1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2017-2018 Yandex LLC 5 * Copyright (c) 2017-2018 Andrey V. Elsukov <ae@FreeBSD.org> 6 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 #include "opt_inet.h" 32 #include "opt_inet6.h" 33 #include "opt_ipfw.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/hash.h> 41 #include <sys/mbuf.h> 42 #include <sys/kernel.h> 43 #include <sys/lock.h> 44 #include <sys/pcpu.h> 45 #include <sys/queue.h> 46 #include <sys/rmlock.h> 47 #include <sys/smp.h> 48 #include <sys/socket.h> 49 #include <sys/sysctl.h> 50 #include <sys/syslog.h> 51 #include <net/ethernet.h> 52 #include <net/if.h> 53 #include <net/if_var.h> 54 #include <net/vnet.h> 55 56 #include <netinet/in.h> 57 #include <netinet/ip.h> 58 #include <netinet/ip_var.h> 59 #include <netinet/ip_fw.h> 60 #include <netinet/udp.h> 61 #include <netinet/tcp.h> 62 63 #include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */ 64 #ifdef INET6 65 #include <netinet6/in6_var.h> 66 #include <netinet6/ip6_var.h> 67 #include <netinet6/scope6_var.h> 68 #endif 69 70 #include <netpfil/ipfw/ip_fw_private.h> 71 72 #include <machine/in_cksum.h> /* XXX for in_cksum */ 73 74 #ifdef MAC 75 #include <security/mac/mac_framework.h> 76 #endif 77 78 /* 79 * Description of dynamic states. 80 * 81 * Dynamic states are stored in lists accessed through a hash tables 82 * whose size is curr_dyn_buckets. This value can be modified through 83 * the sysctl variable dyn_buckets. 84 * 85 * Currently there are four tables: dyn_ipv4, dyn_ipv6, dyn_ipv4_parent, 86 * and dyn_ipv6_parent. 87 * 88 * When a packet is received, its address fields hashed, then matched 89 * against the entries in the corresponding list by addr_type. 90 * Dynamic states can be used for different purposes: 91 * + stateful rules; 92 * + enforcing limits on the number of sessions; 93 * + in-kernel NAT (not implemented yet) 94 * 95 * The lifetime of dynamic states is regulated by dyn_*_lifetime, 96 * measured in seconds and depending on the flags. 97 * 98 * The total number of dynamic states is equal to UMA zone items count. 99 * The max number of dynamic states is dyn_max. When we reach 100 * the maximum number of rules we do not create anymore. This is 101 * done to avoid consuming too much memory, but also too much 102 * time when searching on each packet (ideally, we should try instead 103 * to put a limit on the length of the list on each bucket...). 104 * 105 * Each state holds a pointer to the parent ipfw rule so we know what 106 * action to perform. Dynamic rules are removed when the parent rule is 107 * deleted. 108 * 109 * There are some limitations with dynamic rules -- we do not 110 * obey the 'randomized match', and we do not do multiple 111 * passes through the firewall. XXX check the latter!!! 112 */ 113 114 /* By default use jenkins hash function */ 115 #define IPFIREWALL_JENKINSHASH 116 117 #define DYN_COUNTER_INC(d, dir, pktlen) do { \ 118 (d)->pcnt_ ## dir++; \ 119 (d)->bcnt_ ## dir += pktlen; \ 120 } while (0) 121 122 #define DYN_REFERENCED 0x01 123 /* 124 * DYN_REFERENCED flag is used to show that state keeps reference to named 125 * object, and this reference should be released when state becomes expired. 126 */ 127 128 struct dyn_data { 129 void *parent; /* pointer to parent rule */ 130 uint32_t chain_id; /* cached ruleset id */ 131 uint32_t f_pos; /* cached rule index */ 132 133 uint32_t hashval; /* hash value used for hash resize */ 134 uint16_t fibnum; /* fib used to send keepalives */ 135 uint8_t _pad[3]; 136 uint8_t flags; /* internal flags */ 137 uint16_t rulenum; /* parent rule number */ 138 uint32_t ruleid; /* parent rule id */ 139 140 uint32_t state; /* TCP session state and flags */ 141 uint32_t ack_fwd; /* most recent ACKs in forward */ 142 uint32_t ack_rev; /* and reverse direction (used */ 143 /* to generate keepalives) */ 144 uint32_t sync; /* synchronization time */ 145 uint32_t expire; /* expire time */ 146 147 uint64_t pcnt_fwd; /* bytes counter in forward */ 148 uint64_t bcnt_fwd; /* packets counter in forward */ 149 uint64_t pcnt_rev; /* bytes counter in reverse */ 150 uint64_t bcnt_rev; /* packets counter in reverse */ 151 }; 152 153 #define DPARENT_COUNT_DEC(p) do { \ 154 MPASS(p->count > 0); \ 155 ck_pr_dec_32(&(p)->count); \ 156 } while (0) 157 #define DPARENT_COUNT_INC(p) ck_pr_inc_32(&(p)->count) 158 #define DPARENT_COUNT(p) ck_pr_load_32(&(p)->count) 159 struct dyn_parent { 160 void *parent; /* pointer to parent rule */ 161 uint32_t count; /* number of linked states */ 162 uint8_t _pad[2]; 163 uint16_t rulenum; /* parent rule number */ 164 uint32_t ruleid; /* parent rule id */ 165 uint32_t hashval; /* hash value used for hash resize */ 166 uint32_t expire; /* expire time */ 167 }; 168 169 struct dyn_ipv4_state { 170 uint8_t type; /* State type */ 171 uint8_t proto; /* UL Protocol */ 172 uint16_t kidx; /* named object index */ 173 uint16_t sport, dport; /* ULP source and destination ports */ 174 in_addr_t src, dst; /* IPv4 source and destination */ 175 176 union { 177 struct dyn_data *data; 178 struct dyn_parent *limit; 179 }; 180 CK_SLIST_ENTRY(dyn_ipv4_state) entry; 181 SLIST_ENTRY(dyn_ipv4_state) expired; 182 }; 183 CK_SLIST_HEAD(dyn_ipv4ck_slist, dyn_ipv4_state); 184 VNET_DEFINE_STATIC(struct dyn_ipv4ck_slist *, dyn_ipv4); 185 VNET_DEFINE_STATIC(struct dyn_ipv4ck_slist *, dyn_ipv4_parent); 186 187 SLIST_HEAD(dyn_ipv4_slist, dyn_ipv4_state); 188 VNET_DEFINE_STATIC(struct dyn_ipv4_slist, dyn_expired_ipv4); 189 #define V_dyn_ipv4 VNET(dyn_ipv4) 190 #define V_dyn_ipv4_parent VNET(dyn_ipv4_parent) 191 #define V_dyn_expired_ipv4 VNET(dyn_expired_ipv4) 192 193 #ifdef INET6 194 struct dyn_ipv6_state { 195 uint8_t type; /* State type */ 196 uint8_t proto; /* UL Protocol */ 197 uint16_t kidx; /* named object index */ 198 uint16_t sport, dport; /* ULP source and destination ports */ 199 struct in6_addr src, dst; /* IPv6 source and destination */ 200 uint32_t zoneid; /* IPv6 scope zone id */ 201 union { 202 struct dyn_data *data; 203 struct dyn_parent *limit; 204 }; 205 CK_SLIST_ENTRY(dyn_ipv6_state) entry; 206 SLIST_ENTRY(dyn_ipv6_state) expired; 207 }; 208 CK_SLIST_HEAD(dyn_ipv6ck_slist, dyn_ipv6_state); 209 VNET_DEFINE_STATIC(struct dyn_ipv6ck_slist *, dyn_ipv6); 210 VNET_DEFINE_STATIC(struct dyn_ipv6ck_slist *, dyn_ipv6_parent); 211 212 SLIST_HEAD(dyn_ipv6_slist, dyn_ipv6_state); 213 VNET_DEFINE_STATIC(struct dyn_ipv6_slist, dyn_expired_ipv6); 214 #define V_dyn_ipv6 VNET(dyn_ipv6) 215 #define V_dyn_ipv6_parent VNET(dyn_ipv6_parent) 216 #define V_dyn_expired_ipv6 VNET(dyn_expired_ipv6) 217 #endif /* INET6 */ 218 219 /* 220 * Per-CPU pointer indicates that specified state is currently in use 221 * and must not be reclaimed by expiration callout. 