" : ""); 832 } 833 } 834 835 if (oif || m->m_pkthdr.rcvif) { 836 log(LOG_SECURITY | LOG_INFO, 837 "ipfw: %d %s %s %s via %s%s\n", 838 f ? f->rulenum : -1, 839 action, proto, oif ? "out" : "in", 840 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname, 841 fragment); 842 } else { 843 log(LOG_SECURITY | LOG_INFO, 844 "ipfw: %d %s %s [no if info]%s\n", 845 f ? f->rulenum : -1, 846 action, proto, fragment); 847 } 848 849 if (limit_reached) { 850 log(LOG_SECURITY | LOG_NOTICE, 851 "ipfw: limit %d reached on entry %d\n", 852 limit_reached, f ? f->rulenum : -1); 853 } 854 } 855 856 #undef SNPARGS 857 858 /* 859 * IMPORTANT: the hash function for dynamic rules must be commutative 860 * in source and destination (ip,port), because rules are bidirectional 861 * and we want to find both in the same bucket. 862 */ 863 static __inline int 864 hash_packet(struct ipfw_flow_id *id) 865 { 866 uint32_t i; 867 868 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port); 869 i &= (curr_dyn_buckets - 1); 870 return i; 871 } 872 873 /** 874 * unlink a dynamic rule from a chain. prev is a pointer to 875 * the previous one, q is a pointer to the rule to delete, 876 * head is a pointer to the head of the queue. 877 * Modifies q and potentially also head. 878 */ 879 #define UNLINK_DYN_RULE(prev, head, q) \ 880 do { \ 881 ipfw_dyn_rule *old_q = q; \ 882 \ 883 /* remove a refcount to the parent */ \ 884 if (q->dyn_type == O_LIMIT) \ 885 q->parent->count--; \ 886 DPRINTF("-- unlink entry 0x%08x %d -> 0x%08x %d, %d left\n", \ 887 q->id.src_ip, q->id.src_port, \ 888 q->id.dst_ip, q->id.dst_port, dyn_count - 1); \ 889 if (prev != NULL) \ 890 prev->next = q = q->next; \ 891 else \ 892 head = q = q->next; \ 893 KASSERT(dyn_count > 0, ("invalid dyn count %u", dyn_count)); \ 894 dyn_count--; \ 895 kfree(old_q, M_IPFW); \ 896 } while (0) 897 898 #define TIME_LEQ(a, b) ((int)((a) - (b)) <= 0) 899 900 /** 901 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL. 902 * 903 * If keep_me == NULL, rules are deleted even if not expired, 904 * otherwise only expired rules are removed. 905 * 906 * The value of the second parameter is also used to point to identify 907 * a rule we absolutely do not want to remove (e.g. because we are 908 * holding a reference to it -- this is the case with O_LIMIT_PARENT 909 * rules). The pointer is only used for comparison, so any non-null 910 * value will do. 911 */ 912 static void 913 remove_dyn_rule_locked(struct ip_fw *rule, ipfw_dyn_rule *keep_me) 914 { 915 static uint32_t last_remove = 0; /* XXX */ 916 917 #define FORCE (keep_me == NULL) 918 919 ipfw_dyn_rule *prev, *q; 920 int i, pass = 0, max_pass = 0, unlinked = 0; 921 922 if (ipfw_dyn_v == NULL || dyn_count == 0) 923 return; 924 /* do not expire more than once per second, it is useless */ 925 if (!FORCE && last_remove == time_second) 926 return; 927 last_remove = time_second; 928 929 /* 930 * because O_LIMIT refer to parent rules, during the first pass only 931 * remove child and mark any pending LIMIT_PARENT, and remove 932 * them in a second pass. 933 */ 934 next_pass: 935 for (i = 0; i < curr_dyn_buckets; i++) { 936 for (prev = NULL, q = ipfw_dyn_v[i]; q;) { 937 /* 938 * Logic can become complex here, so we split tests. 939 */ 940 if (q == keep_me) 941 goto next; 942 if (rule != NULL && rule->stub != q->stub) 943 goto next; /* not the one we are looking for */ 944 if (q->dyn_type == O_LIMIT_PARENT) { 945 /* 946 * handle parent in the second pass, 947 * record we need one. 948 */ 949 max_pass = 1; 950 if (pass == 0) 951 goto next; 952 if (FORCE && q->count != 0) { 953 /* XXX should not happen! */ 954 kprintf("OUCH! cannot remove rule, " 955 "count %d\n", q->count); 956 } 957 } else { 958 if (!FORCE && !TIME_LEQ(q->expire, time_second)) 959 goto next; 960 } 961 unlinked = 1; 962 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 963 continue; 964 next: 965 prev = q; 966 q = q->next; 967 } 968 } 969 if (pass++ < max_pass) 970 goto next_pass; 971 972 if (unlinked) 973 ++dyn_buckets_gen; 974 975 #undef FORCE 976 } 977 978 /** 979 * lookup a dynamic rule. 980 */ 981 static ipfw_dyn_rule * 982 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction, 983 struct tcphdr *tcp) 984 { 985 /* 986 * stateful ipfw extensions. 987 * Lookup into dynamic session queue 988 */ 989 #define MATCH_REVERSE 0 990 #define MATCH_FORWARD 1 991 #define MATCH_NONE 2 992 #define MATCH_UNKNOWN 3 993 int i, dir = MATCH_NONE; 994 ipfw_dyn_rule *prev, *q=NULL; 995 996 if (ipfw_dyn_v == NULL) 997 goto done; /* not found */ 998 999 i = hash_packet(pkt); 1000 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) { 1001 if (q->dyn_type == O_LIMIT_PARENT) 1002 goto next; 1003 1004 if (TIME_LEQ(q->expire, time_second)) { 1005 /* 1006 * Entry expired; skip. 1007 * Let ipfw_tick() take care of it 1008 */ 1009 goto next; 1010 } 1011 1012 if (pkt->proto == q->id.proto) { 1013 if (pkt->src_ip == q->id.src_ip && 1014 pkt->dst_ip == q->id.dst_ip && 1015 pkt->src_port == q->id.src_port && 1016 pkt->dst_port == q->id.dst_port) { 1017 dir = MATCH_FORWARD; 1018 break; 1019 } 1020 if (pkt->src_ip == q->id.dst_ip && 1021 pkt->dst_ip == q->id.src_ip && 1022 pkt->src_port == q->id.dst_port && 1023 pkt->dst_port == q->id.src_port) { 1024 dir = MATCH_REVERSE; 1025 break; 1026 } 1027 } 1028 next: 1029 prev = q; 1030 q = q->next; 1031 } 1032 if (q == NULL) 1033 goto done; /* q = NULL, not found */ 1034 1035 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */ 1036 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST); 1037 1038 #define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 1039 #define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 1040 1041 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8); 1042 switch (q->state) { 1043 case TH_SYN: /* opening */ 1044 q->expire = time_second + dyn_syn_lifetime; 1045 break; 1046 1047 case BOTH_SYN: /* move to established */ 1048 case BOTH_SYN | TH_FIN : /* one side tries to close */ 1049 case BOTH_SYN | (TH_FIN << 8) : 1050 if (tcp) { 1051 uint32_t ack = ntohl(tcp->th_ack); 1052 1053 #define _SEQ_GE(a, b) ((int)(a) - (int)(b) >= 0) 1054 1055 if (dir == MATCH_FORWARD) { 1056 if (q->ack_fwd == 0 || 1057 _SEQ_GE(ack, q->ack_fwd)) 1058 q->ack_fwd = ack; 1059 else /* ignore out-of-sequence */ 1060 break; 1061 } else { 1062 if (q->ack_rev == 0 || 1063 _SEQ_GE(ack, q->ack_rev)) 1064 q->ack_rev = ack; 1065 else /* ignore out-of-sequence */ 1066 break; 1067 } 1068 #undef _SEQ_GE 1069 } 1070 q->expire = time_second + dyn_ack_lifetime; 1071 break; 1072 1073 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 1074 KKASSERT(dyn_fin_lifetime < dyn_keepalive_period); 1075 q->expire = time_second + dyn_fin_lifetime; 1076 break; 1077 1078 default: 1079 #if 0 1080 /* 1081 * reset or some invalid combination, but can also 1082 * occur if we use keep-state the wrong way. 1083 */ 1084 if ((q->state & ((TH_RST << 8) | TH_RST)) == 0) 1085 kprintf("invalid state: 0x%x\n", q->state); 1086 #endif 1087 KKASSERT(dyn_rst_lifetime < dyn_keepalive_period); 1088 q->expire = time_second + dyn_rst_lifetime; 1089 break; 1090 } 1091 } else if (pkt->proto == IPPROTO_UDP) { 1092 q->expire = time_second + dyn_udp_lifetime; 1093 } else { 1094 /* other protocols */ 1095 q->expire = time_second + dyn_short_lifetime; 1096 } 1097 done: 1098 if (match_direction) 1099 *match_direction = dir; 1100 return q; 1101 } 1102 1103 static struct ip_fw * 1104 lookup_rule(struct ipfw_flow_id *pkt, int *match_direction, struct tcphdr *tcp, 1105 uint16_t len, int *deny) 1106 { 1107 struct ip_fw *rule = NULL; 1108 ipfw_dyn_rule *q; 1109 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 1110 uint32_t gen; 1111 1112 *deny = 0; 1113 gen = ctx->ipfw_gen; 1114 1115 lockmgr(&dyn_lock, LK_SHARED); 1116 1117 if (ctx->ipfw_gen != gen) { 1118 /* 1119 * Static rules had been change when we were waiting 1120 * for the dynamic hash table lock; deny this packet, 1121 * since it is _not_ known whether it is safe to keep 1122 * iterating the static rules. 1123 */ 1124 *deny = 1; 1125 goto back; 1126 } 1127 1128 q = lookup_dyn_rule(pkt, match_direction, tcp); 1129 if (q == NULL) { 1130 rule = NULL; 1131 } else { 1132 rule = q->stub->rule[mycpuid]; 1133 KKASSERT(rule->stub == q->stub && rule->cpuid == mycpuid); 1134 1135 /* XXX */ 1136 q->pcnt++; 1137 q->bcnt += len; 1138 } 1139 back: 1140 lockmgr(&dyn_lock, LK_RELEASE); 1141 return rule; 1142 } 1143 1144 static void 1145 realloc_dynamic_table(void) 1146 { 1147 ipfw_dyn_rule **old_dyn_v; 1148 uint32_t old_curr_dyn_buckets; 1149 1150 KASSERT(dyn_buckets <= 65536 && (dyn_buckets & (dyn_buckets - 1)) == 0, 1151 ("invalid dyn_buckets %d", dyn_buckets)); 1152 1153 /* Save the current buckets array for later error recovery */ 1154 old_dyn_v = ipfw_dyn_v; 1155 old_curr_dyn_buckets = curr_dyn_buckets; 1156 1157 curr_dyn_buckets = dyn_buckets; 1158 for (;;) { 1159 ipfw_dyn_v = kmalloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *), 1160 M_IPFW, M_NOWAIT | M_ZERO); 1161 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2) 1162 break; 1163 1164 curr_dyn_buckets /= 2; 1165 if (curr_dyn_buckets <= old_curr_dyn_buckets && 1166 old_dyn_v != NULL) { 1167 /* 1168 * Don't try allocating smaller buckets array, reuse 1169 * the old one, which alreay contains enough buckets 1170 */ 1171 break; 1172 } 1173 } 1174 1175 if (ipfw_dyn_v != NULL) { 1176 if (old_dyn_v != NULL) 1177 kfree(old_dyn_v, M_IPFW); 1178 } else { 1179 /* Allocation failed, restore old buckets array */ 1180 ipfw_dyn_v = old_dyn_v; 1181 curr_dyn_buckets = old_curr_dyn_buckets; 1182 } 1183 1184 if (ipfw_dyn_v != NULL) 1185 ++dyn_buckets_gen; 1186 } 1187 1188 /** 1189 * Install state of type 'type' for a dynamic session. 1190 * The hash table contains two type of rules: 1191 * - regular rules (O_KEEP_STATE) 1192 * - rules for sessions with limited number of sess per user 1193 * (O_LIMIT). When they are created, the parent is 1194 * increased by 1, and decreased on delete. In this case, 1195 * the third parameter is the parent rule and not the chain. 1196 * - "parent" rules for the above (O_LIMIT_PARENT). 1197 */ 1198 static ipfw_dyn_rule * 1199 add_dyn_rule(struct ipfw_flow_id *id, uint8_t dyn_type, struct ip_fw *rule) 1200 { 1201 ipfw_dyn_rule *r; 1202 int i; 1203 1204 if (ipfw_dyn_v == NULL || 1205 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) { 1206 realloc_dynamic_table(); 1207 if (ipfw_dyn_v == NULL) 1208 return NULL; /* failed ! */ 1209 } 1210 i = hash_packet(id); 1211 1212 r = kmalloc(sizeof(*r), M_IPFW, M_NOWAIT | M_ZERO); 1213 if (r == NULL) 1214 return NULL; 1215 1216 /* increase refcount on parent, and set pointer */ 1217 if (dyn_type == O_LIMIT) { 1218 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule; 1219 1220 if (parent->dyn_type != O_LIMIT_PARENT) 1221 panic("invalid parent"); 1222 parent->count++; 1223 r->parent = parent; 1224 rule = parent->stub->rule[mycpuid]; 1225 KKASSERT(rule->stub == parent->stub); 1226 } 1227 KKASSERT(rule->cpuid == mycpuid && rule->stub != NULL); 1228 1229 r->id = *id; 1230 r->expire = time_second + dyn_syn_lifetime; 1231 r->stub = rule->stub; 1232 r->dyn_type = dyn_type; 1233 r->pcnt = r->bcnt = 0; 1234 r->count = 0; 1235 1236 r->bucket = i; 1237 r->next = ipfw_dyn_v[i]; 1238 ipfw_dyn_v[i] = r; 1239 dyn_count++; 1240 dyn_buckets_gen++; 1241 DPRINTF("-- add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n", 1242 dyn_type, 1243 r->id.src_ip, r->id.src_port, 1244 r->id.dst_ip, r->id.dst_port, dyn_count); 1245 return r; 1246 } 1247 1248 /** 1249 * lookup dynamic parent rule using pkt and rule as search keys. 1250 * If the lookup fails, then install one. 1251 */ 1252 static ipfw_dyn_rule * 1253 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule) 1254 { 1255 ipfw_dyn_rule *q; 1256 int i; 1257 1258 if (ipfw_dyn_v) { 1259 i = hash_packet(pkt); 1260 for (q = ipfw_dyn_v[i]; q != NULL; q = q->next) { 1261 if (q->dyn_type == O_LIMIT_PARENT && 1262 rule->stub == q->stub && 1263 pkt->proto == q->id.proto && 1264 pkt->src_ip == q->id.src_ip && 1265 pkt->dst_ip == q->id.dst_ip && 1266 pkt->src_port == q->id.src_port && 1267 pkt->dst_port == q->id.dst_port) { 1268 q->expire = time_second + dyn_short_lifetime; 1269 DPRINTF("lookup_dyn_parent found 0x%p\n", q); 1270 return q; 1271 } 1272 } 1273 } 1274 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule); 1275 } 1276 1277 /** 1278 * Install dynamic state for rule type cmd->o.opcode 1279 * 1280 * Returns 1 (failure) if state is not installed because of errors or because 1281 * session limitations are enforced. 1282 */ 1283 static int 1284 install_state_locked(struct ip_fw *rule, ipfw_insn_limit *cmd, 1285 struct ip_fw_args *args) 1286 { 1287 static int last_log; /* XXX */ 1288 1289 ipfw_dyn_rule *q; 1290 1291 DPRINTF("-- install state type %d 0x%08x %u -> 0x%08x %u\n", 1292 cmd->o.opcode, 1293 args->f_id.src_ip, args->f_id.src_port, 1294 args->f_id.dst_ip, args->f_id.dst_port); 1295 1296 q = lookup_dyn_rule(&args->f_id, NULL, NULL); 1297 if (q != NULL) { /* should never occur */ 1298 if (last_log != time_second) { 1299 last_log = time_second; 1300 kprintf(" install_state: entry already present, done\n"); 1301 } 1302 return 0; 1303 } 1304 1305 if (dyn_count >= dyn_max) { 1306 /* 1307 * Run out of slots, try to remove any expired rule. 1308 */ 1309 remove_dyn_rule_locked(NULL, (ipfw_dyn_rule *)1); 1310 if (dyn_count >= dyn_max) { 1311 if (last_log != time_second) { 1312 last_log = time_second; 1313 kprintf("install_state: " 1314 "Too many dynamic rules\n"); 1315 } 1316 return 1; /* cannot install, notify caller */ 1317 } 1318 } 1319 1320 switch (cmd->o.opcode) { 1321 case O_KEEP_STATE: /* bidir rule */ 1322 if (add_dyn_rule(&args->f_id, O_KEEP_STATE, rule) == NULL) 1323 return 1; 1324 break; 1325 1326 case O_LIMIT: /* limit number of sessions */ 1327 { 1328 uint16_t limit_mask = cmd->limit_mask; 1329 struct ipfw_flow_id id; 1330 ipfw_dyn_rule *parent; 1331 1332 DPRINTF("installing dyn-limit rule %d\n", 1333 cmd->conn_limit); 1334 1335 id.dst_ip = id.src_ip = 0; 1336 id.dst_port = id.src_port = 0; 1337 id.proto = args->f_id.proto; 1338 1339 if (limit_mask & DYN_SRC_ADDR) 1340 id.src_ip = args->f_id.src_ip; 1341 if (limit_mask & DYN_DST_ADDR) 1342 id.dst_ip = args->f_id.dst_ip; 1343 if (limit_mask & DYN_SRC_PORT) 1344 id.src_port = args->f_id.src_port; 1345 if (limit_mask & DYN_DST_PORT) 1346 id.dst_port = args->f_id.dst_port; 1347 1348 parent = lookup_dyn_parent(&id, rule); 1349 if (parent == NULL) { 1350 kprintf("add parent failed\n"); 1351 return 1; 1352 } 1353 1354 if (parent->count >= cmd->conn_limit) { 1355 /* 1356 * See if we can remove some expired rule. 1357 */ 1358 remove_dyn_rule_locked(rule, parent); 1359 if (parent->count >= cmd->conn_limit) { 1360 if (fw_verbose && 1361 last_log != time_second) { 1362 last_log = time_second; 1363 log(LOG_SECURITY | LOG_DEBUG, 1364 "drop session, " 1365 "too many entries\n"); 1366 } 1367 return 1; 1368 } 1369 } 1370 if (add_dyn_rule(&args->f_id, O_LIMIT, 1371 (struct ip_fw *)parent) == NULL) 1372 return 1; 1373 } 1374 break; 1375 default: 1376 kprintf("unknown dynamic rule type %u\n", cmd->o.opcode); 1377 return 1; 1378 } 1379 lookup_dyn_rule(&args->f_id, NULL, NULL); /* XXX just set lifetime */ 1380 return 0; 1381 } 1382 1383 static int 1384 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd, 1385 struct ip_fw_args *args, int *deny) 1386 { 1387 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 1388 uint32_t gen; 1389 int ret = 0; 1390 1391 *deny = 0; 1392 gen = ctx->ipfw_gen; 1393 1394 lockmgr(&dyn_lock, LK_EXCLUSIVE); 1395 if (ctx->ipfw_gen != gen) { 1396 /* See the comment in lookup_rule() */ 1397 *deny = 1; 1398 } else { 1399 ret = install_state_locked(rule, cmd, args); 1400 } 1401 lockmgr(&dyn_lock, LK_RELEASE); 1402 1403 return ret; 1404 } 1405 1406 /* 1407 * Transmit a TCP packet, containing either a RST or a keepalive. 1408 * When flags & TH_RST, we are sending a RST packet, because of a 1409 * "reset" action matched the packet. 1410 * Otherwise we are sending a keepalive, and flags & TH_ 1411 */ 1412 static void 1413 send_pkt(struct ipfw_flow_id *id, uint32_t seq, uint32_t ack, int flags) 1414 { 1415 struct mbuf *m; 1416 struct ip *ip; 1417 struct tcphdr *tcp; 1418 struct route sro; /* fake route */ 1419 1420 MGETHDR(m, MB_DONTWAIT, MT_HEADER); 1421 if (m == NULL) 1422 return; 1423 m->m_pkthdr.rcvif = NULL; 1424 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr); 1425 m->m_data += max_linkhdr; 1426 1427 ip = mtod(m, struct ip *); 1428 bzero(ip, m->m_len); 1429 tcp = (struct tcphdr *)(ip + 1); /* no IP options */ 1430 ip->ip_p = IPPROTO_TCP; 1431 tcp->th_off = 5; 1432 1433 /* 1434 * Assume we are sending a RST (or a keepalive in the reverse 1435 * direction), swap src and destination addresses and ports. 1436 */ 1437 ip->ip_src.s_addr = htonl(id->dst_ip); 1438 ip->ip_dst.s_addr = htonl(id->src_ip); 1439 tcp->th_sport = htons(id->dst_port); 1440 tcp->th_dport = htons(id->src_port); 1441 if (flags & TH_RST) { /* we are sending a RST */ 1442 if (flags & TH_ACK) { 1443 tcp->th_seq = htonl(ack); 1444 tcp->th_ack = htonl(0); 1445 tcp->th_flags = TH_RST; 1446 } else { 1447 if (flags & TH_SYN) 1448 seq++; 1449 tcp->th_seq = htonl(0); 1450 tcp->th_ack = htonl(seq); 1451 tcp->th_flags = TH_RST | TH_ACK; 1452 } 1453 } else { 1454 /* 1455 * We are sending a keepalive. flags & TH_SYN determines 1456 * the direction, forward if set, reverse if clear. 1457 * NOTE: seq and ack are always assumed to be correct 1458 * as set by the caller. This may be confusing... 1459 */ 1460 if (flags & TH_SYN) { 1461 /* 1462 * we have to rewrite the correct addresses! 1463 */ 1464 ip->ip_dst.s_addr = htonl(id->dst_ip); 1465 ip->ip_src.s_addr = htonl(id->src_ip); 1466 tcp->th_dport = htons(id->dst_port); 1467 tcp->th_sport = htons(id->src_port); 1468 } 1469 tcp->th_seq = htonl(seq); 1470 tcp->th_ack = htonl(ack); 1471 tcp->th_flags = TH_ACK; 1472 } 1473 1474 /* 1475 * set ip_len to the payload size so we can compute 1476 * the tcp checksum on the pseudoheader 1477 * XXX check this, could save a couple of words ? 1478 */ 1479 ip->ip_len = htons(sizeof(struct tcphdr)); 1480 tcp->th_sum = in_cksum(m, m->m_pkthdr.len); 1481 1482 /* 1483 * now fill fields left out earlier 1484 */ 1485 ip->ip_ttl = ip_defttl; 1486 ip->ip_len = m->m_pkthdr.len; 1487 1488 bzero(&sro, sizeof(sro)); 1489 ip_rtaddr(ip->ip_dst, &sro); 1490 1491 m->m_pkthdr.fw_flags |= IPFW_MBUF_GENERATED; 1492 ip_output(m, NULL, &sro, 0, NULL, NULL); 1493 if (sro.ro_rt) 1494 RTFREE(sro.ro_rt); 1495 } 1496 1497 /* 1498 * sends a reject message, consuming the mbuf passed as an argument. 1499 */ 1500 static void 1501 send_reject(struct ip_fw_args *args, int code, int offset, int ip_len) 1502 { 1503 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */ 1504 /* We need the IP header in host order for icmp_error(). */ 1505 if (args->eh != NULL) { 1506 struct ip *ip = mtod(args->m, struct ip *); 1507 1508 ip->ip_len = ntohs(ip->ip_len); 1509 ip->ip_off = ntohs(ip->ip_off); 1510 } 1511 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0); 1512 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) { 1513 struct tcphdr *const tcp = 1514 L3HDR(struct tcphdr, mtod(args->m, struct ip *)); 1515 1516 if ((tcp->th_flags & TH_RST) == 0) { 1517 send_pkt(&args->f_id, ntohl(tcp->th_seq), 1518 ntohl(tcp->th_ack), tcp->th_flags | TH_RST); 1519 } 1520 m_freem(args->m); 1521 } else { 1522 m_freem(args->m); 1523 } 1524 args->m = NULL; 1525 } 1526 1527 /** 1528 * 1529 * Given an ip_fw *, lookup_next_rule will return a pointer 1530 * to the next rule, which can be either the jump 1531 * target (for skipto instructions) or the next one in the list (in 1532 * all other cases including a missing jump target). 1533 * The result is also written in the "next_rule" field of the rule. 1534 * Backward jumps are not allowed, so start looking from the next 1535 * rule... 1536 * 1537 * This never returns NULL -- in case we do not have an exact match, 1538 * the next rule is returned. When the ruleset is changed, 1539 * pointers are flushed so we are always correct. 1540 */ 1541 1542 static struct ip_fw * 1543 lookup_next_rule(struct ip_fw *me) 1544 { 1545 struct ip_fw *rule = NULL; 1546 ipfw_insn *cmd; 1547 1548 /* look for action, in case it is a skipto */ 1549 cmd = ACTION_PTR(me); 1550 if (cmd->opcode == O_LOG) 1551 cmd += F_LEN(cmd); 1552 if (cmd->opcode == O_SKIPTO) { 1553 for (rule = me->next; rule; rule = rule->next) { 1554 if (rule->rulenum >= cmd->arg1) 1555 break; 1556 } 1557 } 1558 if (rule == NULL) /* failure or not a skipto */ 1559 rule = me->next; 1560 me->next_rule = rule; 1561 return rule; 1562 } 1563 1564 static int 1565 _ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif, 1566 enum ipfw_opcodes opcode, uid_t uid) 1567 { 1568 struct in_addr src_ip, dst_ip; 1569 struct inpcbinfo *pi; 1570 int wildcard; 1571 struct inpcb *pcb; 1572 1573 if (fid->proto == IPPROTO_TCP) { 1574 wildcard = 0; 1575 pi = &tcbinfo[mycpuid]; 1576 } else if (fid->proto == IPPROTO_UDP) { 1577 wildcard = 1; 1578 pi = &udbinfo; 1579 } else { 1580 return 0; 1581 } 1582 1583 /* 1584 * Values in 'fid' are in host byte order 1585 */ 1586 dst_ip.s_addr = htonl(fid->dst_ip); 1587 src_ip.s_addr = htonl(fid->src_ip); 1588 if (oif) { 1589 pcb = in_pcblookup_hash(pi, 1590 dst_ip, htons(fid->dst_port), 1591 src_ip, htons(fid->src_port), 1592 wildcard, oif); 1593 } else { 1594 pcb = in_pcblookup_hash(pi, 1595 src_ip, htons(fid->src_port), 1596 dst_ip, htons(fid->dst_port), 1597 wildcard, NULL); 1598 } 1599 if (pcb == NULL || pcb->inp_socket == NULL) 1600 return 0; 1601 1602 if (opcode == O_UID) { 1603 #define socheckuid(a,b) ((a)->so_cred->cr_uid != (b)) 1604 return !socheckuid(pcb->inp_socket, uid); 1605 #undef socheckuid 1606 } else { 1607 return groupmember(uid, pcb->inp_socket->so_cred); 1608 } 1609 } 1610 1611 static int 1612 ipfw_match_uid(const struct ipfw_flow_id *fid, struct ifnet *oif, 1613 enum ipfw_opcodes opcode, uid_t uid, int *deny) 1614 { 1615 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 1616 uint32_t gen; 1617 int match = 0; 1618 1619 *deny = 0; 1620 gen = ctx->ipfw_gen; 1621 1622 get_mplock(); 1623 if (gen != ctx->ipfw_gen) { 1624 /* See the comment in lookup_rule() */ 1625 *deny = 1; 1626 } else { 1627 match = _ipfw_match_uid(fid, oif, opcode, uid); 1628 } 1629 rel_mplock(); 1630 return match; 1631 } 1632 1633 /* 1634 * The main check routine for the firewall. 1635 * 1636 * All arguments are in args so we can modify them and return them 1637 * back to the caller. 1638 * 1639 * Parameters: 1640 * 1641 * args->m (in/out) The packet; we set to NULL when/if we nuke it. 1642 * Starts with the IP header. 1643 * args->eh (in) Mac header if present, or NULL for layer3 packet. 1644 * args->oif Outgoing interface, or NULL if packet is incoming. 1645 * The incoming interface is in the mbuf. (in) 1646 * 1647 * args->rule Pointer to the last matching rule (in/out) 1648 * args->f_id Addresses grabbed from the packet (out) 1649 * 1650 * Return value: 1651 * 1652 * If the packet was denied/rejected and has been dropped, *m is equal 1653 * to NULL upon return. 1654 * 1655 * IP_FW_DENY the packet must be dropped. 1656 * IP_FW_PASS The packet is to be accepted and routed normally. 1657 * IP_FW_DIVERT Divert the packet to port (args->cookie) 1658 * IP_FW_TEE Tee the packet to port (args->cookie) 1659 * IP_FW_DUMMYNET Send the packet to pipe/queue (args->cookie) 1660 */ 1661 1662 static int 1663 ipfw_chk(struct ip_fw_args *args) 1664 { 1665 /* 1666 * Local variables hold state during the processing of a packet. 1667 * 1668 * IMPORTANT NOTE: to speed up the processing of rules, there 1669 * are some assumption on the values of the variables, which 1670 * are documented here. Should you change them, please check 1671 * the implementation of the various instructions to make sure 1672 * that they still work. 1673 * 1674 * args->eh The MAC header. It is non-null for a layer2 1675 * packet, it is NULL for a layer-3 packet. 1676 * 1677 * m | args->m Pointer to the mbuf, as received from the caller. 1678 * It may change if ipfw_chk() does an m_pullup, or if it 1679 * consumes the packet because it calls send_reject(). 1680 * XXX This has to change, so that ipfw_chk() never modifies 1681 * or consumes the buffer. 1682 * ip is simply an alias of the value of m, and it is kept 1683 * in sync with it (the packet is supposed to start with 1684 * the ip header). 1685 */ 1686 struct mbuf *m = args->m; 1687 struct ip *ip = mtod(m, struct ip *); 1688 1689 /* 1690 * oif | args->oif If NULL, ipfw_chk has been called on the 1691 * inbound path (ether_input, ip_input). 1692 * If non-NULL, ipfw_chk has been called on the outbound path 1693 * (ether_output, ip_output). 