222 */ 223 static void **dyn_hp_cache; 224 DPCPU_DEFINE_STATIC(void *, dyn_hp); 225 #define DYNSTATE_GET(cpu) ck_pr_load_ptr(DPCPU_ID_PTR((cpu), dyn_hp)) 226 #define DYNSTATE_PROTECT(v) ck_pr_store_ptr(DPCPU_PTR(dyn_hp), (v)) 227 #define DYNSTATE_RELEASE() DYNSTATE_PROTECT(NULL) 228 #define DYNSTATE_CRITICAL_ENTER() critical_enter() 229 #define DYNSTATE_CRITICAL_EXIT() do { \ 230 DYNSTATE_RELEASE(); \ 231 critical_exit(); \ 232 } while (0); 233 234 /* 235 * We keep two version numbers, one is updated when new entry added to 236 * the list. Second is updated when an entry deleted from the list. 237 * Versions are updated under bucket lock. 238 * 239 * Bucket "add" version number is used to know, that in the time between 240 * state lookup (i.e. ipfw_dyn_lookup_state()) and the followed state 241 * creation (i.e. ipfw_dyn_install_state()) another concurrent thread did 242 * not install some state in this bucket. Using this info we can avoid 243 * additional state lookup, because we are sure that we will not install 244 * the state twice. 245 * 246 * Also doing the tracking of bucket "del" version during lookup we can 247 * be sure, that state entry was not unlinked and freed in time between 248 * we read the state pointer and protect it with hazard pointer. 249 * 250 * An entry unlinked from CK list keeps unchanged until it is freed. 251 * Unlinked entries are linked into expired lists using "expired" field. 252 */ 253 254 /* 255 * dyn_expire_lock is used to protect access to dyn_expired_xxx lists. 256 * dyn_bucket_lock is used to get write access to lists in specific bucket. 257 * Currently one dyn_bucket_lock is used for all ipv4, ipv4_parent, ipv6, 258 * and ipv6_parent lists. 259 */ 260 VNET_DEFINE_STATIC(struct mtx, dyn_expire_lock); 261 VNET_DEFINE_STATIC(struct mtx *, dyn_bucket_lock); 262 #define V_dyn_expire_lock VNET(dyn_expire_lock) 263 #define V_dyn_bucket_lock VNET(dyn_bucket_lock) 264 265 /* 266 * Bucket's add/delete generation versions. 267 */ 268 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_add); 269 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_del); 270 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_parent_add); 271 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_parent_del); 272 #define V_dyn_ipv4_add VNET(dyn_ipv4_add) 273 #define V_dyn_ipv4_del VNET(dyn_ipv4_del) 274 #define V_dyn_ipv4_parent_add VNET(dyn_ipv4_parent_add) 275 #define V_dyn_ipv4_parent_del VNET(dyn_ipv4_parent_del) 276 277 #ifdef INET6 278 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_add); 279 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_del); 280 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_parent_add); 281 VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_parent_del); 282 #define V_dyn_ipv6_add VNET(dyn_ipv6_add) 283 #define V_dyn_ipv6_del VNET(dyn_ipv6_del) 284 #define V_dyn_ipv6_parent_add VNET(dyn_ipv6_parent_add) 285 #define V_dyn_ipv6_parent_del VNET(dyn_ipv6_parent_del) 286 #endif /* INET6 */ 287 288 #define DYN_BUCKET(h, b) ((h) & (b - 1)) 289 #define DYN_BUCKET_VERSION(b, v) ck_pr_load_32(&V_dyn_ ## v[(b)]) 290 #define DYN_BUCKET_VERSION_BUMP(b, v) ck_pr_inc_32(&V_dyn_ ## v[(b)]) 291 292 #define DYN_BUCKET_LOCK_INIT(lock, b) \ 293 mtx_init(&lock[(b)], "IPFW dynamic bucket", NULL, MTX_DEF) 294 #define DYN_BUCKET_LOCK_DESTROY(lock, b) mtx_destroy(&lock[(b)]) 295 #define DYN_BUCKET_LOCK(b) mtx_lock(&V_dyn_bucket_lock[(b)]) 296 #define DYN_BUCKET_UNLOCK(b) mtx_unlock(&V_dyn_bucket_lock[(b)]) 297 #define DYN_BUCKET_ASSERT(b) mtx_assert(&V_dyn_bucket_lock[(b)], MA_OWNED) 298 299 #define DYN_EXPIRED_LOCK_INIT() \ 300 mtx_init(&V_dyn_expire_lock, "IPFW expired states list", NULL, MTX_DEF) 301 #define DYN_EXPIRED_LOCK_DESTROY() mtx_destroy(&V_dyn_expire_lock) 302 #define DYN_EXPIRED_LOCK() mtx_lock(&V_dyn_expire_lock) 303 #define DYN_EXPIRED_UNLOCK() mtx_unlock(&V_dyn_expire_lock) 304 305 VNET_DEFINE_STATIC(uint32_t, dyn_buckets_max); 306 VNET_DEFINE_STATIC(uint32_t, curr_dyn_buckets); 307 VNET_DEFINE_STATIC(struct callout, dyn_timeout); 308 #define V_dyn_buckets_max VNET(dyn_buckets_max) 309 #define V_curr_dyn_buckets VNET(curr_dyn_buckets) 310 #define V_dyn_timeout VNET(dyn_timeout) 311 312 /* Maximum length of states chain in a bucket */ 313 VNET_DEFINE_STATIC(uint32_t, curr_max_length); 314 #define V_curr_max_length VNET(curr_max_length) 315 316 VNET_DEFINE_STATIC(uint32_t, dyn_keep_states); 317 #define V_dyn_keep_states VNET(dyn_keep_states) 318 319 VNET_DEFINE_STATIC(uma_zone_t, dyn_data_zone); 320 VNET_DEFINE_STATIC(uma_zone_t, dyn_parent_zone); 321 VNET_DEFINE_STATIC(uma_zone_t, dyn_ipv4_zone); 322 #ifdef INET6 323 VNET_DEFINE_STATIC(uma_zone_t, dyn_ipv6_zone); 324 #define V_dyn_ipv6_zone VNET(dyn_ipv6_zone) 325 #endif /* INET6 */ 326 #define V_dyn_data_zone VNET(dyn_data_zone) 327 #define V_dyn_parent_zone VNET(dyn_parent_zone) 328 #define V_dyn_ipv4_zone VNET(dyn_ipv4_zone) 329 330 /* 331 * Timeouts for various events in handing dynamic rules. 332 */ 333 VNET_DEFINE_STATIC(uint32_t, dyn_ack_lifetime); 334 VNET_DEFINE_STATIC(uint32_t, dyn_syn_lifetime); 335 VNET_DEFINE_STATIC(uint32_t, dyn_fin_lifetime); 336 VNET_DEFINE_STATIC(uint32_t, dyn_rst_lifetime); 337 VNET_DEFINE_STATIC(uint32_t, dyn_udp_lifetime); 338 VNET_DEFINE_STATIC(uint32_t, dyn_short_lifetime); 339 340 #define V_dyn_ack_lifetime VNET(dyn_ack_lifetime) 341 #define V_dyn_syn_lifetime VNET(dyn_syn_lifetime) 342 #define V_dyn_fin_lifetime VNET(dyn_fin_lifetime) 343 #define V_dyn_rst_lifetime VNET(dyn_rst_lifetime) 344 #define V_dyn_udp_lifetime VNET(dyn_udp_lifetime) 345 #define V_dyn_short_lifetime VNET(dyn_short_lifetime) 346 347 /* 348 * Keepalives are sent if dyn_keepalive is set. They are sent every 349 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval 350 * seconds of lifetime of a rule. 351 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower 352 * than dyn_keepalive_period. 353 */ 354 VNET_DEFINE_STATIC(uint32_t, dyn_keepalive_interval); 355 VNET_DEFINE_STATIC(uint32_t, dyn_keepalive_period); 356 VNET_DEFINE_STATIC(uint32_t, dyn_keepalive); 357 VNET_DEFINE_STATIC(time_t, dyn_keepalive_last); 358 359 #define V_dyn_keepalive_interval VNET(dyn_keepalive_interval) 360 #define V_dyn_keepalive_period VNET(dyn_keepalive_period) 361 #define V_dyn_keepalive VNET(dyn_keepalive) 362 #define V_dyn_keepalive_last VNET(dyn_keepalive_last) 363 364 VNET_DEFINE_STATIC(uint32_t, dyn_max); /* max # of dynamic states */ 365 VNET_DEFINE_STATIC(uint32_t, dyn_count); /* number of states */ 366 VNET_DEFINE_STATIC(uint32_t, dyn_parent_max); /* max # of parent states */ 367 VNET_DEFINE_STATIC(uint32_t, dyn_parent_count); /* number of parent states */ 368 369 #define V_dyn_max VNET(dyn_max) 370 #define V_dyn_count VNET(dyn_count) 371 #define V_dyn_parent_max VNET(dyn_parent_max) 372 #define V_dyn_parent_count VNET(dyn_parent_count) 373 374 #define DYN_COUNT_DEC(name) do { \ 375 MPASS((V_ ## name) > 0); \ 376 ck_pr_dec_32(&(V_ ## name)); \ 377 } while (0) 378 #define DYN_COUNT_INC(name) ck_pr_inc_32(&(V_ ## name)) 379 #define DYN_COUNT(name) ck_pr_load_32(&(V_ ## name)) 380 381 static time_t last_log; /* Log ratelimiting */ 382 383 /* 384 * Get/set maximum number of dynamic states in given VNET instance. 385 */ 386 static int 387 sysctl_dyn_max(SYSCTL_HANDLER_ARGS) 388 { 389 uint32_t nstates; 390 int error; 391 392 nstates = V_dyn_max; 393 error = sysctl_handle_32(oidp, &nstates, 0, req); 394 /* Read operation or some error */ 395 if ((error != 0) || (req->newptr == NULL)) 396 return (error); 397 398 V_dyn_max = nstates; 399 uma_zone_set_max(V_dyn_data_zone, V_dyn_max); 400 return (0); 401 } 402 403 static int 404 sysctl_dyn_parent_max(SYSCTL_HANDLER_ARGS) 405 { 406 uint32_t nstates; 407 int error; 408 409 nstates = V_dyn_parent_max; 410 error = sysctl_handle_32(oidp, &nstates, 0, req); 411 /* Read operation or some error */ 412 if ((error != 0) || (req->newptr == NULL)) 413 return (error); 414 415 V_dyn_parent_max = nstates; 416 uma_zone_set_max(V_dyn_parent_zone, V_dyn_parent_max); 417 return (0); 418 } 419 420 static int 421 sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS) 422 { 423 uint32_t nbuckets; 424 int error; 425 426 nbuckets = V_dyn_buckets_max; 427 error = sysctl_handle_32(oidp, &nbuckets, 0, req); 428 /* Read operation or some error */ 429 if ((error != 0) || (req->newptr == NULL)) 430 return (error); 431 432 if (nbuckets > 256) 433 V_dyn_buckets_max = 1 << fls(nbuckets - 1); 434 else 435 return (EINVAL); 436 return (0); 437 } 438 439 SYSCTL_DECL(_net_inet_ip_fw); 440 441 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_count, 442 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(dyn_count), 0, 443 "Current number of dynamic states."); 444 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_parent_count, 445 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(dyn_parent_count), 0, 446 "Current number of parent states. "); 447 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, 448 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0, 449 "Current number of buckets for states hash table."); 450 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, curr_max_length, 451 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_max_length), 0, 452 "Current maximum length of states chains in hash buckets."); 453 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, 454 CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 455 0, 0, sysctl_dyn_buckets, "IU", 456 "Max number of buckets for dynamic states hash table."); 457 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max, 458 CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 459 0, 0, sysctl_dyn_max, "IU", 460 "Max number of dynamic states."); 461 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_parent_max, 462 CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 463 0, 0, sysctl_dyn_parent_max, "IU", 464 "Max number of parent dynamic states."); 465 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, 466 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0, 467 "Lifetime of dynamic states for TCP ACK."); 468 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, 469 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0, 470 "Lifetime of dynamic states for TCP SYN."); 471 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, 472 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0, 473 "Lifetime of dynamic states for TCP FIN."); 474 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, 475 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0, 476 "Lifetime of dynamic states for TCP RST."); 477 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, 478 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0, 479 "Lifetime of dynamic states for UDP."); 480 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, 481 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0, 482 "Lifetime of dynamic states for other situations."); 483 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, 484 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0, 485 "Enable keepalives for dynamic states."); 486 SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_keep_states, 487 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keep_states), 0, 488 "Do not flush dynamic states on rule deletion"); 489 490 #ifdef IPFIREWALL_DYNDEBUG 491 #define DYN_DEBUG(fmt, ...) do { \ 492 printf("%s: " fmt "\n", __func__, __VA_ARGS__); \ 493 } while (0) 494 #else 495 #define DYN_DEBUG(fmt, ...) 496 #endif /* !IPFIREWALL_DYNDEBUG */ 497 498 #ifdef INET6 499 /* Functions to work with IPv6 states */ 500 static struct dyn_ipv6_state *dyn_lookup_ipv6_state( 501 const struct ipfw_flow_id *, uint32_t, const void *, 502 struct ipfw_dyn_info *, int); 503 static int dyn_lookup_ipv6_state_locked(const struct ipfw_flow_id *, 504 uint32_t, const void *, int, uint32_t, uint16_t); 505 static struct dyn_ipv6_state *dyn_alloc_ipv6_state( 506 const struct ipfw_flow_id *, uint32_t, uint16_t, uint8_t); 507 static int dyn_add_ipv6_state(void *, uint32_t, uint16_t, 508 const struct ipfw_flow_id *, uint32_t, const void *, int, uint32_t, 509 struct ipfw_dyn_info *, uint16_t, uint16_t, uint8_t); 510 static void dyn_export_ipv6_state(const struct dyn_ipv6_state *, 511 ipfw_dyn_rule *); 512 513 static uint32_t dyn_getscopeid(const struct ip_fw_args *); 514 static void dyn_make_keepalive_ipv6(struct mbuf *, const struct in6_addr *, 515 const struct in6_addr *, uint32_t, uint32_t, uint32_t, uint16_t, 516 uint16_t); 517 static void dyn_enqueue_keepalive_ipv6(struct mbufq *, 518 const struct dyn_ipv6_state *); 519 static void dyn_send_keepalive_ipv6(struct ip_fw_chain *); 520 521 static struct dyn_ipv6_state *dyn_lookup_ipv6_parent( 522 const struct ipfw_flow_id *, uint32_t, const void *, uint32_t, uint16_t, 523 uint32_t); 524 static struct dyn_ipv6_state *dyn_lookup_ipv6_parent_locked( 525 const struct ipfw_flow_id *, uint32_t, const void *, uint32_t, uint16_t, 526 uint32_t); 527 static struct dyn_ipv6_state *dyn_add_ipv6_parent(void *, uint32_t, uint16_t, 528 const struct ipfw_flow_id *, uint32_t, uint32_t, uint32_t, uint16_t); 529 #endif /* INET6 */ 530 531 /* Functions to work with limit states */ 532 static void *dyn_get_parent_state(const struct ipfw_flow_id *, uint32_t, 533 struct ip_fw *, uint32_t, uint32_t, uint16_t); 534 static struct dyn_ipv4_state *dyn_lookup_ipv4_parent( 535 const struct ipfw_flow_id *, const void *, uint32_t, uint16_t, uint32_t); 536 static struct dyn_ipv4_state *dyn_lookup_ipv4_parent_locked( 537 const struct ipfw_flow_id *, const void *, uint32_t, uint16_t, uint32_t); 538 static struct dyn_parent *dyn_alloc_parent(void *, uint32_t, uint16_t, 539 uint32_t); 540 static struct dyn_ipv4_state *dyn_add_ipv4_parent(void *, uint32_t, uint16_t, 541 const struct ipfw_flow_id *, uint32_t, uint32_t, uint16_t); 542 543 static void dyn_tick(void *); 544 static void dyn_expire_states(struct ip_fw_chain *, ipfw_range_tlv *); 545 static void dyn_free_states(struct ip_fw_chain *); 546 static void dyn_export_parent(const struct dyn_parent *, uint16_t, uint8_t, 547 ipfw_dyn_rule *); 548 static void dyn_export_data(const struct dyn_data *, uint16_t, uint8_t, 549 uint8_t, ipfw_dyn_rule *); 550 static uint32_t dyn_update_tcp_state(struct dyn_data *, 551 const struct ipfw_flow_id *, const struct tcphdr *, int); 552 static void dyn_update_proto_state(struct dyn_data *, 553 const struct ipfw_flow_id *, const void *, int, int); 554 555 /* Functions to work with IPv4 states */ 556 struct dyn_ipv4_state *dyn_lookup_ipv4_state(const struct ipfw_flow_id *, 557 const void *, struct ipfw_dyn_info *, int); 558 static int dyn_lookup_ipv4_state_locked(const struct ipfw_flow_id *, 559 const void *, int, uint32_t, uint16_t); 560 static struct dyn_ipv4_state *dyn_alloc_ipv4_state( 561 const struct ipfw_flow_id *, uint16_t, uint8_t); 562 static int dyn_add_ipv4_state(void *, uint32_t, uint16_t, 563 const struct ipfw_flow_id *, const void *, int, uint32_t, 564 struct ipfw_dyn_info *, uint16_t, uint16_t, uint8_t); 565 static void dyn_export_ipv4_state(const struct dyn_ipv4_state *, 566 ipfw_dyn_rule *); 567 568 /* 569 * Named states support. 