1694 */ 1695 struct ifnet *oif = args->oif; 1696 1697 struct ip_fw *f = NULL; /* matching rule */ 1698 int retval = IP_FW_PASS; 1699 struct m_tag *mtag; 1700 struct divert_info *divinfo; 1701 1702 /* 1703 * hlen The length of the IPv4 header. 1704 * hlen >0 means we have an IPv4 packet. 1705 */ 1706 u_int hlen = 0; /* hlen >0 means we have an IP pkt */ 1707 1708 /* 1709 * offset The offset of a fragment. offset != 0 means that 1710 * we have a fragment at this offset of an IPv4 packet. 1711 * offset == 0 means that (if this is an IPv4 packet) 1712 * this is the first or only fragment. 1713 */ 1714 u_short offset = 0; 1715 1716 /* 1717 * Local copies of addresses. They are only valid if we have 1718 * an IP packet. 1719 * 1720 * proto The protocol. Set to 0 for non-ip packets, 1721 * or to the protocol read from the packet otherwise. 1722 * proto != 0 means that we have an IPv4 packet. 1723 * 1724 * src_port, dst_port port numbers, in HOST format. Only 1725 * valid for TCP and UDP packets. 1726 * 1727 * src_ip, dst_ip ip addresses, in NETWORK format. 1728 * Only valid for IPv4 packets. 1729 */ 1730 uint8_t proto; 1731 uint16_t src_port = 0, dst_port = 0; /* NOTE: host format */ 1732 struct in_addr src_ip, dst_ip; /* NOTE: network format */ 1733 uint16_t ip_len = 0; 1734 1735 /* 1736 * dyn_dir = MATCH_UNKNOWN when rules unchecked, 1737 * MATCH_NONE when checked and not matched (dyn_f = NULL), 1738 * MATCH_FORWARD or MATCH_REVERSE otherwise (dyn_f != NULL) 1739 */ 1740 int dyn_dir = MATCH_UNKNOWN; 1741 struct ip_fw *dyn_f = NULL; 1742 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 1743 1744 if (m->m_pkthdr.fw_flags & IPFW_MBUF_GENERATED) 1745 return IP_FW_PASS; /* accept */ 1746 1747 if (args->eh == NULL || /* layer 3 packet */ 1748 (m->m_pkthdr.len >= sizeof(struct ip) && 1749 ntohs(args->eh->ether_type) == ETHERTYPE_IP)) 1750 hlen = ip->ip_hl << 2; 1751 1752 /* 1753 * Collect parameters into local variables for faster matching. 1754 */ 1755 if (hlen == 0) { /* do not grab addresses for non-ip pkts */ 1756 proto = args->f_id.proto = 0; /* mark f_id invalid */ 1757 goto after_ip_checks; 1758 } 1759 1760 proto = args->f_id.proto = ip->ip_p; 1761 src_ip = ip->ip_src; 1762 dst_ip = ip->ip_dst; 1763 if (args->eh != NULL) { /* layer 2 packets are as on the wire */ 1764 offset = ntohs(ip->ip_off) & IP_OFFMASK; 1765 ip_len = ntohs(ip->ip_len); 1766 } else { 1767 offset = ip->ip_off & IP_OFFMASK; 1768 ip_len = ip->ip_len; 1769 } 1770 1771 #define PULLUP_TO(len) \ 1772 do { \ 1773 if (m->m_len < (len)) { \ 1774 args->m = m = m_pullup(m, (len));\ 1775 if (m == NULL) \ 1776 goto pullup_failed; \ 1777 ip = mtod(m, struct ip *); \ 1778 } \ 1779 } while (0) 1780 1781 if (offset == 0) { 1782 switch (proto) { 1783 case IPPROTO_TCP: 1784 { 1785 struct tcphdr *tcp; 1786 1787 PULLUP_TO(hlen + sizeof(struct tcphdr)); 1788 tcp = L3HDR(struct tcphdr, ip); 1789 dst_port = tcp->th_dport; 1790 src_port = tcp->th_sport; 1791 args->f_id.flags = tcp->th_flags; 1792 } 1793 break; 1794 1795 case IPPROTO_UDP: 1796 { 1797 struct udphdr *udp; 1798 1799 PULLUP_TO(hlen + sizeof(struct udphdr)); 1800 udp = L3HDR(struct udphdr, ip); 1801 dst_port = udp->uh_dport; 1802 src_port = udp->uh_sport; 1803 } 1804 break; 1805 1806 case IPPROTO_ICMP: 1807 PULLUP_TO(hlen + 4); /* type, code and checksum. */ 1808 args->f_id.flags = L3HDR(struct icmp, ip)->icmp_type; 1809 break; 1810 1811 default: 1812 break; 1813 } 1814 } 1815 1816 #undef PULLUP_TO 1817 1818 args->f_id.src_ip = ntohl(src_ip.s_addr); 1819 args->f_id.dst_ip = ntohl(dst_ip.s_addr); 1820 args->f_id.src_port = src_port = ntohs(src_port); 1821 args->f_id.dst_port = dst_port = ntohs(dst_port); 1822 1823 after_ip_checks: 1824 if (args->rule) { 1825 /* 1826 * Packet has already been tagged. Look for the next rule 1827 * to restart processing. 1828 * 1829 * If fw_one_pass != 0 then just accept it. 1830 * XXX should not happen here, but optimized out in 1831 * the caller. 1832 */ 1833 if (fw_one_pass) 1834 return IP_FW_PASS; 1835 1836 /* This rule is being/has been flushed */ 1837 if (ipfw_flushing) 1838 return IP_FW_DENY; 1839 1840 KASSERT(args->rule->cpuid == mycpuid, 1841 ("rule used on cpu%d", mycpuid)); 1842 1843 /* This rule was deleted */ 1844 if (args->rule->rule_flags & IPFW_RULE_F_INVALID) 1845 return IP_FW_DENY; 1846 1847 f = args->rule->next_rule; 1848 if (f == NULL) 1849 f = lookup_next_rule(args->rule); 1850 } else { 1851 /* 1852 * Find the starting rule. It can be either the first 1853 * one, or the one after divert_rule if asked so. 1854 */ 1855 int skipto; 1856 1857 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL); 1858 if (mtag != NULL) { 1859 divinfo = m_tag_data(mtag); 1860 skipto = divinfo->skipto; 1861 } else { 1862 skipto = 0; 1863 } 1864 1865 f = ctx->ipfw_layer3_chain; 1866 if (args->eh == NULL && skipto != 0) { 1867 /* No skipto during rule flushing */ 1868 if (ipfw_flushing) 1869 return IP_FW_DENY; 1870 1871 if (skipto >= IPFW_DEFAULT_RULE) 1872 return IP_FW_DENY; /* invalid */ 1873 1874 while (f && f->rulenum <= skipto) 1875 f = f->next; 1876 if (f == NULL) /* drop packet */ 1877 return IP_FW_DENY; 1878 } else if (ipfw_flushing) { 1879 /* Rules are being flushed; skip to default rule */ 1880 f = ctx->ipfw_default_rule; 1881 } 1882 } 1883 if ((mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL)) != NULL) 1884 m_tag_delete(m, mtag); 1885 1886 /* 1887 * Now scan the rules, and parse microinstructions for each rule. 1888 */ 1889 for (; f; f = f->next) { 1890 int l, cmdlen; 1891 ipfw_insn *cmd; 1892 int skip_or; /* skip rest of OR block */ 1893 1894 again: 1895 if (ctx->ipfw_set_disable & (1 << f->set)) 1896 continue; 1897 1898 skip_or = 0; 1899 for (l = f->cmd_len, cmd = f->cmd; l > 0; 1900 l -= cmdlen, cmd += cmdlen) { 1901 int match, deny; 1902 1903 /* 1904 * check_body is a jump target used when we find a 1905 * CHECK_STATE, and need to jump to the body of 1906 * the target rule. 1907 */ 1908 1909 check_body: 1910 cmdlen = F_LEN(cmd); 1911 /* 1912 * An OR block (insn_1 || .. || insn_n) has the 1913 * F_OR bit set in all but the last instruction. 1914 * The first match will set "skip_or", and cause 1915 * the following instructions to be skipped until 1916 * past the one with the F_OR bit clear. 1917 */ 1918 if (skip_or) { /* skip this instruction */ 1919 if ((cmd->len & F_OR) == 0) 1920 skip_or = 0; /* next one is good */ 1921 continue; 1922 } 1923 match = 0; /* set to 1 if we succeed */ 1924 1925 switch (cmd->opcode) { 1926 /* 1927 * The first set of opcodes compares the packet's 1928 * fields with some pattern, setting 'match' if a 1929 * match is found. At the end of the loop there is 1930 * logic to deal with F_NOT and F_OR flags associated 1931 * with the opcode. 1932 */ 1933 case O_NOP: 1934 match = 1; 1935 break; 1936 1937 case O_FORWARD_MAC: 1938 kprintf("ipfw: opcode %d unimplemented\n", 1939 cmd->opcode); 1940 break; 1941 1942 case O_GID: 1943 case O_UID: 1944 /* 1945 * We only check offset == 0 && proto != 0, 1946 * as this ensures that we have an IPv4 1947 * packet with the ports info. 1948 */ 1949 if (offset!=0) 1950 break; 1951 1952 match = ipfw_match_uid(&args->f_id, oif, 1953 cmd->opcode, 1954 (uid_t)((ipfw_insn_u32 *)cmd)->d[0], 1955 &deny); 1956 if (deny) 1957 return IP_FW_DENY; 1958 break; 1959 1960 case O_RECV: 1961 match = iface_match(m->m_pkthdr.rcvif, 1962 (ipfw_insn_if *)cmd); 1963 break; 1964 1965 case O_XMIT: 1966 match = iface_match(oif, (ipfw_insn_if *)cmd); 1967 break; 1968 1969 case O_VIA: 1970 match = iface_match(oif ? oif : 1971 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd); 1972 break; 1973 1974 case O_MACADDR2: 1975 if (args->eh != NULL) { /* have MAC header */ 1976 uint32_t *want = (uint32_t *) 1977 ((ipfw_insn_mac *)cmd)->addr; 1978 uint32_t *mask = (uint32_t *) 1979 ((ipfw_insn_mac *)cmd)->mask; 1980 uint32_t *hdr = (uint32_t *)args->eh; 1981 1982 match = 1983 (want[0] == (hdr[0] & mask[0]) && 1984 want[1] == (hdr[1] & mask[1]) && 1985 want[2] == (hdr[2] & mask[2])); 1986 } 1987 break; 1988 1989 case O_MAC_TYPE: 1990 if (args->eh != NULL) { 1991 uint16_t t = 1992 ntohs(args->eh->ether_type); 1993 uint16_t *p = 1994 ((ipfw_insn_u16 *)cmd)->ports; 1995 int i; 1996 1997 /* Special vlan handling */ 1998 if (m->m_flags & M_VLANTAG) 1999 t = ETHERTYPE_VLAN; 2000 2001 for (i = cmdlen - 1; !match && i > 0; 2002 i--, p += 2) { 2003 match = 2004 (t >= p[0] && t <= p[1]); 2005 } 2006 } 2007 break; 2008 2009 case O_FRAG: 2010 match = (hlen > 0 && offset != 0); 2011 break; 2012 2013 case O_IN: /* "out" is "not in" */ 2014 match = (oif == NULL); 2015 break; 2016 2017 case O_LAYER2: 2018 match = (args->eh != NULL); 2019 break; 2020 2021 case O_PROTO: 2022 /* 2023 * We do not allow an arg of 0 so the 2024 * check of "proto" only suffices. 2025 */ 2026 match = (proto == cmd->arg1); 2027 break; 2028 2029 case O_IP_SRC: 2030 match = (hlen > 0 && 2031 ((ipfw_insn_ip *)cmd)->addr.s_addr == 2032 src_ip.s_addr); 2033 break; 2034 2035 case O_IP_SRC_MASK: 2036 match = (hlen > 0 && 2037 ((ipfw_insn_ip *)cmd)->addr.s_addr == 2038 (src_ip.s_addr & 2039 ((ipfw_insn_ip *)cmd)->mask.s_addr)); 2040 break; 2041 2042 case O_IP_SRC_ME: 2043 if (hlen > 0) { 2044 struct ifnet *tif; 2045 2046 tif = INADDR_TO_IFP(&src_ip); 2047 match = (tif != NULL); 2048 } 2049 break; 2050 2051 case O_IP_DST_SET: 2052 case O_IP_SRC_SET: 2053 if (hlen > 0) { 2054 uint32_t *d = (uint32_t *)(cmd + 1); 2055 uint32_t addr = 2056 cmd->opcode == O_IP_DST_SET ? 2057 args->f_id.dst_ip : 2058 args->f_id.src_ip; 2059 2060 if (addr < d[0]) 2061 break; 2062 addr -= d[0]; /* subtract base */ 2063 match = 2064 (addr < cmd->arg1) && 2065 (d[1 + (addr >> 5)] & 2066 (1 << (addr & 0x1f))); 2067 } 2068 break; 2069 2070 case O_IP_DST: 2071 match = (hlen > 0 && 2072 ((ipfw_insn_ip *)cmd)->addr.s_addr == 2073 dst_ip.s_addr); 2074 break; 2075 2076 case O_IP_DST_MASK: 2077 match = (hlen > 0) && 2078 (((ipfw_insn_ip *)cmd)->addr.s_addr == 2079 (dst_ip.s_addr & 2080 ((ipfw_insn_ip *)cmd)->mask.s_addr)); 2081 break; 2082 2083 case O_IP_DST_ME: 2084 if (hlen > 0) { 2085 struct ifnet *tif; 2086 2087 tif = INADDR_TO_IFP(&dst_ip); 2088 match = (tif != NULL); 2089 } 2090 break; 2091 2092 case O_IP_SRCPORT: 2093 case O_IP_DSTPORT: 2094 /* 2095 * offset == 0 && proto != 0 is enough 2096 * to guarantee that we have an IPv4 2097 * packet with port info. 2098 */ 2099 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP) 2100 && offset == 0) { 2101 uint16_t x = 2102 (cmd->opcode == O_IP_SRCPORT) ? 2103 src_port : dst_port ; 2104 uint16_t *p = 2105 ((ipfw_insn_u16 *)cmd)->ports; 2106 int i; 2107 2108 for (i = cmdlen - 1; !match && i > 0; 2109 i--, p += 2) { 2110 match = 2111 (x >= p[0] && x <= p[1]); 2112 } 2113 } 2114 break; 2115 2116 case O_ICMPTYPE: 2117 match = (offset == 0 && proto==IPPROTO_ICMP && 2118 icmptype_match(ip, (ipfw_insn_u32 *)cmd)); 2119 break; 2120 2121 case O_IPOPT: 2122 match = (hlen > 0 && ipopts_match(ip, cmd)); 2123 break; 2124 2125 case O_IPVER: 2126 match = (hlen > 0 && cmd->arg1 == ip->ip_v); 2127 break; 2128 2129 case O_IPTTL: 2130 match = (hlen > 0 && cmd->arg1 == ip->ip_ttl); 2131 break; 2132 2133 case O_IPID: 2134 match = (hlen > 0 && 2135 cmd->arg1 == ntohs(ip->ip_id)); 2136 break; 2137 2138 case O_IPLEN: 2139 match = (hlen > 0 && cmd->arg1 == ip_len); 2140 break; 2141 2142 case O_IPPRECEDENCE: 2143 match = (hlen > 0 && 2144 (cmd->arg1 == (ip->ip_tos & 0xe0))); 2145 break; 2146 2147 case O_IPTOS: 2148 match = (hlen > 0 && 2149 flags_match(cmd, ip->ip_tos)); 2150 break; 2151 2152 case O_TCPFLAGS: 2153 match = (proto == IPPROTO_TCP && offset == 0 && 2154 flags_match(cmd, 2155 L3HDR(struct tcphdr,ip)->th_flags)); 2156 break; 2157 2158 case O_TCPOPTS: 2159 match = (proto == IPPROTO_TCP && offset == 0 && 2160 tcpopts_match(ip, cmd)); 2161 break; 2162 2163 case O_TCPSEQ: 2164 match = (proto == IPPROTO_TCP && offset == 0 && 2165 ((ipfw_insn_u32 *)cmd)->d[0] == 2166 L3HDR(struct tcphdr,ip)->th_seq); 2167 break; 2168 2169 case O_TCPACK: 2170 match = (proto == IPPROTO_TCP && offset == 0 && 2171 ((ipfw_insn_u32 *)cmd)->d[0] == 2172 L3HDR(struct tcphdr,ip)->th_ack); 2173 break; 2174 2175 case O_TCPWIN: 2176 match = (proto == IPPROTO_TCP && offset == 0 && 2177 cmd->arg1 == 2178 L3HDR(struct tcphdr,ip)->th_win); 2179 break; 2180 2181 case O_ESTAB: 2182 /* reject packets which have SYN only */ 2183 /* XXX should i also check for TH_ACK ? */ 2184 match = (proto == IPPROTO_TCP && offset == 0 && 2185 (L3HDR(struct tcphdr,ip)->th_flags & 2186 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN); 2187 break; 2188 2189 case O_LOG: 2190 if (fw_verbose) 2191 ipfw_log(f, hlen, args->eh, m, oif); 2192 match = 1; 2193 break; 2194 2195 case O_PROB: 2196 match = (krandom() < 2197 ((ipfw_insn_u32 *)cmd)->d[0]); 2198 break; 2199 2200 /* 2201 * The second set of opcodes represents 'actions', 2202 * i.e. the terminal part of a rule once the packet 2203 * matches all previous patterns. 