570 */ 571 static char *default_state_name = "default"; 572 struct dyn_state_obj { 573 struct named_object no; 574 char name[64]; 575 }; 576 577 #define DYN_STATE_OBJ(ch, cmd) \ 578 ((struct dyn_state_obj *)SRV_OBJECT(ch, (cmd)->arg1)) 579 /* 580 * Classifier callback. 581 * Return 0 if opcode contains object that should be referenced 582 * or rewritten. 583 */ 584 static int 585 dyn_classify(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype) 586 { 587 588 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); 589 /* Don't rewrite "check-state any" */ 590 if (cmd->arg1 == 0 && 591 cmd->opcode == O_CHECK_STATE) 592 return (1); 593 594 *puidx = cmd->arg1; 595 *ptype = 0; 596 return (0); 597 } 598 599 static void 600 dyn_update(ipfw_insn *cmd, uint16_t idx) 601 { 602 603 cmd->arg1 = idx; 604 DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); 605 } 606 607 static int 608 dyn_findbyname(struct ip_fw_chain *ch, struct tid_info *ti, 609 struct named_object **pno) 610 { 611 ipfw_obj_ntlv *ntlv; 612 const char *name; 613 614 DYN_DEBUG("uidx %d", ti->uidx); 615 if (ti->uidx != 0) { 616 if (ti->tlvs == NULL) 617 return (EINVAL); 618 /* Search ntlv in the buffer provided by user */ 619 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, 620 IPFW_TLV_STATE_NAME); 621 if (ntlv == NULL) 622 return (EINVAL); 623 name = ntlv->name; 624 } else 625 name = default_state_name; 626 /* 627 * Search named object with corresponding name. 628 * Since states objects are global - ignore the set value 629 * and use zero instead. 630 */ 631 *pno = ipfw_objhash_lookup_name_type(CHAIN_TO_SRV(ch), 0, 632 IPFW_TLV_STATE_NAME, name); 633 /* 634 * We always return success here. 635 * The caller will check *pno and mark object as unresolved, 636 * then it will automatically create "default" object. 637 */ 638 return (0); 639 } 640 641 static struct named_object * 642 dyn_findbykidx(struct ip_fw_chain *ch, uint16_t idx) 643 { 644 645 DYN_DEBUG("kidx %d", idx); 646 return (ipfw_objhash_lookup_kidx(CHAIN_TO_SRV(ch), idx)); 647 } 648 649 static int 650 dyn_create(struct ip_fw_chain *ch, struct tid_info *ti, 651 uint16_t *pkidx) 652 { 653 struct namedobj_instance *ni; 654 struct dyn_state_obj *obj; 655 struct named_object *no; 656 ipfw_obj_ntlv *ntlv; 657 char *name; 658 659 DYN_DEBUG("uidx %d", ti->uidx); 660 if (ti->uidx != 0) { 661 if (ti->tlvs == NULL) 662 return (EINVAL); 663 ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, 664 IPFW_TLV_STATE_NAME); 665 if (ntlv == NULL) 666 return (EINVAL); 667 name = ntlv->name; 668 } else 669 name = default_state_name; 670 671 ni = CHAIN_TO_SRV(ch); 672 obj = malloc(sizeof(*obj), M_IPFW, M_WAITOK | M_ZERO); 673 obj->no.name = obj->name; 674 obj->no.etlv = IPFW_TLV_STATE_NAME; 675 strlcpy(obj->name, name, sizeof(obj->name)); 676 677 IPFW_UH_WLOCK(ch); 678 no = ipfw_objhash_lookup_name_type(ni, 0, 679 IPFW_TLV_STATE_NAME, name); 680 if (no != NULL) { 681 /* 682 * Object is already created. 683 * Just return its kidx and bump refcount. 684 */ 685 *pkidx = no->kidx; 686 no->refcnt++; 687 IPFW_UH_WUNLOCK(ch); 688 free(obj, M_IPFW); 689 DYN_DEBUG("\tfound kidx %d", *pkidx); 690 return (0); 691 } 692 if (ipfw_objhash_alloc_idx(ni, &obj->no.kidx) != 0) { 693 DYN_DEBUG("\talloc_idx failed for %s", name); 694 IPFW_UH_WUNLOCK(ch); 695 free(obj, M_IPFW); 696 return (ENOSPC); 697 } 698 ipfw_objhash_add(ni, &obj->no); 699 SRV_OBJECT(ch, obj->no.kidx) = obj; 700 obj->no.refcnt++; 701 *pkidx = obj->no.kidx; 702 IPFW_UH_WUNLOCK(ch); 703 DYN_DEBUG("\tcreated kidx %d", *pkidx); 704 return (0); 705 } 706 707 static void 708 dyn_destroy(struct ip_fw_chain *ch, struct named_object *no) 709 { 710 struct dyn_state_obj *obj; 711 712 IPFW_UH_WLOCK_ASSERT(ch); 713 714 KASSERT(no->etlv == IPFW_TLV_STATE_NAME, 715 ("%s: wrong object type %u", __func__, no->etlv)); 716 KASSERT(no->refcnt == 1, 717 ("Destroying object '%s' (type %u, idx %u) with refcnt %u", 718 no->name, no->etlv, no->kidx, no->refcnt)); 719 DYN_DEBUG("kidx %d", no->kidx); 720 obj = SRV_OBJECT(ch, no->kidx); 721 SRV_OBJECT(ch, no->kidx) = NULL; 722 ipfw_objhash_del(CHAIN_TO_SRV(ch), no); 723 ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), no->kidx); 724 725 free(obj, M_IPFW); 726 } 727 728 static struct opcode_obj_rewrite dyn_opcodes[] = { 729 { 730 O_KEEP_STATE, IPFW_TLV_STATE_NAME, 731 dyn_classify, dyn_update, 732 dyn_findbyname, dyn_findbykidx, 733 dyn_create, dyn_destroy 734 }, 735 { 736 O_CHECK_STATE, IPFW_TLV_STATE_NAME, 737 dyn_classify, dyn_update, 738 dyn_findbyname, dyn_findbykidx, 739 dyn_create, dyn_destroy 740 }, 741 { 742 O_PROBE_STATE, IPFW_TLV_STATE_NAME, 743 dyn_classify, dyn_update, 744 dyn_findbyname, dyn_findbykidx, 745 dyn_create, dyn_destroy 746 }, 747 { 748 O_LIMIT, IPFW_TLV_STATE_NAME, 749 dyn_classify, dyn_update, 750 dyn_findbyname, dyn_findbykidx, 751 dyn_create, dyn_destroy 752 }, 753 }; 754 755 /* 756 * IMPORTANT: the hash function for dynamic rules must be commutative 757 * in source and destination (ip,port), because rules are bidirectional 758 * and we want to find both in the same bucket. 759 */ 760 #ifndef IPFIREWALL_JENKINSHASH 761 static __inline uint32_t 762 hash_packet(const struct ipfw_flow_id *id) 763 { 764 uint32_t i; 765 766 #ifdef INET6 767 if (IS_IP6_FLOW_ID(id)) 768 i = ntohl((id->dst_ip6.__u6_addr.__u6_addr32[2]) ^ 769 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^ 770 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^ 771 (id->src_ip6.__u6_addr.