2204 * Typically there is only one action for each rule, 2205 * and the opcode is stored at the end of the rule 2206 * (but there are exceptions -- see below). 2207 * 2208 * In general, here we set retval and terminate the 2209 * outer loop (would be a 'break 3' in some language, 2210 * but we need to do a 'goto done'). 2211 * 2212 * Exceptions: 2213 * O_COUNT and O_SKIPTO actions: 2214 * instead of terminating, we jump to the next rule 2215 * ('goto next_rule', equivalent to a 'break 2'), 2216 * or to the SKIPTO target ('goto again' after 2217 * having set f, cmd and l), respectively. 2218 * 2219 * O_LIMIT and O_KEEP_STATE: these opcodes are 2220 * not real 'actions', and are stored right 2221 * before the 'action' part of the rule. 2222 * These opcodes try to install an entry in the 2223 * state tables; if successful, we continue with 2224 * the next opcode (match=1; break;), otherwise 2225 * the packet must be dropped ('goto done' after 2226 * setting retval). If static rules are changed 2227 * during the state installation, the packet will 2228 * be dropped and rule's stats will not beupdated 2229 * ('return IP_FW_DENY'). 2230 * 2231 * O_PROBE_STATE and O_CHECK_STATE: these opcodes 2232 * cause a lookup of the state table, and a jump 2233 * to the 'action' part of the parent rule 2234 * ('goto check_body') if an entry is found, or 2235 * (CHECK_STATE only) a jump to the next rule if 2236 * the entry is not found ('goto next_rule'). 2237 * The result of the lookup is cached to make 2238 * further instances of these opcodes are 2239 * effectively NOPs. If static rules are changed 2240 * during the state looking up, the packet will 2241 * be dropped and rule's stats will not be updated 2242 * ('return IP_FW_DENY'). 2243 */ 2244 case O_LIMIT: 2245 case O_KEEP_STATE: 2246 if (!(f->rule_flags & IPFW_RULE_F_STATE)) { 2247 kprintf("%s rule (%d) is not ready " 2248 "on cpu%d\n", 2249 cmd->opcode == O_LIMIT ? 2250 "limit" : "keep state", 2251 f->rulenum, f->cpuid); 2252 goto next_rule; 2253 } 2254 if (install_state(f, 2255 (ipfw_insn_limit *)cmd, args, &deny)) { 2256 if (deny) 2257 return IP_FW_DENY; 2258 2259 retval = IP_FW_DENY; 2260 goto done; /* error/limit violation */ 2261 } 2262 if (deny) 2263 return IP_FW_DENY; 2264 match = 1; 2265 break; 2266 2267 case O_PROBE_STATE: 2268 case O_CHECK_STATE: 2269 /* 2270 * dynamic rules are checked at the first 2271 * keep-state or check-state occurrence, 2272 * with the result being stored in dyn_dir. 2273 * The compiler introduces a PROBE_STATE 2274 * instruction for us when we have a 2275 * KEEP_STATE (because PROBE_STATE needs 2276 * to be run first). 2277 */ 2278 if (dyn_dir == MATCH_UNKNOWN) { 2279 dyn_f = lookup_rule(&args->f_id, 2280 &dyn_dir, 2281 proto == IPPROTO_TCP ? 2282 L3HDR(struct tcphdr, ip) : NULL, 2283 ip_len, &deny); 2284 if (deny) 2285 return IP_FW_DENY; 2286 if (dyn_f != NULL) { 2287 /* 2288 * Found a rule from a dynamic 2289 * entry; jump to the 'action' 2290 * part of the rule. 2291 */ 2292 f = dyn_f; 2293 cmd = ACTION_PTR(f); 2294 l = f->cmd_len - f->act_ofs; 2295 goto check_body; 2296 } 2297 } 2298 /* 2299 * Dynamic entry not found. If CHECK_STATE, 2300 * skip to next rule, if PROBE_STATE just 2301 * ignore and continue with next opcode. 2302 */ 2303 if (cmd->opcode == O_CHECK_STATE) 2304 goto next_rule; 2305 else if (!(f->rule_flags & IPFW_RULE_F_STATE)) 2306 goto next_rule; /* not ready yet */ 2307 match = 1; 2308 break; 2309 2310 case O_ACCEPT: 2311 retval = IP_FW_PASS; /* accept */ 2312 goto done; 2313 2314 case O_PIPE: 2315 case O_QUEUE: 2316 args->rule = f; /* report matching rule */ 2317 args->cookie = cmd->arg1; 2318 retval = IP_FW_DUMMYNET; 2319 goto done; 2320 2321 case O_DIVERT: 2322 case O_TEE: 2323 if (args->eh) /* not on layer 2 */ 2324 break; 2325 2326 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, 2327 sizeof(*divinfo), MB_DONTWAIT); 2328 if (mtag == NULL) { 2329 retval = IP_FW_DENY; 2330 goto done; 2331 } 2332 divinfo = m_tag_data(mtag); 2333 2334 divinfo->skipto = f->rulenum; 2335 divinfo->port = cmd->arg1; 2336 divinfo->tee = (cmd->opcode == O_TEE); 2337 m_tag_prepend(m, mtag); 2338 2339 args->cookie = cmd->arg1; 2340 retval = (cmd->opcode == O_DIVERT) ? 2341 IP_FW_DIVERT : IP_FW_TEE; 2342 goto done; 2343 2344 case O_COUNT: 2345 case O_SKIPTO: 2346 f->pcnt++; /* update stats */ 2347 f->bcnt += ip_len; 2348 f->timestamp = time_second; 2349 if (cmd->opcode == O_COUNT) 2350 goto next_rule; 2351 /* handle skipto */ 2352 if (f->next_rule == NULL) 2353 lookup_next_rule(f); 2354 f = f->next_rule; 2355 goto again; 2356 2357 case O_REJECT: 2358 /* 2359 * Drop the packet and send a reject notice 2360 * if the packet is not ICMP (or is an ICMP 2361 * query), and it is not multicast/broadcast. 2362 */ 2363 if (hlen > 0 && 2364 (proto != IPPROTO_ICMP || 2365 is_icmp_query(ip)) && 2366 !(m->m_flags & (M_BCAST|M_MCAST)) && 2367 !IN_MULTICAST(ntohl(dst_ip.s_addr))) { 2368 /* 2369 * Update statistics before the possible 2370 * blocking 'send_reject' 2371 */ 2372 f->pcnt++; 2373 f->bcnt += ip_len; 2374 f->timestamp = time_second; 2375 2376 send_reject(args, cmd->arg1, 2377 offset,ip_len); 2378 m = args->m; 2379 2380 /* 2381 * Return directly here, rule stats 2382 * have been updated above. 2383 */ 2384 return IP_FW_DENY; 2385 } 2386 /* FALLTHROUGH */ 2387 case O_DENY: 2388 retval = IP_FW_DENY; 2389 goto done; 2390 2391 case O_FORWARD_IP: 2392 if (args->eh) /* not valid on layer2 pkts */ 2393 break; 2394 if (!dyn_f || dyn_dir == MATCH_FORWARD) { 2395 struct sockaddr_in *sin; 2396 2397 mtag = m_tag_get(PACKET_TAG_IPFORWARD, 2398 sizeof(*sin), MB_DONTWAIT); 2399 if (mtag == NULL) { 2400 retval = IP_FW_DENY; 2401 goto done; 2402 } 2403 sin = m_tag_data(mtag); 2404 2405 /* Structure copy */ 2406 *sin = ((ipfw_insn_sa *)cmd)->sa; 2407 2408 m_tag_prepend(m, mtag); 2409 m->m_pkthdr.fw_flags |= 2410 IPFORWARD_MBUF_TAGGED; 2411 m->m_pkthdr.fw_flags &= 2412 ~BRIDGE_MBUF_TAGGED; 2413 } 2414 retval = IP_FW_PASS; 2415 goto done; 2416 2417 default: 2418 panic("-- unknown opcode %d", cmd->opcode); 2419 } /* end of switch() on opcodes */ 2420 2421 if (cmd->len & F_NOT) 2422 match = !match; 2423 2424 if (match) { 2425 if (cmd->len & F_OR) 2426 skip_or = 1; 2427 } else { 2428 if (!(cmd->len & F_OR)) /* not an OR block, */ 2429 break; /* try next rule */ 2430 } 2431 2432 } /* end of inner for, scan opcodes */ 2433 2434 next_rule:; /* try next rule */ 2435 2436 } /* end of outer for, scan rules */ 2437 kprintf("+++ ipfw: ouch!, skip past end of rules, denying packet\n"); 2438 return IP_FW_DENY; 2439 2440 done: 2441 /* Update statistics */ 2442 f->pcnt++; 2443 f->bcnt += ip_len; 2444 f->timestamp = time_second; 2445 return retval; 2446 2447 pullup_failed: 2448 if (fw_verbose) 2449 kprintf("pullup failed\n"); 2450 return IP_FW_DENY; 2451 } 2452 2453 static void 2454 ipfw_dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa) 2455 { 2456 struct m_tag *mtag; 2457 struct dn_pkt *pkt; 2458 ipfw_insn *cmd; 2459 const struct ipfw_flow_id *id; 2460 struct dn_flow_id *fid; 2461 2462 M_ASSERTPKTHDR(m); 2463 2464 mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*pkt), MB_DONTWAIT); 2465 if (mtag == NULL) { 2466 m_freem(m); 2467 return; 2468 } 2469 m_tag_prepend(m, mtag); 2470 2471 pkt = m_tag_data(mtag); 2472 bzero(pkt, sizeof(*pkt)); 2473 2474 cmd = fwa->rule->cmd + fwa->rule->act_ofs; 2475 if (cmd->opcode == O_LOG) 2476 cmd += F_LEN(cmd); 2477 KASSERT(cmd->opcode == O_PIPE || cmd->opcode == O_QUEUE, 2478 ("Rule is not PIPE or QUEUE, opcode %d", cmd->opcode)); 2479 2480 pkt->dn_m = m; 2481 pkt->dn_flags = (dir & DN_FLAGS_DIR_MASK); 2482 pkt->ifp = fwa->oif; 2483 pkt->pipe_nr = pipe_nr; 2484 2485 pkt->cpuid = mycpuid; 2486 pkt->msgport = cur_netport(); 2487 2488 id = &fwa->f_id; 2489 fid = &pkt->id; 2490 fid->fid_dst_ip = id->dst_ip; 2491 fid->fid_src_ip = id->src_ip; 2492 fid->fid_dst_port = id->dst_port; 2493 fid->fid_src_port = id->src_port; 2494 fid->fid_proto = id->proto; 2495 fid->fid_flags = id->flags; 2496 2497 ipfw_ref_rule(fwa->rule); 2498 pkt->dn_priv = fwa->rule; 2499 pkt->dn_unref_priv = ipfw_unref_rule; 2500 2501 if (cmd->opcode == O_PIPE) 2502 pkt->dn_flags |= DN_FLAGS_IS_PIPE; 2503 2504 m->m_pkthdr.fw_flags |= DUMMYNET_MBUF_TAGGED; 2505 } 2506 2507 /* 2508 * When a rule is added/deleted, clear the next_rule pointers in all rules. 2509 * These will be reconstructed on the fly as packets are matched. 2510 * Must be called at splimp(). 2511 */ 2512 static void 2513 ipfw_flush_rule_ptrs(struct ipfw_context *ctx) 2514 { 2515 struct ip_fw *rule; 2516 2517 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) 2518 rule->next_rule = NULL; 2519 } 2520 2521 static __inline void 2522 ipfw_inc_static_count(struct ip_fw *rule) 2523 { 2524 /* Static rule's counts are updated only on CPU0 */ 2525 KKASSERT(mycpuid == 0); 2526 2527 static_count++; 2528 static_ioc_len += IOC_RULESIZE(rule); 2529 } 2530 2531 static __inline void 2532 ipfw_dec_static_count(struct ip_fw *rule) 2533 { 2534 int l = IOC_RULESIZE(rule); 2535 2536 /* Static rule's counts are updated only on CPU0 */ 2537 KKASSERT(mycpuid == 0); 2538 2539 KASSERT(static_count > 0, ("invalid static count %u", static_count)); 2540 static_count--; 2541 2542 KASSERT(static_ioc_len >= l, 2543 ("invalid static len %u", static_ioc_len)); 2544 static_ioc_len -= l; 2545 } 2546 2547 static void 2548 ipfw_link_sibling(struct netmsg_ipfw *fwmsg, struct ip_fw *rule) 2549 { 2550 if (fwmsg->sibling != NULL) { 2551 KKASSERT(mycpuid > 0 && fwmsg->sibling->cpuid == mycpuid - 1); 2552 fwmsg->sibling->sibling = rule; 2553 } 2554 fwmsg->sibling = rule; 2555 } 2556 2557 static struct ip_fw * 2558 ipfw_create_rule(const struct ipfw_ioc_rule *ioc_rule, struct ip_fw_stub *stub) 2559 { 2560 struct ip_fw *rule; 2561 2562 rule = kmalloc(RULESIZE(ioc_rule), M_IPFW, M_WAITOK | M_ZERO); 2563 2564 rule->act_ofs = ioc_rule->act_ofs; 2565 rule->cmd_len = ioc_rule->cmd_len; 2566 rule->rulenum = ioc_rule->rulenum; 2567 rule->set = ioc_rule->set; 2568 rule->usr_flags = ioc_rule->usr_flags; 2569 2570 bcopy(ioc_rule->cmd, rule->cmd, rule->cmd_len * 4 /* XXX */); 2571 2572 rule->refcnt = 1; 2573 rule->cpuid = mycpuid; 2574 2575 rule->stub = stub; 2576 if (stub != NULL) 2577 stub->rule[mycpuid] = rule; 2578 2579 return rule; 2580 } 2581 2582 static void 2583 ipfw_add_rule_dispatch(netmsg_t nmsg) 2584 { 2585 struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg; 2586 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2587 struct ip_fw *rule; 2588 2589 rule = ipfw_create_rule(fwmsg->ioc_rule, fwmsg->stub); 2590 2591 /* 2592 * Bump generation after ipfw_create_rule(), 2593 * since this function is blocking 2594 */ 2595 ctx->ipfw_gen++; 2596 2597 /* 2598 * Insert rule into the pre-determined position 2599 */ 2600 if (fwmsg->prev_rule != NULL) { 2601 struct ip_fw *prev, *next; 2602 2603 prev = fwmsg->prev_rule; 2604 KKASSERT(prev->cpuid == mycpuid); 2605 2606 next = fwmsg->next_rule; 2607 KKASSERT(next->cpuid == mycpuid); 2608 2609 rule->next = next; 2610 prev->next = rule; 2611 2612 /* 2613 * Move to the position on the next CPU 2614 * before the msg is forwarded. 2615 */ 2616 fwmsg->prev_rule = prev->sibling; 2617 fwmsg->next_rule = next->sibling; 2618 } else { 2619 KKASSERT(fwmsg->next_rule == NULL); 2620 rule->next = ctx->ipfw_layer3_chain; 2621 ctx->ipfw_layer3_chain = rule; 2622 } 2623 2624 /* Link rule CPU sibling */ 2625 ipfw_link_sibling(fwmsg, rule); 2626 2627 ipfw_flush_rule_ptrs(ctx); 2628 2629 if (mycpuid == 0) { 2630 /* Statistics only need to be updated once */ 2631 ipfw_inc_static_count(rule); 2632 2633 /* Return the rule on CPU0 */ 2634 nmsg->lmsg.u.ms_resultp = rule; 2635 } 2636 2637 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 2638 } 2639 2640 static void 2641 ipfw_enable_state_dispatch(netmsg_t nmsg) 2642 { 2643 struct lwkt_msg *lmsg = &nmsg->lmsg; 2644 struct ip_fw *rule = lmsg->u.ms_resultp; 2645 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2646 2647 ctx->ipfw_gen++; 2648 2649 KKASSERT(rule->cpuid == mycpuid); 2650 KKASSERT(rule->stub != NULL && rule->stub->rule[mycpuid] == rule); 2651 KKASSERT(!(rule->rule_flags & IPFW_RULE_F_STATE)); 2652 rule->rule_flags |= IPFW_RULE_F_STATE; 2653 lmsg->u.ms_resultp = rule->sibling; 2654 2655 ifnet_forwardmsg(lmsg, mycpuid + 1); 2656 } 2657 2658 /* 2659 * Add a new rule to the list. Copy the rule into a malloc'ed area, 2660 * then possibly create a rule number and add the rule to the list. 2661 * Update the rule_number in the input struct so the caller knows 2662 * it as well. 2663 */ 2664 static void 2665 ipfw_add_rule(struct ipfw_ioc_rule *ioc_rule, uint32_t rule_flags) 2666 { 2667 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2668 struct netmsg_ipfw fwmsg; 2669 struct netmsg_base *nmsg; 2670 struct ip_fw *f, *prev, *rule; 2671 struct ip_fw_stub *stub; 2672 2673 IPFW_ASSERT_CFGPORT(&curthread->td_msgport); 2674 2675 /* 2676 * If rulenum is 0, find highest numbered rule before the 2677 * default rule, and add rule number incremental step. 2678 */ 2679 if (ioc_rule->rulenum == 0) { 2680 int step = autoinc_step; 2681 2682 KKASSERT(step >= IPFW_AUTOINC_STEP_MIN && 2683 step <= IPFW_AUTOINC_STEP_MAX); 2684 2685 /* 2686 * Locate the highest numbered rule before default 2687 */ 2688 for (f = ctx->ipfw_layer3_chain; f; f = f->next) { 2689 if (f->rulenum == IPFW_DEFAULT_RULE) 2690 break; 2691 ioc_rule->rulenum = f->rulenum; 2692 } 2693 if (ioc_rule->rulenum < IPFW_DEFAULT_RULE - step) 2694 ioc_rule->rulenum += step; 2695 } 2696 KASSERT(ioc_rule->rulenum != IPFW_DEFAULT_RULE && 2697 ioc_rule->rulenum != 0, 2698 ("invalid rule num %d", ioc_rule->rulenum)); 2699 2700 /* 2701 * Now find the right place for the new rule in the sorted list. 2702 */ 2703 for (prev = NULL, f = ctx->ipfw_layer3_chain; f; 2704 prev = f, f = f->next) { 2705 if (f->rulenum > ioc_rule->rulenum) { 2706 /* Found the location */ 2707 break; 2708 } 2709 } 2710 KASSERT(f != NULL, ("no default rule?!")); 2711 2712 if (rule_flags & IPFW_RULE_F_STATE) { 2713 int size; 2714 2715 /* 2716 * If the new rule will create states, then allocate 2717 * a rule stub, which will be referenced by states 2718 * (dyn rules) 2719 */ 2720 size = sizeof(*stub) + ((ncpus - 1) * sizeof(struct ip_fw *)); 2721 stub = kmalloc(size, M_IPFW, M_WAITOK | M_ZERO); 2722 } else { 2723 stub = NULL; 2724 } 2725 2726 /* 2727 * Duplicate the rule onto each CPU. 2728 * The rule duplicated on CPU0 will be returned. 2729 */ 2730 bzero(&fwmsg, sizeof(fwmsg)); 2731 nmsg = &fwmsg.base; 2732 netmsg_init(nmsg, NULL, &curthread->td_msgport, 2733 0, ipfw_add_rule_dispatch); 2734 fwmsg.ioc_rule = ioc_rule; 2735 fwmsg.prev_rule = prev; 2736 fwmsg.next_rule = prev == NULL ? NULL : f; 2737 fwmsg.stub = stub; 2738 2739 ifnet_domsg(&nmsg->lmsg, 0); 2740 KKASSERT(fwmsg.prev_rule == NULL && fwmsg.next_rule == NULL); 2741 2742 rule = nmsg->lmsg.u.ms_resultp; 2743 KKASSERT(rule != NULL && rule->cpuid == mycpuid); 2744 2745 if (rule_flags & IPFW_RULE_F_STATE) { 2746 /* 2747 * Turn on state flag, _after_ everything on all 2748 * CPUs have been setup. 2749 */ 2750 bzero(nmsg, sizeof(*nmsg)); 2751 netmsg_init(nmsg, NULL, &curthread->td_msgport, 2752 0, ipfw_enable_state_dispatch); 2753 nmsg->lmsg.u.ms_resultp = rule; 2754 2755 ifnet_domsg(&nmsg->lmsg, 0); 2756 KKASSERT(nmsg->lmsg.u.ms_resultp == NULL); 2757 } 2758 2759 DPRINTF("++ installed rule %d, static count now %d\n", 2760 rule->rulenum, static_count); 2761 } 2762 2763 /** 2764 * Free storage associated with a static rule (including derived 2765 * dynamic rules). 2766 * The caller is in charge of clearing rule pointers to avoid 2767 * dangling pointers. 2768 * @return a pointer to the next entry. 2769 * Arguments are not checked, so they better be correct. 2770 * Must be called at splimp(). 2771 */ 2772 static struct ip_fw * 2773 ipfw_delete_rule(struct ipfw_context *ctx, 2774 struct ip_fw *prev, struct ip_fw *rule) 2775 { 2776 struct ip_fw *n; 2777 struct ip_fw_stub *stub; 2778 2779 ctx->ipfw_gen++; 2780 2781 /* STATE flag should have been cleared before we reach here */ 2782 KKASSERT((rule->rule_flags & IPFW_RULE_F_STATE) == 0); 2783 2784 stub = rule->stub; 2785 n = rule->next; 2786 if (prev == NULL) 2787 ctx->ipfw_layer3_chain = n; 2788 else 2789 prev->next = n; 2790 2791 /* Mark the rule as invalid */ 2792 rule->rule_flags |= IPFW_RULE_F_INVALID; 2793 rule->next_rule = NULL; 2794 rule->sibling = NULL; 2795 rule->stub = NULL; 2796 #ifdef foo 2797 /* Don't reset cpuid here; keep various assertion working */ 2798 rule->cpuid = -1; 2799 #endif 2800 2801 /* Statistics only need to be updated once */ 2802 if (mycpuid == 0) 2803 ipfw_dec_static_count(rule); 2804 2805 /* Free 'stub' on the last CPU */ 2806 if (stub != NULL && mycpuid == ncpus - 1) 2807 kfree(stub, M_IPFW); 2808 2809 /* Try to free this rule */ 2810 ipfw_free_rule(rule); 2811 2812 /* Return the next rule */ 2813 return n; 2814 } 2815 2816 static void 2817 ipfw_flush_dispatch(netmsg_t nmsg) 2818 { 2819 struct lwkt_msg *lmsg = &nmsg->lmsg; 2820 int kill_default = lmsg->u.ms_result; 2821 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2822 struct ip_fw *rule; 2823 2824 ipfw_flush_rule_ptrs(ctx); /* more efficient to do outside the loop */ 2825 2826 while ((rule = ctx->ipfw_layer3_chain) != NULL && 2827 (kill_default || rule->rulenum != IPFW_DEFAULT_RULE)) 2828 ipfw_delete_rule(ctx, NULL, rule); 2829 2830 ifnet_forwardmsg(lmsg, mycpuid + 1); 2831 } 2832 2833 static void 2834 ipfw_disable_rule_state_dispatch(netmsg_t nmsg) 2835 { 2836 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg; 2837 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2838 struct ip_fw *rule; 2839 2840 ctx->ipfw_gen++; 2841 2842 rule = dmsg->start_rule; 2843 if (rule != NULL) { 2844 KKASSERT(rule->cpuid == mycpuid); 2845 2846 /* 2847 * Move to the position on the next CPU 2848 * before the msg is forwarded. 2849 */ 2850 dmsg->start_rule = rule->sibling; 2851 } else { 2852 KKASSERT(dmsg->rulenum == 0); 2853 rule = ctx->ipfw_layer3_chain; 2854 } 2855 2856 while (rule != NULL) { 2857 if (dmsg->rulenum && rule->rulenum != dmsg->rulenum) 2858 break; 2859 rule->rule_flags &= ~IPFW_RULE_F_STATE; 2860 rule = rule->next; 2861 } 2862 2863 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 2864 } 2865 2866 /* 2867 * Deletes all rules from a chain (including the default rule 2868 * if the second argument is set). 2869 * Must be called at splimp(). 2870 */ 2871 static void 2872 ipfw_flush(int kill_default) 2873 { 2874 struct netmsg_del dmsg; 2875 struct netmsg_base nmsg; 2876 struct lwkt_msg *lmsg; 2877 struct ip_fw *rule; 2878 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2879 2880 IPFW_ASSERT_CFGPORT(&curthread->td_msgport); 2881 2882 /* 2883 * If 'kill_default' then caller has done the necessary 2884 * msgport syncing; unnecessary to do it again. 2885 */ 2886 if (!kill_default) { 2887 /* 2888 * Let ipfw_chk() know the rules are going to 2889 * be flushed, so it could jump directly to 2890 * the default rule. 2891 */ 2892 ipfw_flushing = 1; 2893 netmsg_service_sync(); 2894 } 2895 2896 /* 2897 * Clear STATE flag on rules, so no more states (dyn rules) 2898 * will be created. 2899 */ 2900 bzero(&dmsg, sizeof(dmsg)); 2901 netmsg_init(&dmsg.base, NULL, &curthread->td_msgport, 2902 0, ipfw_disable_rule_state_dispatch); 2903 ifnet_domsg(&dmsg.base.lmsg, 0); 2904 2905 /* 2906 * This actually nukes all states (dyn rules) 2907 */ 2908 lockmgr(&dyn_lock, LK_EXCLUSIVE); 2909 for (rule = ctx->ipfw_layer3_chain; rule != NULL; rule = rule->next) { 2910 /* 2911 * Can't check IPFW_RULE_F_STATE here, 2912 * since it has been cleared previously. 2913 * Check 'stub' instead. 2914 */ 2915 if (rule->stub != NULL) { 2916 /* Force removal */ 2917 remove_dyn_rule_locked(rule, NULL); 2918 } 2919 } 2920 lockmgr(&dyn_lock, LK_RELEASE); 2921 2922 /* 2923 * Press the 'flush' button 2924 */ 2925 bzero(&nmsg, sizeof(nmsg)); 2926 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 2927 0, ipfw_flush_dispatch); 2928 lmsg = &nmsg.lmsg; 2929 lmsg->u.ms_result = kill_default; 2930 ifnet_domsg(lmsg, 0); 2931 2932 KASSERT(dyn_count == 0, ("%u dyn rule remains", dyn_count)); 2933 2934 if (kill_default) { 2935 if (ipfw_dyn_v != NULL) { 2936 /* 2937 * Free dynamic rules(state) hash table 2938 */ 2939 kfree(ipfw_dyn_v, M_IPFW); 2940 ipfw_dyn_v = NULL; 2941 } 2942 2943 KASSERT(static_count == 0, 2944 ("%u static rules remain", static_count)); 2945 KASSERT(static_ioc_len == 0, 2946 ("%u bytes of static rules remain", static_ioc_len)); 2947 } else { 2948 KASSERT(static_count == 1, 2949 ("%u static rules remain", static_count)); 2950 KASSERT(static_ioc_len == IOC_RULESIZE(ctx->ipfw_default_rule), 2951 ("%u bytes of static rules remain, should be %lu", 2952 static_ioc_len, 2953 (u_long)IOC_RULESIZE(ctx->ipfw_default_rule))); 2954 } 2955 2956 /* Flush is done */ 2957 ipfw_flushing = 0; 2958 } 2959 2960 static void 2961 ipfw_alt_delete_rule_dispatch(netmsg_t nmsg) 2962 { 2963 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg; 2964 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2965 struct ip_fw *rule, *prev; 2966 2967 rule = dmsg->start_rule; 2968 KKASSERT(rule->cpuid == mycpuid); 2969 dmsg->start_rule = rule->sibling; 2970 2971 prev = dmsg->prev_rule; 2972 if (prev != NULL) { 2973 KKASSERT(prev->cpuid == mycpuid); 2974 2975 /* 2976 * Move to the position on the next CPU 2977 * before the msg is forwarded. 2978 */ 2979 dmsg->prev_rule = prev->sibling; 2980 } 2981 2982 /* 2983 * flush pointers outside the loop, then delete all matching 2984 * rules. 'prev' remains the same throughout the cycle. 2985 */ 2986 ipfw_flush_rule_ptrs(ctx); 2987 while (rule && rule->rulenum == dmsg->rulenum) 2988 rule = ipfw_delete_rule(ctx, prev, rule); 2989 2990 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 2991 } 2992 2993 static int 2994 ipfw_alt_delete_rule(uint16_t rulenum) 2995 { 2996 struct ip_fw *prev, *rule, *f; 2997 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 2998 struct netmsg_del dmsg; 2999 struct netmsg_base *nmsg; 3000 int state; 3001 3002 /* 3003 * Locate first rule to delete 3004 */ 3005 for (prev = NULL, rule = ctx->ipfw_layer3_chain; 3006 rule && rule->rulenum < rulenum; 3007 prev = rule, rule = rule->next) 3008 ; /* EMPTY */ 3009 if (rule->rulenum != rulenum) 3010 return EINVAL; 3011 3012 /* 3013 * Check whether any rules with the given number will 3014 * create states. 3015 */ 3016 state = 0; 3017 for (f = rule; f && f->rulenum == rulenum; f = f->next) { 3018 if (f->rule_flags & IPFW_RULE_F_STATE) { 3019 state = 1; 3020 break; 3021 } 3022 } 3023 3024 if (state) { 3025 /* 3026 * Clear the STATE flag, so no more states will be 3027 * created based the rules numbered 'rulenum'. 3028 */ 3029 bzero(&dmsg, sizeof(dmsg)); 3030 nmsg = &dmsg.base; 3031 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3032 0, ipfw_disable_rule_state_dispatch); 3033 dmsg.start_rule = rule; 3034 dmsg.rulenum = rulenum; 3035 3036 ifnet_domsg(&nmsg->lmsg, 0); 3037 KKASSERT(dmsg.start_rule == NULL); 3038 3039 /* 3040 * Nuke all related states 3041 */ 3042 lockmgr(&dyn_lock, LK_EXCLUSIVE); 3043 for (f = rule; f && f->rulenum == rulenum; f = f->next) { 3044 /* 3045 * Can't check IPFW_RULE_F_STATE here, 3046 * since it has been cleared previously. 3047 * Check 'stub' instead. 