__u6_addr32[3])); 772 else 773 #endif /* INET6 */ 774 i = (id->dst_ip) ^ (id->src_ip); 775 i ^= (id->dst_port) ^ (id->src_port); 776 return (i); 777 } 778 779 static __inline uint32_t 780 hash_parent(const struct ipfw_flow_id *id, const void *rule) 781 { 782 783 return (hash_packet(id) ^ ((uintptr_t)rule)); 784 } 785 786 #else /* IPFIREWALL_JENKINSHASH */ 787 788 VNET_DEFINE_STATIC(uint32_t, dyn_hashseed); 789 #define V_dyn_hashseed VNET(dyn_hashseed) 790 791 static __inline int 792 addrcmp4(const struct ipfw_flow_id *id) 793 { 794 795 if (id->src_ip < id->dst_ip) 796 return (0); 797 if (id->src_ip > id->dst_ip) 798 return (1); 799 if (id->src_port <= id->dst_port) 800 return (0); 801 return (1); 802 } 803 804 #ifdef INET6 805 static __inline int 806 addrcmp6(const struct ipfw_flow_id *id) 807 { 808 int ret; 809 810 ret = memcmp(&id->src_ip6, &id->dst_ip6, sizeof(struct in6_addr)); 811 if (ret < 0) 812 return (0); 813 if (ret > 0) 814 return (1); 815 if (id->src_port <= id->dst_port) 816 return (0); 817 return (1); 818 } 819 820 static __inline uint32_t 821 hash_packet6(const struct ipfw_flow_id *id) 822 { 823 struct tuple6 { 824 struct in6_addr addr[2]; 825 uint16_t port[2]; 826 } t6; 827 828 if (addrcmp6(id) == 0) { 829 t6.addr[0] = id->src_ip6; 830 t6.addr[1] = id->dst_ip6; 831 t6.port[0] = id->src_port; 832 t6.port[1] = id->dst_port; 833 } else { 834 t6.addr[0] = id->dst_ip6; 835 t6.addr[1] = id->src_ip6; 836 t6.port[0] = id->dst_port; 837 t6.port[1] = id->src_port; 838 } 839 return (jenkins_hash32((const uint32_t *)&t6, 840 sizeof(t6) / sizeof(uint32_t), V_dyn_hashseed)); 841 } 842 #endif 843 844 static __inline uint32_t 845 hash_packet(const struct ipfw_flow_id *id) 846 { 847 struct tuple4 { 848 in_addr_t addr[2]; 849 uint16_t port[2]; 850 } t4; 851 852 if (IS_IP4_FLOW_ID(id)) { 853 /* All fields are in host byte order */ 854 if (addrcmp4(id) == 0) { 855 t4.addr[0] = id->src_ip; 856 t4.addr[1] = id->dst_ip; 857 t4.port[0] = id->src_port; 858 t4.port[1] = id->dst_port; 859 } else { 860 t4.addr[0] = id->dst_ip; 861 t4.addr[1] = id->src_ip; 862 t4.port[0] = id->dst_port; 863 t4.port[1] = id->src_port; 864 } 865 return (jenkins_hash32((const uint32_t *)&t4, 866 sizeof(t4) / sizeof(uint32_t), V_dyn_hashseed)); 867 } else 868 #ifdef INET6 869 if (IS_IP6_FLOW_ID(id)) 870 return (hash_packet6(id)); 871 #endif 872 return (0); 873 } 874 875 static __inline uint32_t 876 hash_parent(const struct ipfw_flow_id *id, const void *rule) 877 { 878 879 return (jenkins_hash32((const uint32_t *)&rule, 880 sizeof(rule) / sizeof(uint32_t), hash_packet(id))); 881 } 882 #endif /* IPFIREWALL_JENKINSHASH */ 883 884 /* 885 * Print customizable flow id description via log(9) facility. 886 */ 887 static void 888 print_dyn_rule_flags(const struct ipfw_flow_id *id, int dyn_type, 889 int log_flags, char *prefix, char *postfix) 890 { 891 struct in_addr da; 892 #ifdef INET6 893 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN]; 894 #else 895 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; 896 #endif 897 898 #ifdef INET6 899 if (IS_IP6_FLOW_ID(id)) { 900 ip6_sprintf(src, &id->src_ip6); 901 ip6_sprintf(dst, &id->dst_ip6); 902 } else 903 #endif 904 { 905 da.s_addr = htonl(id->src_ip); 906 inet_ntop(AF_INET, &da, src, sizeof(src)); 907 da.s_addr = htonl(id->dst_ip); 908 inet_ntop(AF_INET, &da, dst, sizeof(dst)); 909 } 910 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n", 911 prefix, dyn_type, src, id->src_port, dst, 912 id->dst_port, V_dyn_count, postfix); 913 } 914 915 #define print_dyn_rule(id, dtype, prefix, postfix) \ 916 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix) 917 918 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) 919 #define TIME_LE(a,b) ((int)((a)-(b)) < 0) 920 #define _SEQ_GE(a,b) ((int)((a)-(b)) >= 0) 921 #define BOTH_SYN (TH_SYN | (TH_SYN << 8)) 922 #define BOTH_FIN (TH_FIN | (TH_FIN << 8)) 923 #define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8)) 924 #define ACK_FWD 0x00010000 /* fwd ack seen */ 925 #define ACK_REV 0x00020000 /* rev ack seen */ 926 #define ACK_BOTH (ACK_FWD | ACK_REV) 927 928 static uint32_t 929 dyn_update_tcp_state(struct dyn_data *data, const struct ipfw_flow_id *pkt, 930 const struct tcphdr *tcp, int dir) 931 { 932 uint32_t ack, expire; 933 uint32_t state, old; 934 uint8_t th_flags; 935 936 expire = data->expire; 937 old = state = data->state; 938 th_flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST); 939 state |= (dir == MATCH_FORWARD) ? th_flags: (th_flags << 8); 940 switch (state & TCP_FLAGS) { 941 case TH_SYN: /* opening */ 942 expire = time_uptime + V_dyn_syn_lifetime; 943 break; 944 945 case BOTH_SYN: /* move to established */ 946 case BOTH_SYN | TH_FIN: /* one side tries to close */ 947 case BOTH_SYN | (TH_FIN << 8): 948 if (tcp == NULL) 949 break; 950 ack = ntohl(tcp->th_ack); 951 if (dir == MATCH_FORWARD) { 952 if (data->ack_fwd == 0 || 953 _SEQ_GE(ack, data->ack_fwd)) { 954 state |= ACK_FWD; 955 if (data->ack_fwd != ack) 956 ck_pr_store_32(&data->ack_fwd, ack); 957 } 958 } else { 959 if (data->ack_rev == 0 || 960 _SEQ_GE(ack, data->ack_rev)) { 961 state |= ACK_REV; 962 if (data->ack_rev != ack) 963 ck_pr_store_32(&data->ack_rev, ack); 964 } 965 } 966 if ((state & ACK_BOTH) == ACK_BOTH) { 967 /* 968 * Set expire time to V_dyn_ack_lifetime only if 969 * we got ACKs for both directions. 970 * We use XOR here to avoid possible state 971 * overwriting in concurrent thread. 972 */ 973 expire = time_uptime + V_dyn_ack_lifetime; 974 ck_pr_xor_32(&data->state, ACK_BOTH); 975 } else if ((data->state & ACK_BOTH) != (state & ACK_BOTH)) 976 ck_pr_or_32(&data->state, state & ACK_BOTH); 977 break; 978 979 case BOTH_SYN | BOTH_FIN: /* both sides closed */ 980 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period) 981 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1; 982 expire = time_uptime + V_dyn_fin_lifetime; 983 break; 984 985 default: 986 if (V_dyn_keepalive != 0 && 987 V_dyn_rst_lifetime >= V_dyn_keepalive_period) 988 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1; 989 expire = time_uptime + V_dyn_rst_lifetime; 990 } 991 /* Save TCP state if it was changed */ 992 if ((state & TCP_FLAGS) != (old & TCP_FLAGS)) 993 ck_pr_or_32(&data->state, state & TCP_FLAGS); 994 return (expire); 995 } 996 997 /* 998 * Update ULP specific state. 999 * For TCP we keep sequence numbers and flags. For other protocols 1000 * currently we update only expire time. Packets and bytes counters 1001 * are also updated here. 1002 */ 1003 static void 1004 dyn_update_proto_state(struct dyn_data *data, const struct ipfw_flow_id *pkt, 1005 const void *ulp, int pktlen, int dir) 1006 { 1007 uint32_t expire; 1008 1009 /* NOTE: we are in critical section here. */ 1010 switch (pkt->proto) { 1011 case IPPROTO_UDP: 1012 case IPPROTO_UDPLITE: 1013 expire = time_uptime + V_dyn_udp_lifetime; 1014 break; 1015 case IPPROTO_TCP: 1016 expire = dyn_update_tcp_state(data, pkt, ulp, dir); 1017 break; 1018 default: 1019 expire = time_uptime + V_dyn_short_lifetime; 1020 } 1021 /* 1022 * Expiration timer has the per-second granularity, no need to update 1023 * it every time when state is matched. 1024 */ 1025 if (data->expire != expire) 1026 ck_pr_store_32(&data->expire, expire); 1027 1028 if (dir == MATCH_FORWARD) 1029 DYN_COUNTER_INC(data, fwd, pktlen); 1030 else 1031 DYN_COUNTER_INC(data, rev, pktlen); 1032 } 1033 1034 /* 1035 * Lookup IPv4 state. 1036 * Must be called in critical section. 1037 */ 1038 struct dyn_ipv4_state * 1039 dyn_lookup_ipv4_state(const struct ipfw_flow_id *pkt, const void *ulp, 1040 struct ipfw_dyn_info *info, int pktlen) 1041 { 1042 struct dyn_ipv4_state *s; 1043 uint32_t version, bucket; 1044 1045 bucket = DYN_BUCKET(info->hashval, V_curr_dyn_buckets); 1046 info->version = DYN_BUCKET_VERSION(bucket, ipv4_add); 1047 restart: 1048 version = DYN_BUCKET_VERSION(bucket, ipv4_del); 1049 CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) { 1050 DYNSTATE_PROTECT(s); 1051 if (version != DYN_BUCKET_VERSION(bucket, ipv4_del)) 1052 goto restart; 1053 if (s->proto != pkt->proto) 1054 continue; 1055 if (info->kidx != 0 && s->kidx != info->kidx) 1056 continue; 1057 if (s->sport == pkt->src_port && s->dport == pkt->dst_port && 1058 s->src == pkt->src_ip && s->dst == pkt->dst_ip) { 1059 info->direction = MATCH_FORWARD; 1060 break; 1061 } 1062 if (s->sport == pkt->dst_port && s->dport == pkt->src_port && 1063 s->src == pkt->dst_ip && s->dst == pkt->src_ip) { 1064 info->direction = MATCH_REVERSE; 1065 break; 1066 } 1067 } 1068 1069 if (s != NULL) 1070 dyn_update_proto_state(s->data, pkt, ulp, pktlen, 1071 info->direction); 1072 return (s); 1073 } 1074 1075 /* 1076 * Lookup IPv4 state. 