3048 */ 3049 if (f->stub != NULL) { 3050 /* Force removal */ 3051 remove_dyn_rule_locked(f, NULL); 3052 } 3053 } 3054 lockmgr(&dyn_lock, LK_RELEASE); 3055 } 3056 3057 /* 3058 * Get rid of the rule duplications on all CPUs 3059 */ 3060 bzero(&dmsg, sizeof(dmsg)); 3061 nmsg = &dmsg.base; 3062 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3063 0, ipfw_alt_delete_rule_dispatch); 3064 dmsg.prev_rule = prev; 3065 dmsg.start_rule = rule; 3066 dmsg.rulenum = rulenum; 3067 3068 ifnet_domsg(&nmsg->lmsg, 0); 3069 KKASSERT(dmsg.prev_rule == NULL && dmsg.start_rule == NULL); 3070 return 0; 3071 } 3072 3073 static void 3074 ipfw_alt_delete_ruleset_dispatch(netmsg_t nmsg) 3075 { 3076 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg; 3077 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3078 struct ip_fw *prev, *rule; 3079 #ifdef INVARIANTS 3080 int del = 0; 3081 #endif 3082 3083 ipfw_flush_rule_ptrs(ctx); 3084 3085 prev = NULL; 3086 rule = ctx->ipfw_layer3_chain; 3087 while (rule != NULL) { 3088 if (rule->set == dmsg->from_set) { 3089 rule = ipfw_delete_rule(ctx, prev, rule); 3090 #ifdef INVARIANTS 3091 del = 1; 3092 #endif 3093 } else { 3094 prev = rule; 3095 rule = rule->next; 3096 } 3097 } 3098 KASSERT(del, ("no match set?!")); 3099 3100 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 3101 } 3102 3103 static void 3104 ipfw_disable_ruleset_state_dispatch(netmsg_t nmsg) 3105 { 3106 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg; 3107 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3108 struct ip_fw *rule; 3109 #ifdef INVARIANTS 3110 int cleared = 0; 3111 #endif 3112 3113 ctx->ipfw_gen++; 3114 3115 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) { 3116 if (rule->set == dmsg->from_set) { 3117 #ifdef INVARIANTS 3118 cleared = 1; 3119 #endif 3120 rule->rule_flags &= ~IPFW_RULE_F_STATE; 3121 } 3122 } 3123 KASSERT(cleared, ("no match set?!")); 3124 3125 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 3126 } 3127 3128 static int 3129 ipfw_alt_delete_ruleset(uint8_t set) 3130 { 3131 struct netmsg_del dmsg; 3132 struct netmsg_base *nmsg; 3133 int state, del; 3134 struct ip_fw *rule; 3135 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3136 3137 /* 3138 * Check whether the 'set' exists. If it exists, 3139 * then check whether any rules within the set will 3140 * try to create states. 3141 */ 3142 state = 0; 3143 del = 0; 3144 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) { 3145 if (rule->set == set) { 3146 del = 1; 3147 if (rule->rule_flags & IPFW_RULE_F_STATE) { 3148 state = 1; 3149 break; 3150 } 3151 } 3152 } 3153 if (!del) 3154 return 0; /* XXX EINVAL? */ 3155 3156 if (state) { 3157 /* 3158 * Clear the STATE flag, so no more states will be 3159 * created based the rules in this set. 3160 */ 3161 bzero(&dmsg, sizeof(dmsg)); 3162 nmsg = &dmsg.base; 3163 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3164 0, ipfw_disable_ruleset_state_dispatch); 3165 dmsg.from_set = set; 3166 3167 ifnet_domsg(&nmsg->lmsg, 0); 3168 3169 /* 3170 * Nuke all related states 3171 */ 3172 lockmgr(&dyn_lock, LK_EXCLUSIVE); 3173 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) { 3174 if (rule->set != set) 3175 continue; 3176 3177 /* 3178 * Can't check IPFW_RULE_F_STATE here, 3179 * since it has been cleared previously. 3180 * Check 'stub' instead. 3181 */ 3182 if (rule->stub != NULL) { 3183 /* Force removal */ 3184 remove_dyn_rule_locked(rule, NULL); 3185 } 3186 } 3187 lockmgr(&dyn_lock, LK_RELEASE); 3188 } 3189 3190 /* 3191 * Delete this set 3192 */ 3193 bzero(&dmsg, sizeof(dmsg)); 3194 nmsg = &dmsg.base; 3195 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3196 0, ipfw_alt_delete_ruleset_dispatch); 3197 dmsg.from_set = set; 3198 3199 ifnet_domsg(&nmsg->lmsg, 0); 3200 return 0; 3201 } 3202 3203 static void 3204 ipfw_alt_move_rule_dispatch(netmsg_t nmsg) 3205 { 3206 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg; 3207 struct ip_fw *rule; 3208 3209 rule = dmsg->start_rule; 3210 KKASSERT(rule->cpuid == mycpuid); 3211 3212 /* 3213 * Move to the position on the next CPU 3214 * before the msg is forwarded. 3215 */ 3216 dmsg->start_rule = rule->sibling; 3217 3218 while (rule && rule->rulenum <= dmsg->rulenum) { 3219 if (rule->rulenum == dmsg->rulenum) 3220 rule->set = dmsg->to_set; 3221 rule = rule->next; 3222 } 3223 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 3224 } 3225 3226 static int 3227 ipfw_alt_move_rule(uint16_t rulenum, uint8_t set) 3228 { 3229 struct netmsg_del dmsg; 3230 struct netmsg_base *nmsg; 3231 struct ip_fw *rule; 3232 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3233 3234 /* 3235 * Locate first rule to move 3236 */ 3237 for (rule = ctx->ipfw_layer3_chain; rule && rule->rulenum <= rulenum; 3238 rule = rule->next) { 3239 if (rule->rulenum == rulenum && rule->set != set) 3240 break; 3241 } 3242 if (rule == NULL || rule->rulenum > rulenum) 3243 return 0; /* XXX error? */ 3244 3245 bzero(&dmsg, sizeof(dmsg)); 3246 nmsg = &dmsg.base; 3247 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3248 0, ipfw_alt_move_rule_dispatch); 3249 dmsg.start_rule = rule; 3250 dmsg.rulenum = rulenum; 3251 dmsg.to_set = set; 3252 3253 ifnet_domsg(&nmsg->lmsg, 0); 3254 KKASSERT(dmsg.start_rule == NULL); 3255 return 0; 3256 } 3257 3258 static void 3259 ipfw_alt_move_ruleset_dispatch(netmsg_t nmsg) 3260 { 3261 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg; 3262 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3263 struct ip_fw *rule; 3264 3265 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) { 3266 if (rule->set == dmsg->from_set) 3267 rule->set = dmsg->to_set; 3268 } 3269 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 3270 } 3271 3272 static int 3273 ipfw_alt_move_ruleset(uint8_t from_set, uint8_t to_set) 3274 { 3275 struct netmsg_del dmsg; 3276 struct netmsg_base *nmsg; 3277 3278 bzero(&dmsg, sizeof(dmsg)); 3279 nmsg = &dmsg.base; 3280 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3281 0, ipfw_alt_move_ruleset_dispatch); 3282 dmsg.from_set = from_set; 3283 dmsg.to_set = to_set; 3284 3285 ifnet_domsg(&nmsg->lmsg, 0); 3286 return 0; 3287 } 3288 3289 static void 3290 ipfw_alt_swap_ruleset_dispatch(netmsg_t nmsg) 3291 { 3292 struct netmsg_del *dmsg = (struct netmsg_del *)nmsg; 3293 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3294 struct ip_fw *rule; 3295 3296 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) { 3297 if (rule->set == dmsg->from_set) 3298 rule->set = dmsg->to_set; 3299 else if (rule->set == dmsg->to_set) 3300 rule->set = dmsg->from_set; 3301 } 3302 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 3303 } 3304 3305 static int 3306 ipfw_alt_swap_ruleset(uint8_t set1, uint8_t set2) 3307 { 3308 struct netmsg_del dmsg; 3309 struct netmsg_base *nmsg; 3310 3311 bzero(&dmsg, sizeof(dmsg)); 3312 nmsg = &dmsg.base; 3313 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3314 0, ipfw_alt_swap_ruleset_dispatch); 3315 dmsg.from_set = set1; 3316 dmsg.to_set = set2; 3317 3318 ifnet_domsg(&nmsg->lmsg, 0); 3319 return 0; 3320 } 3321 3322 /** 3323 * Remove all rules with given number, and also do set manipulation. 3324 * 3325 * The argument is an uint32_t. The low 16 bit are the rule or set number, 3326 * the next 8 bits are the new set, the top 8 bits are the command: 3327 * 3328 * 0 delete rules with given number 3329 * 1 delete rules with given set number 3330 * 2 move rules with given number to new set 3331 * 3 move rules with given set number to new set 3332 * 4 swap sets with given numbers 3333 */ 3334 static int 3335 ipfw_ctl_alter(uint32_t arg) 3336 { 3337 uint16_t rulenum; 3338 uint8_t cmd, new_set; 3339 int error = 0; 3340 3341 rulenum = arg & 0xffff; 3342 cmd = (arg >> 24) & 0xff; 3343 new_set = (arg >> 16) & 0xff; 3344 3345 if (cmd > 4) 3346 return EINVAL; 3347 if (new_set >= IPFW_DEFAULT_SET) 3348 return EINVAL; 3349 if (cmd == 0 || cmd == 2) { 3350 if (rulenum == IPFW_DEFAULT_RULE) 3351 return EINVAL; 3352 } else { 3353 if (rulenum >= IPFW_DEFAULT_SET) 3354 return EINVAL; 3355 } 3356 3357 switch (cmd) { 3358 case 0: /* delete rules with given number */ 3359 error = ipfw_alt_delete_rule(rulenum); 3360 break; 3361 3362 case 1: /* delete all rules with given set number */ 3363 error = ipfw_alt_delete_ruleset(rulenum); 3364 break; 3365 3366 case 2: /* move rules with given number to new set */ 3367 error = ipfw_alt_move_rule(rulenum, new_set); 3368 break; 3369 3370 case 3: /* move rules with given set number to new set */ 3371 error = ipfw_alt_move_ruleset(rulenum, new_set); 3372 break; 3373 3374 case 4: /* swap two sets */ 3375 error = ipfw_alt_swap_ruleset(rulenum, new_set); 3376 break; 3377 } 3378 return error; 3379 } 3380 3381 /* 3382 * Clear counters for a specific rule. 3383 */ 3384 static void 3385 clear_counters(struct ip_fw *rule, int log_only) 3386 { 3387 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule); 3388 3389 if (log_only == 0) { 3390 rule->bcnt = rule->pcnt = 0; 3391 rule->timestamp = 0; 3392 } 3393 if (l->o.opcode == O_LOG) 3394 l->log_left = l->max_log; 3395 } 3396 3397 static void 3398 ipfw_zero_entry_dispatch(netmsg_t nmsg) 3399 { 3400 struct netmsg_zent *zmsg = (struct netmsg_zent *)nmsg; 3401 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3402 struct ip_fw *rule; 3403 3404 if (zmsg->rulenum == 0) { 3405 KKASSERT(zmsg->start_rule == NULL); 3406 3407 ctx->ipfw_norule_counter = 0; 3408 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) 3409 clear_counters(rule, zmsg->log_only); 3410 } else { 3411 struct ip_fw *start = zmsg->start_rule; 3412 3413 KKASSERT(start->cpuid == mycpuid); 3414 KKASSERT(start->rulenum == zmsg->rulenum); 3415 3416 /* 3417 * We can have multiple rules with the same number, so we 3418 * need to clear them all. 3419 */ 3420 for (rule = start; rule && rule->rulenum == zmsg->rulenum; 3421 rule = rule->next) 3422 clear_counters(rule, zmsg->log_only); 3423 3424 /* 3425 * Move to the position on the next CPU 3426 * before the msg is forwarded. 3427 */ 3428 zmsg->start_rule = start->sibling; 3429 } 3430 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 3431 } 3432 3433 /** 3434 * Reset some or all counters on firewall rules. 3435 * @arg frwl is null to clear all entries, or contains a specific 3436 * rule number. 3437 * @arg log_only is 1 if we only want to reset logs, zero otherwise. 3438 */ 3439 static int 3440 ipfw_ctl_zero_entry(int rulenum, int log_only) 3441 { 3442 struct netmsg_zent zmsg; 3443 struct netmsg_base *nmsg; 3444 const char *msg; 3445 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3446 3447 bzero(&zmsg, sizeof(zmsg)); 3448 nmsg = &zmsg.base; 3449 netmsg_init(nmsg, NULL, &curthread->td_msgport, 3450 0, ipfw_zero_entry_dispatch); 3451 zmsg.log_only = log_only; 3452 3453 if (rulenum == 0) { 3454 msg = log_only ? "ipfw: All logging counts reset.\n" 3455 : "ipfw: Accounting cleared.\n"; 3456 } else { 3457 struct ip_fw *rule; 3458 3459 /* 3460 * Locate the first rule with 'rulenum' 3461 */ 3462 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) { 3463 if (rule->rulenum == rulenum) 3464 break; 3465 } 3466 if (rule == NULL) /* we did not find any matching rules */ 3467 return (EINVAL); 3468 zmsg.start_rule = rule; 3469 zmsg.rulenum = rulenum; 3470 3471 msg = log_only ? "ipfw: Entry %d logging count reset.\n" 3472 : "ipfw: Entry %d cleared.\n"; 3473 } 3474 ifnet_domsg(&nmsg->lmsg, 0); 3475 KKASSERT(zmsg.start_rule == NULL); 3476 3477 if (fw_verbose) 3478 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum); 3479 return (0); 3480 } 3481 3482 /* 3483 * Check validity of the structure before insert. 3484 * Fortunately rules are simple, so this mostly need to check rule sizes. 3485 */ 3486 static int 3487 ipfw_check_ioc_rule(struct ipfw_ioc_rule *rule, int size, uint32_t *rule_flags) 3488 { 3489 int l, cmdlen = 0; 3490 int have_action = 0; 3491 ipfw_insn *cmd; 3492 3493 *rule_flags = 0; 3494 3495 /* Check for valid size */ 3496 if (size < sizeof(*rule)) { 3497 kprintf("ipfw: rule too short\n"); 3498 return EINVAL; 3499 } 3500 l = IOC_RULESIZE(rule); 3501 if (l != size) { 3502 kprintf("ipfw: size mismatch (have %d want %d)\n", size, l); 3503 return EINVAL; 3504 } 3505 3506 /* Check rule number */ 3507 if (rule->rulenum == IPFW_DEFAULT_RULE) { 3508 kprintf("ipfw: invalid rule number\n"); 3509 return EINVAL; 3510 } 3511 3512 /* 3513 * Now go for the individual checks. Very simple ones, basically only 3514 * instruction sizes. 3515 */ 3516 for (l = rule->cmd_len, cmd = rule->cmd; l > 0; 3517 l -= cmdlen, cmd += cmdlen) { 3518 cmdlen = F_LEN(cmd); 3519 if (cmdlen > l) { 3520 kprintf("ipfw: opcode %d size truncated\n", 3521 cmd->opcode); 3522 return EINVAL; 3523 } 3524 3525 DPRINTF("ipfw: opcode %d\n", cmd->opcode); 3526 3527 if (cmd->opcode == O_KEEP_STATE || cmd->opcode == O_LIMIT) { 3528 /* This rule will create states */ 3529 *rule_flags |= IPFW_RULE_F_STATE; 3530 } 3531 3532 switch (cmd->opcode) { 3533 case O_NOP: 3534 case O_PROBE_STATE: 3535 case O_KEEP_STATE: 3536 case O_PROTO: 3537 case O_IP_SRC_ME: 3538 case O_IP_DST_ME: 3539 case O_LAYER2: 3540 case O_IN: 3541 case O_FRAG: 3542 case O_IPOPT: 3543 case O_IPLEN: 3544 case O_IPID: 3545 case O_IPTOS: 3546 case O_IPPRECEDENCE: 3547 case O_IPTTL: 3548 case O_IPVER: 3549 case O_TCPWIN: 3550 case O_TCPFLAGS: 3551 case O_TCPOPTS: 3552 case O_ESTAB: 3553 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 3554 goto bad_size; 3555 break; 3556 3557 case O_UID: 3558 case O_GID: 3559 case O_IP_SRC: 3560 case O_IP_DST: 3561 case O_TCPSEQ: 3562 case O_TCPACK: 3563 case O_PROB: 3564 case O_ICMPTYPE: 3565 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32)) 3566 goto bad_size; 3567 break; 3568 3569 case O_LIMIT: 3570 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit)) 3571 goto bad_size; 3572 break; 3573 3574 case O_LOG: 3575 if (cmdlen != F_INSN_SIZE(ipfw_insn_log)) 3576 goto bad_size; 3577 3578 ((ipfw_insn_log *)cmd)->log_left = 3579 ((ipfw_insn_log *)cmd)->max_log; 3580 3581 break; 3582 3583 case O_IP_SRC_MASK: 3584 case O_IP_DST_MASK: 3585 if (cmdlen != F_INSN_SIZE(ipfw_insn_ip)) 3586 goto bad_size; 3587 if (((ipfw_insn_ip *)cmd)->mask.s_addr == 0) { 3588 kprintf("ipfw: opcode %d, useless rule\n", 3589 cmd->opcode); 3590 return EINVAL; 3591 } 3592 break; 3593 3594 case O_IP_SRC_SET: 3595 case O_IP_DST_SET: 3596 if (cmd->arg1 == 0 || cmd->arg1 > 256) { 3597 kprintf("ipfw: invalid set size %d\n", 3598 cmd->arg1); 3599 return EINVAL; 3600 } 3601 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 3602 (cmd->arg1+31)/32 ) 3603 goto bad_size; 3604 break; 3605 3606 case O_MACADDR2: 3607 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac)) 3608 goto bad_size; 3609 break; 3610 3611 case O_MAC_TYPE: 3612 case O_IP_SRCPORT: 3613 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */ 3614 if (cmdlen < 2 || cmdlen > 31) 3615 goto bad_size; 3616 break; 3617 3618 case O_RECV: 3619 case O_XMIT: 3620 case O_VIA: 3621 if (cmdlen != F_INSN_SIZE(ipfw_insn_if)) 3622 goto bad_size; 3623 break; 3624 3625 case O_PIPE: 3626 case O_QUEUE: 3627 if (cmdlen != F_INSN_SIZE(ipfw_insn_pipe)) 3628 goto bad_size; 3629 goto check_action; 3630 3631 case O_FORWARD_IP: 3632 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) { 3633 goto bad_size; 3634 } else { 3635 in_addr_t fwd_addr; 3636 3637 fwd_addr = ((ipfw_insn_sa *)cmd)-> 3638 sa.sin_addr.s_addr; 3639 if (IN_MULTICAST(ntohl(fwd_addr))) { 3640 kprintf("ipfw: try forwarding to " 3641 "multicast address\n"); 3642 return EINVAL; 3643 } 3644 } 3645 goto check_action; 3646 3647 case O_FORWARD_MAC: /* XXX not implemented yet */ 3648 case O_CHECK_STATE: 3649 case O_COUNT: 3650 case O_ACCEPT: 3651 case O_DENY: 3652 case O_REJECT: 3653 case O_SKIPTO: 3654 case O_DIVERT: 3655 case O_TEE: 3656 if (cmdlen != F_INSN_SIZE(ipfw_insn)) 3657 goto bad_size; 3658 check_action: 3659 if (have_action) { 3660 kprintf("ipfw: opcode %d, multiple actions" 3661 " not allowed\n", 3662 cmd->opcode); 3663 return EINVAL; 3664 } 3665 have_action = 1; 3666 if (l != cmdlen) { 3667 kprintf("ipfw: opcode %d, action must be" 3668 " last opcode\n", 3669 cmd->opcode); 3670 return EINVAL; 3671 } 3672 break; 3673 default: 3674 kprintf("ipfw: opcode %d, unknown opcode\n", 3675 cmd->opcode); 3676 return EINVAL; 3677 } 3678 } 3679 if (have_action == 0) { 3680 kprintf("ipfw: missing action\n"); 3681 return EINVAL; 3682 } 3683 return 0; 3684 3685 bad_size: 3686 kprintf("ipfw: opcode %d size %d wrong\n", 3687 cmd->opcode, cmdlen); 3688 return EINVAL; 3689 } 3690 3691 static int 3692 ipfw_ctl_add_rule(struct sockopt *sopt) 3693 { 3694 struct ipfw_ioc_rule *ioc_rule; 3695 size_t size; 3696 uint32_t rule_flags; 3697 int error; 3698 3699 size = sopt->sopt_valsize; 3700 if (size > (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX) || 3701 size < sizeof(*ioc_rule)) { 3702 return EINVAL; 3703 } 3704 if (size != (sizeof(uint32_t) * IPFW_RULE_SIZE_MAX)) { 3705 sopt->sopt_val = krealloc(sopt->sopt_val, sizeof(uint32_t) * 3706 IPFW_RULE_SIZE_MAX, M_TEMP, M_WAITOK); 3707 } 3708 ioc_rule = sopt->sopt_val; 3709 3710 error = ipfw_check_ioc_rule(ioc_rule, size, &rule_flags); 3711 if (error) 3712 return error; 3713 3714 ipfw_add_rule(ioc_rule, rule_flags); 3715 3716 if (sopt->sopt_dir == SOPT_GET) 3717 sopt->sopt_valsize = IOC_RULESIZE(ioc_rule); 3718 return 0; 3719 } 3720 3721 static void * 3722 ipfw_copy_rule(const struct ip_fw *rule, struct ipfw_ioc_rule *ioc_rule) 3723 { 3724 const struct ip_fw *sibling; 3725 #ifdef INVARIANTS 3726 int i; 3727 #endif 3728 3729 KKASSERT(rule->cpuid == IPFW_CFGCPUID); 3730 3731 ioc_rule->act_ofs = rule->act_ofs; 3732 ioc_rule->cmd_len = rule->cmd_len; 3733 ioc_rule->rulenum = rule->rulenum; 3734 ioc_rule->set = rule->set; 3735 ioc_rule->usr_flags = rule->usr_flags; 3736 3737 ioc_rule->set_disable = ipfw_ctx[mycpuid]->ipfw_set_disable; 3738 ioc_rule->static_count = static_count; 3739 ioc_rule->static_len = static_ioc_len; 3740 3741 /* 3742 * Visit (read-only) all of the rule's duplications to get 3743 * the necessary statistics 3744 */ 3745 #ifdef INVARIANTS 3746 i = 0; 3747 #endif 3748 ioc_rule->pcnt = 0; 3749 ioc_rule->bcnt = 0; 3750 ioc_rule->timestamp = 0; 3751 for (sibling = rule; sibling != NULL; sibling = sibling->sibling) { 3752 ioc_rule->pcnt += sibling->pcnt; 3753 ioc_rule->bcnt += sibling->bcnt; 3754 if (sibling->timestamp > ioc_rule->timestamp) 3755 ioc_rule->timestamp = sibling->timestamp; 3756 #ifdef INVARIANTS 3757 ++i; 3758 #endif 3759 } 3760 KASSERT(i == ncpus, ("static rule is not duplicated on every cpu")); 3761 3762 bcopy(rule->cmd, ioc_rule->cmd, ioc_rule->cmd_len * 4 /* XXX */); 3763 3764 return ((uint8_t *)ioc_rule + IOC_RULESIZE(ioc_rule)); 3765 } 3766 3767 static void 3768 ipfw_copy_state(const ipfw_dyn_rule *dyn_rule, 3769 struct ipfw_ioc_state *ioc_state) 3770 { 3771 const struct ipfw_flow_id *id; 3772 struct ipfw_ioc_flowid *ioc_id; 3773 3774 ioc_state->expire = TIME_LEQ(dyn_rule->expire, time_second) ? 3775 0 : dyn_rule->expire - time_second; 3776 ioc_state->pcnt = dyn_rule->pcnt; 3777 ioc_state->bcnt = dyn_rule->bcnt; 3778 3779 ioc_state->dyn_type = dyn_rule->dyn_type; 3780 ioc_state->count = dyn_rule->count; 3781 3782 ioc_state->rulenum = dyn_rule->stub->rule[mycpuid]->rulenum; 3783 3784 id = &dyn_rule->id; 3785 ioc_id = &ioc_state->id; 3786 3787 ioc_id->type = ETHERTYPE_IP; 3788 ioc_id->u.ip.dst_ip = id->dst_ip; 3789 ioc_id->u.ip.src_ip = id->src_ip; 3790 ioc_id->u.ip.dst_port = id->dst_port; 3791 ioc_id->u.ip.src_port = id->src_port; 3792 ioc_id->u.ip.proto = id->proto; 3793 } 3794 3795 static int 3796 ipfw_ctl_get_rules(struct sockopt *sopt) 3797 { 3798 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3799 struct ip_fw *rule; 3800 void *bp; 3801 size_t size; 3802 uint32_t dcount = 0; 3803 3804 /* 3805 * pass up a copy of the current rules. Static rules 3806 * come first (the last of which has number IPFW_DEFAULT_RULE), 3807 * followed by a possibly empty list of dynamic rule. 3808 */ 3809 3810 size = static_ioc_len; /* size of static rules */ 3811 if (ipfw_dyn_v) { /* add size of dyn.rules */ 3812 dcount = dyn_count; 3813 size += dcount * sizeof(struct ipfw_ioc_state); 3814 } 3815 3816 if (sopt->sopt_valsize < size) { 3817 /* short length, no need to return incomplete rules */ 3818 /* XXX: if superuser, no need to zero buffer */ 3819 bzero(sopt->sopt_val, sopt->sopt_valsize); 3820 return 0; 3821 } 3822 bp = sopt->sopt_val; 3823 3824 for (rule = ctx->ipfw_layer3_chain; rule; rule = rule->next) 3825 bp = ipfw_copy_rule(rule, bp); 3826 3827 if (ipfw_dyn_v && dcount != 0) { 3828 struct ipfw_ioc_state *ioc_state = bp; 3829 uint32_t dcount2 = 0; 3830 #ifdef INVARIANTS 3831 size_t old_size = size; 3832 #endif 3833 int i; 3834 3835 lockmgr(&dyn_lock, LK_SHARED); 3836 3837 /* Check 'ipfw_dyn_v' again with lock held */ 3838 if (ipfw_dyn_v == NULL) 3839 goto skip; 3840 3841 for (i = 0; i < curr_dyn_buckets; i++) { 3842 ipfw_dyn_rule *p; 3843 3844 /* 3845 * The # of dynamic rules may have grown after the 3846 * snapshot of 'dyn_count' was taken, so we will have 3847 * to check 'dcount' (snapshot of dyn_count) here to 3848 * make sure that we don't overflow the pre-allocated 3849 * buffer. 3850 */ 3851 for (p = ipfw_dyn_v[i]; p != NULL && dcount != 0; 3852 p = p->next, ioc_state++, dcount--, dcount2++) 3853 ipfw_copy_state(p, ioc_state); 3854 } 3855 skip: 3856 lockmgr(&dyn_lock, LK_RELEASE); 3857 3858 /* 3859 * The # of dynamic rules may be shrinked after the 3860 * snapshot of 'dyn_count' was taken. To give user a 3861 * correct dynamic rule count, we use the 'dcount2' 3862 * calculated above (with shared lockmgr lock held). 3863 */ 3864 size = static_ioc_len + 3865 (dcount2 * sizeof(struct ipfw_ioc_state)); 3866 KKASSERT(size <= old_size); 3867 } 3868 3869 sopt->sopt_valsize = size; 3870 return 0; 3871 } 3872 3873 static void 3874 ipfw_set_disable_dispatch(netmsg_t nmsg) 3875 { 3876 struct lwkt_msg *lmsg = &nmsg->lmsg; 3877 struct ipfw_context *ctx = ipfw_ctx[mycpuid]; 3878 3879 ctx->ipfw_gen++; 3880 ctx->ipfw_set_disable = lmsg->u.ms_result32; 3881 3882 ifnet_forwardmsg(lmsg, mycpuid + 1); 3883 } 3884 3885 static void 3886 ipfw_ctl_set_disable(uint32_t disable, uint32_t enable) 3887 { 3888 struct netmsg_base nmsg; 3889 struct lwkt_msg *lmsg; 3890 uint32_t set_disable; 3891 3892 /* IPFW_DEFAULT_SET is always enabled */ 3893 enable |= (1 << IPFW_DEFAULT_SET); 3894 set_disable = (ipfw_ctx[mycpuid]->ipfw_set_disable | disable) & ~enable; 3895 3896 bzero(&nmsg, sizeof(nmsg)); 3897 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 3898 0, ipfw_set_disable_dispatch); 3899 lmsg = &nmsg.lmsg; 3900 lmsg->u.ms_result32 = set_disable; 3901 3902 ifnet_domsg(lmsg, 0); 3903 } 3904 3905 /** 3906 * {set|get}sockopt parser. 3907 */ 3908 static int 3909 ipfw_ctl(struct sockopt *sopt) 3910 { 3911 int error, rulenum; 3912 uint32_t *masks; 3913 size_t size; 3914 3915 error = 0; 3916 3917 switch (sopt->sopt_name) { 3918 case IP_FW_GET: 3919 error = ipfw_ctl_get_rules(sopt); 3920 break; 3921 3922 case IP_FW_FLUSH: 3923 ipfw_flush(0 /* keep default rule */); 3924 break; 3925 3926 case IP_FW_ADD: 3927 error = ipfw_ctl_add_rule(sopt); 3928 break; 3929 3930 case IP_FW_DEL: 3931 /* 3932 * IP_FW_DEL is used for deleting single rules or sets, 3933 * and (ab)used to atomically manipulate sets. 3934 * Argument size is used to distinguish between the two: 3935 * sizeof(uint32_t) 3936 * delete single rule or set of rules, 3937 * or reassign rules (or sets) to a different set. 3938 * 2 * sizeof(uint32_t) 3939 * atomic disable/enable sets. 3940 * first uint32_t contains sets to be disabled, 3941 * second uint32_t contains sets to be enabled. 3942 */ 3943 masks = sopt->sopt_val; 3944 size = sopt->sopt_valsize; 3945 if (size == sizeof(*masks)) { 3946 /* 3947 * Delete or reassign static rule 3948 */ 3949 error = ipfw_ctl_alter(masks[0]); 3950 } else if (size == (2 * sizeof(*masks))) { 3951 /* 3952 * Set enable/disable 3953 */ 3954 ipfw_ctl_set_disable(masks[0], masks[1]); 3955 } else { 3956 error = EINVAL; 3957 } 3958 break; 3959 3960 case IP_FW_ZERO: 3961 case IP_FW_RESETLOG: /* argument is an int, the rule number */ 3962 rulenum = 0; 3963 3964 if (sopt->sopt_val != 0) { 3965 error = soopt_to_kbuf(sopt, &rulenum, 3966 sizeof(int), sizeof(int)); 3967 if (error) 3968 break; 3969 } 3970 error = ipfw_ctl_zero_entry(rulenum, 3971 sopt->sopt_name == IP_FW_RESETLOG); 3972 break; 3973 3974 default: 3975 kprintf("ipfw_ctl invalid option %d\n", sopt->sopt_name); 3976 error = EINVAL; 3977 } 3978 return error; 3979 } 3980 3981 /* 3982 * This procedure is only used to handle keepalives. It is invoked 3983 * every dyn_keepalive_period 3984 */ 3985 static void 3986 ipfw_tick_dispatch(netmsg_t nmsg) 3987 { 3988 time_t keep_alive; 3989 uint32_t gen; 3990 int i; 3991 3992 IPFW_ASSERT_CFGPORT(&curthread->td_msgport); 3993 KKASSERT(IPFW_LOADED); 3994 3995 /* Reply ASAP */ 3996 crit_enter(); 3997 lwkt_replymsg(&nmsg->lmsg, 0); 3998 crit_exit(); 3999 4000 if (ipfw_dyn_v == NULL || dyn_count == 0) 4001 goto done; 4002 4003 keep_alive = time_second; 4004 4005 lockmgr(&dyn_lock, LK_EXCLUSIVE); 4006 again: 4007 if (ipfw_dyn_v == NULL || dyn_count == 0) { 4008 lockmgr(&dyn_lock, LK_RELEASE); 4009 goto done; 4010 } 4011 gen = dyn_buckets_gen; 4012 4013 for (i = 0; i < curr_dyn_buckets; i++) { 4014 ipfw_dyn_rule *q, *prev; 4015 4016 for (prev = NULL, q = ipfw_dyn_v[i]; q != NULL;) { 4017 uint32_t ack_rev, ack_fwd; 4018 struct ipfw_flow_id id; 4019 4020 if (q->dyn_type == O_LIMIT_PARENT) 4021 goto next; 4022 4023 if (TIME_LEQ(q->expire, time_second)) { 4024 /* State expired */ 4025 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q); 4026 continue; 4027 } 4028 4029 /* 4030 * Keep alive processing 4031 */ 4032 4033 if (!dyn_keepalive) 4034 goto next; 4035 if (q->id.proto != IPPROTO_TCP) 4036 goto next; 4037 if ((q->state & BOTH_SYN) != BOTH_SYN) 4038 goto next; 4039 if (TIME_LEQ(time_second + dyn_keepalive_interval, 4040 q->expire)) 4041 goto next; /* too early */ 4042 if (q->keep_alive == keep_alive) 4043 goto next; /* alreay done */ 4044 4045 /* 4046 * Save necessary information, so that they could 4047 * survive after possible blocking in send_pkt() 4048 */ 4049 id = q->id; 4050 ack_rev = q->ack_rev; 4051 ack_fwd = q->ack_fwd; 4052 4053 /* Sending has been started */ 4054 q->keep_alive = keep_alive; 4055 4056 /* Release lock to avoid possible dead lock */ 4057 lockmgr(&dyn_lock, LK_RELEASE); 4058 send_pkt(&id, ack_rev - 1, ack_fwd, TH_SYN); 4059 send_pkt(&id, ack_fwd - 1, ack_rev, 0); 4060 lockmgr(&dyn_lock, LK_EXCLUSIVE); 4061 4062 if (gen != dyn_buckets_gen) { 4063 /* 4064 * Dyn bucket array has been changed during 4065 * the above two sending; reiterate. 