1077 * Simplifed version is used to check that matching state doesn't exist. 1078 */ 1079 static int 1080 dyn_lookup_ipv4_state_locked(const struct ipfw_flow_id *pkt, 1081 const void *ulp, int pktlen, uint32_t bucket, uint16_t kidx) 1082 { 1083 struct dyn_ipv4_state *s; 1084 int dir; 1085 1086 dir = MATCH_NONE; 1087 DYN_BUCKET_ASSERT(bucket); 1088 CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) { 1089 if (s->proto != pkt->proto || 1090 s->kidx != kidx) 1091 continue; 1092 if (s->sport == pkt->src_port && 1093 s->dport == pkt->dst_port && 1094 s->src == pkt->src_ip && s->dst == pkt->dst_ip) { 1095 dir = MATCH_FORWARD; 1096 break; 1097 } 1098 if (s->sport == pkt->dst_port && s->dport == pkt->src_port && 1099 s->src == pkt->dst_ip && s->dst == pkt->src_ip) { 1100 dir = MATCH_REVERSE; 1101 break; 1102 } 1103 } 1104 if (s != NULL) 1105 dyn_update_proto_state(s->data, pkt, ulp, pktlen, dir); 1106 return (s != NULL); 1107 } 1108 1109 struct dyn_ipv4_state * 1110 dyn_lookup_ipv4_parent(const struct ipfw_flow_id *pkt, const void *rule, 1111 uint32_t ruleid, uint16_t rulenum, uint32_t hashval) 1112 { 1113 struct dyn_ipv4_state *s; 1114 uint32_t version, bucket; 1115 1116 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); 1117 restart: 1118 version = DYN_BUCKET_VERSION(bucket, ipv4_parent_del); 1119 CK_SLIST_FOREACH(s, &V_dyn_ipv4_parent[bucket], entry) { 1120 DYNSTATE_PROTECT(s); 1121 if (version != DYN_BUCKET_VERSION(bucket, ipv4_parent_del)) 1122 goto restart; 1123 /* 1124 * NOTE: we do not need to check kidx, because parent rule 1125 * can not create states with different kidx. 1126 * And parent rule always created for forward direction. 1127 */ 1128 if (s->limit->parent == rule && 1129 s->limit->ruleid == ruleid && 1130 s->limit->rulenum == rulenum && 1131 s->proto == pkt->proto && 1132 s->sport == pkt->src_port && 1133 s->dport == pkt->dst_port && 1134 s->src == pkt->src_ip && s->dst == pkt->dst_ip) { 1135 if (s->limit->expire != time_uptime + 1136 V_dyn_short_lifetime) 1137 ck_pr_store_32(&s->limit->expire, 1138 time_uptime + V_dyn_short_lifetime); 1139 break; 1140 } 1141 } 1142 return (s); 1143 } 1144 1145 static struct dyn_ipv4_state * 1146 dyn_lookup_ipv4_parent_locked(const struct ipfw_flow_id *pkt, 1147 const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t bucket) 1148 { 1149 struct dyn_ipv4_state *s; 1150 1151 DYN_BUCKET_ASSERT(bucket); 1152 CK_SLIST_FOREACH(s, &V_dyn_ipv4_parent[bucket], entry) { 1153 if (s->limit->parent == rule && 1154 s->limit->ruleid == ruleid && 1155 s->limit->rulenum == rulenum && 1156 s->proto == pkt->proto && 1157 s->sport == pkt->src_port && 1158 s->dport == pkt->dst_port && 1159 s->src == pkt->src_ip && s->dst == pkt->dst_ip) 1160 break; 1161 } 1162 return (s); 1163 } 1164 1165 #ifdef INET6 1166 static uint32_t 1167 dyn_getscopeid(const struct ip_fw_args *args) 1168 { 1169 1170 /* 1171 * If source or destination address is an scopeid address, we need 1172 * determine the scope zone id to resolve address scope ambiguity. 1173 */ 1174 if (IN6_IS_ADDR_LINKLOCAL(&args->f_id.src_ip6) || 1175 IN6_IS_ADDR_LINKLOCAL(&args->f_id.dst_ip6)) 1176 return (in6_getscopezone(args->ifp, IPV6_ADDR_SCOPE_LINKLOCAL)); 1177 1178 return (0); 1179 } 1180 1181 /* 1182 * Lookup IPv6 state. 1183 * Must be called in critical section. 1184 */ 1185 static struct dyn_ipv6_state * 1186 dyn_lookup_ipv6_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, 1187 const void *ulp, struct ipfw_dyn_info *info, int pktlen) 1188 { 1189 struct dyn_ipv6_state *s; 1190 uint32_t version, bucket; 1191 1192 bucket = DYN_BUCKET(info->hashval, V_curr_dyn_buckets); 1193 info->version = DYN_BUCKET_VERSION(bucket, ipv6_add); 1194 restart: 1195 version = DYN_BUCKET_VERSION(bucket, ipv6_del); 1196 CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) { 1197 DYNSTATE_PROTECT(s); 1198 if (version != DYN_BUCKET_VERSION(bucket, ipv6_del)) 1199 goto restart; 1200 if (s->proto != pkt->proto || s->zoneid != zoneid) 1201 continue; 1202 if (info->kidx != 0 && s->kidx != info->kidx) 1203 continue; 1204 if (s->sport == pkt->src_port && s->dport == pkt->dst_port && 1205 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && 1206 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) { 1207 info->direction = MATCH_FORWARD; 1208 break; 1209 } 1210 if (s->sport == pkt->dst_port && s->dport == pkt->src_port && 1211 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->dst_ip6) && 1212 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->src_ip6)) { 1213 info->direction = MATCH_REVERSE; 1214 break; 1215 } 1216 } 1217 if (s != NULL) 1218 dyn_update_proto_state(s->data, pkt, ulp, pktlen, 1219 info->direction); 1220 return (s); 1221 } 1222 1223 /* 1224 * Lookup IPv6 state. 1225 * Simplifed version is used to check that matching state doesn't exist. 1226 */ 1227 static int 1228 dyn_lookup_ipv6_state_locked(const struct ipfw_flow_id *pkt, uint32_t zoneid, 1229 const void *ulp, int pktlen, uint32_t bucket, uint16_t kidx) 1230 { 1231 struct dyn_ipv6_state *s; 1232 int dir; 1233 1234 dir = MATCH_NONE; 1235 DYN_BUCKET_ASSERT(bucket); 1236 CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) { 1237 if (s->proto != pkt->proto || s->kidx != kidx || 1238 s->zoneid != zoneid) 1239 continue; 1240 if (s->sport == pkt->src_port && s->dport == pkt->dst_port && 1241 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && 1242 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) { 1243 dir = MATCH_FORWARD; 1244 break; 1245 } 1246 if (s->sport == pkt->dst_port && s->dport == pkt->src_port && 1247 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->dst_ip6) && 1248 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->src_ip6)) { 1249 dir = MATCH_REVERSE; 1250 break; 1251 } 1252 } 1253 if (s != NULL) 1254 dyn_update_proto_state(s->data, pkt, ulp, pktlen, dir); 1255 return (s != NULL); 1256 } 1257 1258 static struct dyn_ipv6_state * 1259 dyn_lookup_ipv6_parent(const struct ipfw_flow_id *pkt, uint32_t zoneid, 1260 const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t hashval) 1261 { 1262 struct dyn_ipv6_state *s; 1263 uint32_t version, bucket; 1264 1265 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); 1266 restart: 1267 version = DYN_BUCKET_VERSION(bucket, ipv6_parent_del); 1268 CK_SLIST_FOREACH(s, &V_dyn_ipv6_parent[bucket], entry) { 1269 DYNSTATE_PROTECT(s); 1270 if (version != DYN_BUCKET_VERSION(bucket, ipv6_parent_del)) 1271 goto restart; 1272 /* 1273 * NOTE: we do not need to check kidx, because parent rule 1274 * can not create states with different kidx. 1275 * Also parent rule always created for forward direction. 1276 */ 1277 if (s->limit->parent == rule && 1278 s->limit->ruleid == ruleid && 1279 s->limit->rulenum == rulenum && 1280 s->proto == pkt->proto && 1281 s->sport == pkt->src_port && 1282 s->dport == pkt->dst_port && s->zoneid == zoneid && 1283 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && 1284 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) { 1285 if (s->limit->expire != time_uptime + 1286 V_dyn_short_lifetime) 1287 ck_pr_store_32(&s->limit->expire, 1288 time_uptime + V_dyn_short_lifetime); 1289 break; 1290 } 1291 } 1292 return (s); 1293 } 1294 1295 static struct dyn_ipv6_state * 1296 dyn_lookup_ipv6_parent_locked(const struct ipfw_flow_id *pkt, uint32_t zoneid, 1297 const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t bucket) 1298 { 1299 struct dyn_ipv6_state *s; 1300 1301 DYN_BUCKET_ASSERT(bucket); 1302 CK_SLIST_FOREACH(s, &V_dyn_ipv6_parent[bucket], entry) { 1303 if (s->limit->parent == rule && 1304 s->limit->ruleid == ruleid && 1305 s->limit->rulenum == rulenum && 1306 s->proto == pkt->proto && 1307 s->sport == pkt->src_port && 1308 s->dport == pkt->dst_port && s->zoneid == zoneid && 1309 IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && 1310 IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) 1311 break; 1312 } 1313 return (s); 1314 } 1315 1316 #endif /* INET6 */ 1317 1318 /* 1319 * Lookup dynamic state. 1320 * pkt - filled by ipfw_chk() ipfw_flow_id; 1321 * ulp - determined by ipfw_chk() upper level protocol header; 1322 * dyn_info - info about matched state to return back; 1323 * Returns pointer to state's parent rule and dyn_info. If there is 1324 * no state, NULL is returned. 1325 * On match ipfw_dyn_lookup() updates state's counters. 1326 */ 1327 struct ip_fw * 1328 ipfw_dyn_lookup_state(const struct ip_fw_args *args, const void *ulp, 1329 int pktlen, const ipfw_insn *cmd, struct ipfw_dyn_info *info) 1330 { 1331 struct dyn_data *data; 1332 struct ip_fw *rule; 1333 1334 IPFW_RLOCK_ASSERT(&V_layer3_chain); 1335 1336 data = NULL; 1337 rule = NULL; 1338 info->kidx = cmd->arg1; 1339 info->direction = MATCH_NONE; 1340 info->hashval = hash_packet(&args->f_id); 1341 1342 DYNSTATE_CRITICAL_ENTER(); 1343 if (IS_IP4_FLOW_ID(&args->f_id)) { 1344 struct dyn_ipv4_state *s; 1345 1346 s = dyn_lookup_ipv4_state(&args->f_id, ulp, info, pktlen); 1347 if (s != NULL) { 1348 /* 1349 * Dynamic states are created using the same 5-tuple, 1350 * so it is assumed, that parent rule for O_LIMIT 1351 * state has the same address family. 1352 */ 1353 data = s->data; 1354 if (s->type == O_LIMIT) { 1355 s = data->parent; 1356 rule = s->limit->parent; 1357 } else 1358 rule = data->parent; 1359 } 1360 } 1361 #ifdef INET6 1362 else if (IS_IP6_FLOW_ID(&args->f_id)) { 1363 struct dyn_ipv6_state *s; 1364 1365 s = dyn_lookup_ipv6_state(&args->f_id, dyn_getscopeid(args), 1366 ulp, info, pktlen); 1367 if (s != NULL) { 1368 data = s->data; 1369 if (s->type == O_LIMIT) { 1370 s = data->parent; 1371 rule = s->limit->parent; 1372 } else 1373 rule = data->parent; 1374 } 1375 } 1376 #endif 1377 if (data != NULL) { 1378 /* 1379 * If cached chain id is the same, we can avoid rule index 1380 * lookup. Otherwise do lookup and update chain_id and f_pos. 1381 * It is safe even if there is concurrent thread that want 1382 * update the same state, because chain->id can be changed 1383 * only under IPFW_WLOCK(). 1384 */ 1385 if (data->chain_id != V_layer3_chain.id) { 1386 data->f_pos = ipfw_find_rule(&V_layer3_chain, 1387 data->rulenum, data->ruleid); 1388 /* 1389 * Check that found state has not orphaned. 1390 * When chain->id being changed the parent 1391 * rule can be deleted. If found rule doesn't 1392 * match the parent pointer, consider this 1393 * result as MATCH_NONE and return NULL. 1394 * 1395 * This will lead to creation of new similar state 1396 * that will be added into head of this bucket. 1397 * And the state that we currently have matched 1398 * should be deleted by dyn_expire_states(). 1399 * 1400 * In case when dyn_keep_states is enabled, return 1401 * pointer to deleted rule and f_pos value 1402 * corresponding to penultimate rule. 1403 * When we have enabled V_dyn_keep_states, states 1404 * that become orphaned will get the DYN_REFERENCED 1405 * flag and rule will keep around. So we can return 1406 * it. But since it is not in the rules map, we need 1407 * return such f_pos value, so after the state 1408 * handling if the search will continue, the next rule 1409 * will be the last one - the default rule. 1410 */ 1411 if (V_layer3_chain.map[data->f_pos] == rule) { 1412 data->chain_id = V_layer3_chain.id; 1413 info->f_pos = data->f_pos; 1414 } else if (V_dyn_keep_states != 0) { 1415 /* 1416 * The original rule pointer is still usable. 1417 * So, we return it, but f_pos need to be 1418 * changed to point to the penultimate rule. 1419 */ 1420 MPASS(V_layer3_chain.n_rules > 1); 1421 data->chain_id = V_layer3_chain.id; 1422 data->f_pos = V_layer3_chain.n_rules - 2; 1423 info->f_pos = data->f_pos; 1424 } else { 1425 rule = NULL; 1426 info->direction = MATCH_NONE; 1427 DYN_DEBUG("rule %p [%u, %u] is considered " 1428 "invalid in data %p", rule, data->ruleid, 1429 data->rulenum, data); 1430 /* info->f_pos doesn't matter here. */ 1431 } 1432 } else 1433 info->f_pos = data->f_pos; 1434 } 1435 DYNSTATE_CRITICAL_EXIT(); 1436 #if 0 1437 /* 1438 * Return MATCH_NONE if parent rule is in disabled set. 1439 * This will lead to creation of new similar state that 1440 * will be added into head of this bucket. 1441 * 1442 * XXXAE: we need to be able update state's set when parent 1443 * rule set is changed. 1444 */ 1445 if (rule != NULL && (V_set_disable & (1 << rule->set))) { 1446 rule = NULL; 1447 info->direction = MATCH_NONE; 1448 } 1449 #endif 1450 return (rule); 1451 } 1452 1453 static struct dyn_parent * 1454 dyn_alloc_parent(void *parent, uint32_t ruleid, uint16_t rulenum, 1455 uint32_t hashval) 1456 { 1457 struct dyn_parent *limit; 1458 1459 limit = uma_zalloc(V_dyn_parent_zone, M_NOWAIT | M_ZERO); 1460 if (limit == NULL) { 1461 if (last_log != time_uptime) { 1462 last_log = time_uptime; 1463 log(LOG_DEBUG, 1464 "ipfw: Cannot allocate parent dynamic state, " 1465 "consider increasing " 1466 "net.inet.ip.fw.dyn_parent_max\n"); 1467 } 1468 return (NULL); 1469 } 1470 1471 limit->parent = parent; 1472 limit->ruleid = ruleid; 1473 limit->rulenum = rulenum; 1474 limit->hashval = hashval; 1475 limit->expire = time_uptime + V_dyn_short_lifetime; 1476 return (limit); 1477 } 1478 1479 static struct dyn_data * 1480 dyn_alloc_dyndata(void *parent, uint32_t ruleid, uint16_t rulenum, 1481 const struct ipfw_flow_id *pkt, const void *ulp, int pktlen, 1482 uint32_t hashval, uint16_t fibnum) 1483 { 1484 struct dyn_data *data; 1485 1486 data = uma_zalloc(V_dyn_data_zone, M_NOWAIT | M_ZERO); 1487 if (data == NULL) { 1488 if (last_log != time_uptime) { 1489 last_log = time_uptime; 1490 log(LOG_DEBUG, 1491 "ipfw: Cannot allocate dynamic state, " 1492 "consider increasing net.inet.ip.fw.dyn_max\n"); 1493 } 1494 return (NULL); 1495 } 1496 1497 data->parent = parent; 1498 data->ruleid = ruleid; 1499 data->rulenum = rulenum; 1500 data->fibnum = fibnum; 1501 data->hashval = hashval; 1502 data->expire = time_uptime + V_dyn_syn_lifetime; 1503 dyn_update_proto_state(data, pkt, ulp, pktlen, MATCH_FORWARD); 1504 return (data); 1505 } 1506 1507 static struct dyn_ipv4_state * 1508 dyn_alloc_ipv4_state(const struct ipfw_flow_id *pkt, uint16_t kidx, 1509 uint8_t type) 1510 { 1511 struct dyn_ipv4_state *s; 1512 1513 s = uma_zalloc(V_dyn_ipv4_zone, M_NOWAIT | M_ZERO); 1514 if (s == NULL) 1515 return (NULL); 1516 1517 s->type = type; 1518 s->kidx = kidx; 1519 s->proto = pkt->proto; 1520 s->sport = pkt->src_port; 1521 s->dport = pkt->dst_port; 1522 s->src = pkt->src_ip; 1523 s->dst = pkt->dst_ip; 1524 return (s); 1525 } 1526 1527 /* 1528 * Add IPv4 parent state. 