4066 */ 4067 goto again; 4068 } 4069 next: 4070 prev = q; 4071 q = q->next; 4072 } 4073 } 4074 lockmgr(&dyn_lock, LK_RELEASE); 4075 done: 4076 callout_reset(&ipfw_timeout_h, dyn_keepalive_period * hz, 4077 ipfw_tick, NULL); 4078 } 4079 4080 /* 4081 * This procedure is only used to handle keepalives. It is invoked 4082 * every dyn_keepalive_period 4083 */ 4084 static void 4085 ipfw_tick(void *dummy __unused) 4086 { 4087 struct lwkt_msg *lmsg = &ipfw_timeout_netmsg.lmsg; 4088 4089 KKASSERT(mycpuid == IPFW_CFGCPUID); 4090 4091 crit_enter(); 4092 4093 KKASSERT(lmsg->ms_flags & MSGF_DONE); 4094 if (IPFW_LOADED) { 4095 lwkt_sendmsg(IPFW_CFGPORT, lmsg); 4096 /* ipfw_timeout_netmsg's handler reset this callout */ 4097 } 4098 4099 crit_exit(); 4100 } 4101 4102 static int 4103 ipfw_check_in(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir) 4104 { 4105 struct ip_fw_args args; 4106 struct mbuf *m = *m0; 4107 struct m_tag *mtag; 4108 int tee = 0, error = 0, ret; 4109 4110 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 4111 /* Extract info from dummynet tag */ 4112 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 4113 KKASSERT(mtag != NULL); 4114 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 4115 KKASSERT(args.rule != NULL); 4116 4117 m_tag_delete(m, mtag); 4118 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 4119 } else { 4120 args.rule = NULL; 4121 } 4122 4123 args.eh = NULL; 4124 args.oif = NULL; 4125 args.m = m; 4126 ret = ipfw_chk(&args); 4127 m = args.m; 4128 4129 if (m == NULL) { 4130 error = EACCES; 4131 goto back; 4132 } 4133 4134 switch (ret) { 4135 case IP_FW_PASS: 4136 break; 4137 4138 case IP_FW_DENY: 4139 m_freem(m); 4140 m = NULL; 4141 error = EACCES; 4142 break; 4143 4144 case IP_FW_DUMMYNET: 4145 /* Send packet to the appropriate pipe */ 4146 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_IN, &args); 4147 break; 4148 4149 case IP_FW_TEE: 4150 tee = 1; 4151 /* FALL THROUGH */ 4152 4153 case IP_FW_DIVERT: 4154 /* 4155 * Must clear bridge tag when changing 4156 */ 4157 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 4158 if (ip_divert_p != NULL) { 4159 m = ip_divert_p(m, tee, 1); 4160 } else { 4161 m_freem(m); 4162 m = NULL; 4163 /* not sure this is the right error msg */ 4164 error = EACCES; 4165 } 4166 break; 4167 4168 default: 4169 panic("unknown ipfw return value: %d", ret); 4170 } 4171 back: 4172 *m0 = m; 4173 return error; 4174 } 4175 4176 static int 4177 ipfw_check_out(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir) 4178 { 4179 struct ip_fw_args args; 4180 struct mbuf *m = *m0; 4181 struct m_tag *mtag; 4182 int tee = 0, error = 0, ret; 4183 4184 if (m->m_pkthdr.fw_flags & DUMMYNET_MBUF_TAGGED) { 4185 /* Extract info from dummynet tag */ 4186 mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 4187 KKASSERT(mtag != NULL); 4188 args.rule = ((struct dn_pkt *)m_tag_data(mtag))->dn_priv; 4189 KKASSERT(args.rule != NULL); 4190 4191 m_tag_delete(m, mtag); 4192 m->m_pkthdr.fw_flags &= ~DUMMYNET_MBUF_TAGGED; 4193 } else { 4194 args.rule = NULL; 4195 } 4196 4197 args.eh = NULL; 4198 args.m = m; 4199 args.oif = ifp; 4200 ret = ipfw_chk(&args); 4201 m = args.m; 4202 4203 if (m == NULL) { 4204 error = EACCES; 4205 goto back; 4206 } 4207 4208 switch (ret) { 4209 case IP_FW_PASS: 4210 break; 4211 4212 case IP_FW_DENY: 4213 m_freem(m); 4214 m = NULL; 4215 error = EACCES; 4216 break; 4217 4218 case IP_FW_DUMMYNET: 4219 ipfw_dummynet_io(m, args.cookie, DN_TO_IP_OUT, &args); 4220 break; 4221 4222 case IP_FW_TEE: 4223 tee = 1; 4224 /* FALL THROUGH */ 4225 4226 case IP_FW_DIVERT: 4227 if (ip_divert_p != NULL) { 4228 m = ip_divert_p(m, tee, 0); 4229 } else { 4230 m_freem(m); 4231 m = NULL; 4232 /* not sure this is the right error msg */ 4233 error = EACCES; 4234 } 4235 break; 4236 4237 default: 4238 panic("unknown ipfw return value: %d", ret); 4239 } 4240 back: 4241 *m0 = m; 4242 return error; 4243 } 4244 4245 static void 4246 ipfw_hook(void) 4247 { 4248 struct pfil_head *pfh; 4249 4250 IPFW_ASSERT_CFGPORT(&curthread->td_msgport); 4251 4252 pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET); 4253 if (pfh == NULL) 4254 return; 4255 4256 pfil_add_hook(ipfw_check_in, NULL, PFIL_IN | PFIL_MPSAFE, pfh); 4257 pfil_add_hook(ipfw_check_out, NULL, PFIL_OUT | PFIL_MPSAFE, pfh); 4258 } 4259 4260 static void 4261 ipfw_dehook(void) 4262 { 4263 struct pfil_head *pfh; 4264 4265 IPFW_ASSERT_CFGPORT(&curthread->td_msgport); 4266 4267 pfh = pfil_head_get(PFIL_TYPE_AF, AF_INET); 4268 if (pfh == NULL) 4269 return; 4270 4271 pfil_remove_hook(ipfw_check_in, NULL, PFIL_IN, pfh); 4272 pfil_remove_hook(ipfw_check_out, NULL, PFIL_OUT, pfh); 4273 } 4274 4275 static void 4276 ipfw_sysctl_enable_dispatch(netmsg_t nmsg) 4277 { 4278 struct lwkt_msg *lmsg = &nmsg->lmsg; 4279 int enable = lmsg->u.ms_result; 4280 4281 if (fw_enable == enable) 4282 goto reply; 4283 4284 fw_enable = enable; 4285 if (fw_enable) 4286 ipfw_hook(); 4287 else 4288 ipfw_dehook(); 4289 reply: 4290 lwkt_replymsg(lmsg, 0); 4291 } 4292 4293 static int 4294 ipfw_sysctl_enable(SYSCTL_HANDLER_ARGS) 4295 { 4296 struct netmsg_base nmsg; 4297 struct lwkt_msg *lmsg; 4298 int enable, error; 4299 4300 enable = fw_enable; 4301 error = sysctl_handle_int(oidp, &enable, 0, req); 4302 if (error || req->newptr == NULL) 4303 return error; 4304 4305 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 4306 0, ipfw_sysctl_enable_dispatch); 4307 lmsg = &nmsg.lmsg; 4308 lmsg->u.ms_result = enable; 4309 4310 return lwkt_domsg(IPFW_CFGPORT, lmsg, 0); 4311 } 4312 4313 static int 4314 ipfw_sysctl_autoinc_step(SYSCTL_HANDLER_ARGS) 4315 { 4316 return sysctl_int_range(oidp, arg1, arg2, req, 4317 IPFW_AUTOINC_STEP_MIN, IPFW_AUTOINC_STEP_MAX); 4318 } 4319 4320 static int 4321 ipfw_sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS) 4322 { 4323 int error, value; 4324 4325 lockmgr(&dyn_lock, LK_EXCLUSIVE); 4326 4327 value = dyn_buckets; 4328 error = sysctl_handle_int(oidp, &value, 0, req); 4329 if (error || !req->newptr) 4330 goto back; 4331 4332 /* 4333 * Make sure we have a power of 2 and 4334 * do not allow more than 64k entries. 4335 */ 4336 error = EINVAL; 4337 if (value <= 1 || value > 65536) 4338 goto back; 4339 if ((value & (value - 1)) != 0) 4340 goto back; 4341 4342 error = 0; 4343 dyn_buckets = value; 4344 back: 4345 lockmgr(&dyn_lock, LK_RELEASE); 4346 return error; 4347 } 4348 4349 static int 4350 ipfw_sysctl_dyn_fin(SYSCTL_HANDLER_ARGS) 4351 { 4352 return sysctl_int_range(oidp, arg1, arg2, req, 4353 1, dyn_keepalive_period - 1); 4354 } 4355 4356 static int 4357 ipfw_sysctl_dyn_rst(SYSCTL_HANDLER_ARGS) 4358 { 4359 return sysctl_int_range(oidp, arg1, arg2, req, 4360 1, dyn_keepalive_period - 1); 4361 } 4362 4363 static void 4364 ipfw_ctx_init_dispatch(netmsg_t nmsg) 4365 { 4366 struct netmsg_ipfw *fwmsg = (struct netmsg_ipfw *)nmsg; 4367 struct ipfw_context *ctx; 4368 struct ip_fw *def_rule; 4369 4370 ctx = kmalloc(sizeof(*ctx), M_IPFW, M_WAITOK | M_ZERO); 4371 ipfw_ctx[mycpuid] = ctx; 4372 4373 def_rule = kmalloc(sizeof(*def_rule), M_IPFW, M_WAITOK | M_ZERO); 4374 4375 def_rule->act_ofs = 0; 4376 def_rule->rulenum = IPFW_DEFAULT_RULE; 4377 def_rule->cmd_len = 1; 4378 def_rule->set = IPFW_DEFAULT_SET; 4379 4380 def_rule->cmd[0].len = 1; 4381 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT 4382 def_rule->cmd[0].opcode = O_ACCEPT; 4383 #else 4384 def_rule->cmd[0].opcode = O_DENY; 4385 #endif 4386 4387 def_rule->refcnt = 1; 4388 def_rule->cpuid = mycpuid; 4389 4390 /* Install the default rule */ 4391 ctx->ipfw_default_rule = def_rule; 4392 ctx->ipfw_layer3_chain = def_rule; 4393 4394 /* Link rule CPU sibling */ 4395 ipfw_link_sibling(fwmsg, def_rule); 4396 4397 /* Statistics only need to be updated once */ 4398 if (mycpuid == 0) 4399 ipfw_inc_static_count(def_rule); 4400 4401 ifnet_forwardmsg(&nmsg->lmsg, mycpuid + 1); 4402 } 4403 4404 static void 4405 ipfw_init_dispatch(netmsg_t nmsg) 4406 { 4407 struct netmsg_ipfw fwmsg; 4408 int error = 0; 4409 4410 if (IPFW_LOADED) { 4411 kprintf("IP firewall already loaded\n"); 4412 error = EEXIST; 4413 goto reply; 4414 } 4415 4416 bzero(&fwmsg, sizeof(fwmsg)); 4417 netmsg_init(&fwmsg.base, NULL, &curthread->td_msgport, 4418 0, ipfw_ctx_init_dispatch); 4419 ifnet_domsg(&fwmsg.base.lmsg, 0); 4420 4421 ip_fw_chk_ptr = ipfw_chk; 4422 ip_fw_ctl_ptr = ipfw_ctl; 4423 ip_fw_dn_io_ptr = ipfw_dummynet_io; 4424 4425 kprintf("ipfw2 initialized, default to %s, logging ", 4426 ipfw_ctx[mycpuid]->ipfw_default_rule->cmd[0].opcode == 4427 O_ACCEPT ? "accept" : "deny"); 4428 4429 #ifdef IPFIREWALL_VERBOSE 4430 fw_verbose = 1; 4431 #endif 4432 #ifdef IPFIREWALL_VERBOSE_LIMIT 4433 verbose_limit = IPFIREWALL_VERBOSE_LIMIT; 4434 #endif 4435 if (fw_verbose == 0) { 4436 kprintf("disabled\n"); 4437 } else if (verbose_limit == 0) { 4438 kprintf("unlimited\n"); 4439 } else { 4440 kprintf("limited to %d packets/entry by default\n", 4441 verbose_limit); 4442 } 4443 4444 callout_init_mp(&ipfw_timeout_h); 4445 netmsg_init(&ipfw_timeout_netmsg, NULL, &netisr_adone_rport, 4446 MSGF_DROPABLE | MSGF_PRIORITY, 4447 ipfw_tick_dispatch); 4448 lockinit(&dyn_lock, "ipfw_dyn", 0, 0); 4449 4450 ip_fw_loaded = 1; 4451 callout_reset(&ipfw_timeout_h, hz, ipfw_tick, NULL); 4452 4453 if (fw_enable) 4454 ipfw_hook(); 4455 reply: 4456 lwkt_replymsg(&nmsg->lmsg, error); 4457 } 4458 4459 static int 4460 ipfw_init(void) 4461 { 4462 struct netmsg_base smsg; 4463 4464 netmsg_init(&smsg, NULL, &curthread->td_msgport, 4465 0, ipfw_init_dispatch); 4466 return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0); 4467 } 4468 4469 #ifdef KLD_MODULE 4470 4471 static void 4472 ipfw_fini_dispatch(netmsg_t nmsg) 4473 { 4474 int error = 0, cpu; 4475 4476 if (ipfw_refcnt != 0) { 4477 error = EBUSY; 4478 goto reply; 4479 } 4480 4481 ip_fw_loaded = 0; 4482 4483 ipfw_dehook(); 4484 callout_stop(&ipfw_timeout_h); 4485 4486 netmsg_service_sync(); 4487 4488 crit_enter(); 4489 if ((ipfw_timeout_netmsg.lmsg.ms_flags & MSGF_DONE) == 0) { 4490 /* 4491 * Callout message is pending; drop it 4492 */ 4493 lwkt_dropmsg(&ipfw_timeout_netmsg.lmsg); 4494 } 4495 crit_exit(); 4496 4497 ip_fw_chk_ptr = NULL; 4498 ip_fw_ctl_ptr = NULL; 4499 ip_fw_dn_io_ptr = NULL; 4500 ipfw_flush(1 /* kill default rule */); 4501 4502 /* Free pre-cpu context */ 4503 for (cpu = 0; cpu < ncpus; ++cpu) 4504 kfree(ipfw_ctx[cpu], M_IPFW); 4505 4506 kprintf("IP firewall unloaded\n"); 4507 reply: 4508 lwkt_replymsg(&nmsg->lmsg, error); 4509 } 4510 4511 static int 4512 ipfw_fini(void) 4513 { 4514 struct netmsg_base smsg; 4515 4516 netmsg_init(&smsg, NULL, &curthread->td_msgport, 4517 0, ipfw_fini_dispatch); 4518 return lwkt_domsg(IPFW_CFGPORT, &smsg.lmsg, 0); 4519 } 4520 4521 #endif /* KLD_MODULE */ 4522 4523 static int 4524 ipfw_modevent(module_t mod, int type, void *unused) 4525 { 4526 int err = 0; 4527 4528 switch (type) { 4529 case MOD_LOAD: 4530 err = ipfw_init(); 4531 break; 4532 4533 case MOD_UNLOAD: 4534 #ifndef KLD_MODULE 4535 kprintf("ipfw statically compiled, cannot unload\n"); 4536 err = EBUSY; 4537 #else 4538 err = ipfw_fini(); 4539 #endif 4540 break; 4541 default: 4542 break; 4543 } 4544 return err; 4545 } 4546 4547 static moduledata_t ipfwmod = { 4548 "ipfw", 4549 ipfw_modevent, 4550 0 4551 }; 4552 DECLARE_MODULE(ipfw, ipfwmod, SI_SUB_PROTO_END, SI_ORDER_ANY); 4553 MODULE_VERSION(ipfw, 1); 4554