1529 * Returns pointer to parent state. When it is not NULL we are in 1530 * critical section and pointer protected by hazard pointer. 1531 * When some error occurs, it returns NULL and exit from critical section 1532 * is not needed. 1533 */ 1534 static struct dyn_ipv4_state * 1535 dyn_add_ipv4_parent(void *rule, uint32_t ruleid, uint16_t rulenum, 1536 const struct ipfw_flow_id *pkt, uint32_t hashval, uint32_t version, 1537 uint16_t kidx) 1538 { 1539 struct dyn_ipv4_state *s; 1540 struct dyn_parent *limit; 1541 uint32_t bucket; 1542 1543 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); 1544 DYN_BUCKET_LOCK(bucket); 1545 if (version != DYN_BUCKET_VERSION(bucket, ipv4_parent_add)) { 1546 /* 1547 * Bucket version has been changed since last lookup, 1548 * do lookup again to be sure that state does not exist. 1549 */ 1550 s = dyn_lookup_ipv4_parent_locked(pkt, rule, ruleid, 1551 rulenum, bucket); 1552 if (s != NULL) { 1553 /* 1554 * Simultaneous thread has already created this 1555 * state. Just return it. 1556 */ 1557 DYNSTATE_CRITICAL_ENTER(); 1558 DYNSTATE_PROTECT(s); 1559 DYN_BUCKET_UNLOCK(bucket); 1560 return (s); 1561 } 1562 } 1563 1564 limit = dyn_alloc_parent(rule, ruleid, rulenum, hashval); 1565 if (limit == NULL) { 1566 DYN_BUCKET_UNLOCK(bucket); 1567 return (NULL); 1568 } 1569 1570 s = dyn_alloc_ipv4_state(pkt, kidx, O_LIMIT_PARENT); 1571 if (s == NULL) { 1572 DYN_BUCKET_UNLOCK(bucket); 1573 uma_zfree(V_dyn_parent_zone, limit); 1574 return (NULL); 1575 } 1576 1577 s->limit = limit; 1578 CK_SLIST_INSERT_HEAD(&V_dyn_ipv4_parent[bucket], s, entry); 1579 DYN_COUNT_INC(dyn_parent_count); 1580 DYN_BUCKET_VERSION_BUMP(bucket, ipv4_parent_add); 1581 DYNSTATE_CRITICAL_ENTER(); 1582 DYNSTATE_PROTECT(s); 1583 DYN_BUCKET_UNLOCK(bucket); 1584 return (s); 1585 } 1586 1587 static int 1588 dyn_add_ipv4_state(void *parent, uint32_t ruleid, uint16_t rulenum, 1589 const struct ipfw_flow_id *pkt, const void *ulp, int pktlen, 1590 uint32_t hashval, struct ipfw_dyn_info *info, uint16_t fibnum, 1591 uint16_t kidx, uint8_t type) 1592 { 1593 struct dyn_ipv4_state *s; 1594 void *data; 1595 uint32_t bucket; 1596 1597 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); 1598 DYN_BUCKET_LOCK(bucket); 1599 if (info->direction == MATCH_UNKNOWN || 1600 info->kidx != kidx || 1601 info->hashval != hashval || 1602 info->version != DYN_BUCKET_VERSION(bucket, ipv4_add)) { 1603 /* 1604 * Bucket version has been changed since last lookup, 1605 * do lookup again to be sure that state does not exist. 1606 */ 1607 if (dyn_lookup_ipv4_state_locked(pkt, ulp, pktlen, 1608 bucket, kidx) != 0) { 1609 DYN_BUCKET_UNLOCK(bucket); 1610 return (EEXIST); 1611 } 1612 } 1613 1614 data = dyn_alloc_dyndata(parent, ruleid, rulenum, pkt, ulp, 1615 pktlen, hashval, fibnum); 1616 if (data == NULL) { 1617 DYN_BUCKET_UNLOCK(bucket); 1618 return (ENOMEM); 1619 } 1620 1621 s = dyn_alloc_ipv4_state(pkt, kidx, type); 1622 if (s == NULL) { 1623 DYN_BUCKET_UNLOCK(bucket); 1624 uma_zfree(V_dyn_data_zone, data); 1625 return (ENOMEM); 1626 } 1627 1628 s->data = data; 1629 CK_SLIST_INSERT_HEAD(&V_dyn_ipv4[bucket], s, entry); 1630 DYN_COUNT_INC(dyn_count); 1631 DYN_BUCKET_VERSION_BUMP(bucket, ipv4_add); 1632 DYN_BUCKET_UNLOCK(bucket); 1633 return (0); 1634 } 1635 1636 #ifdef INET6 1637 static struct dyn_ipv6_state * 1638 dyn_alloc_ipv6_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, 1639 uint16_t kidx, uint8_t type) 1640 { 1641 struct dyn_ipv6_state *s; 1642 1643 s = uma_zalloc(V_dyn_ipv6_zone, M_NOWAIT | M_ZERO); 1644 if (s == NULL) 1645 return (NULL); 1646 1647 s->type = type; 1648 s->kidx = kidx; 1649 s->zoneid = zoneid; 1650 s->proto = pkt->proto; 1651 s->sport = pkt->src_port; 1652 s->dport = pkt->dst_port; 1653 s->src = pkt->src_ip6; 1654 s->dst = pkt->dst_ip6; 1655 return (s); 1656 } 1657 1658 /* 1659 * Add IPv6 parent state. 1660 * Returns pointer to parent state. When it is not NULL we are in 1661 * critical section and pointer protected by hazard pointer. 1662 * When some error occurs, it return NULL and exit from critical section 1663 * is not needed. 1664 */ 1665 static struct dyn_ipv6_state * 1666 dyn_add_ipv6_parent(void *rule, uint32_t ruleid, uint16_t rulenum, 1667 const struct ipfw_flow_id *pkt, uint32_t zoneid, uint32_t hashval, 1668 uint32_t version, uint16_t kidx) 1669 { 1670 struct dyn_ipv6_state *s; 1671 struct dyn_parent *limit; 1672 uint32_t bucket; 1673 1674 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); 1675 DYN_BUCKET_LOCK(bucket); 1676 if (version != DYN_BUCKET_VERSION(bucket, ipv6_parent_add)) { 1677 /* 1678 * Bucket version has been changed since last lookup, 1679 * do lookup again to be sure that state does not exist. 1680 */ 1681 s = dyn_lookup_ipv6_parent_locked(pkt, zoneid, rule, ruleid, 1682 rulenum, bucket); 1683 if (s != NULL) { 1684 /* 1685 * Simultaneous thread has already created this 1686 * state. Just return it. 1687 */ 1688 DYNSTATE_CRITICAL_ENTER(); 1689 DYNSTATE_PROTECT(s); 1690 DYN_BUCKET_UNLOCK(bucket); 1691 return (s); 1692 } 1693 } 1694 1695 limit = dyn_alloc_parent(rule, ruleid, rulenum, hashval); 1696 if (limit == NULL) { 1697 DYN_BUCKET_UNLOCK(bucket); 1698 return (NULL); 1699 } 1700 1701 s = dyn_alloc_ipv6_state(pkt, zoneid, kidx, O_LIMIT_PARENT); 1702 if (s == NULL) { 1703 DYN_BUCKET_UNLOCK(bucket); 1704 uma_zfree(V_dyn_parent_zone, limit); 1705 return (NULL); 1706 } 1707 1708 s->limit = limit; 1709 CK_SLIST_INSERT_HEAD(&V_dyn_ipv6_parent[bucket], s, entry); 1710 DYN_COUNT_INC(dyn_parent_count); 1711 DYN_BUCKET_VERSION_BUMP(bucket, ipv6_parent_add); 1712 DYNSTATE_CRITICAL_ENTER(); 1713 DYNSTATE_PROTECT(s); 1714 DYN_BUCKET_UNLOCK(bucket); 1715 return (s); 1716 } 1717 1718 static int 1719 dyn_add_ipv6_state(void *parent, uint32_t ruleid, uint16_t rulenum, 1720 const struct ipfw_flow_id *pkt, uint32_t zoneid, const void *ulp, 1721 int pktlen, uint32_t hashval, struct ipfw_dyn_info *info, 1722 uint16_t fibnum, uint16_t kidx, uint8_t type) 1723 { 1724 struct dyn_ipv6_state *s; 1725 struct dyn_data *data; 1726 uint32_t bucket; 1727 1728 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); 1729 DYN_BUCKET_LOCK(bucket); 1730 if (info->direction == MATCH_UNKNOWN || 1731 info->kidx != kidx || 1732 info->hashval != hashval || 1733 info->version != DYN_BUCKET_VERSION(bucket, ipv6_add)) { 1734 /* 1735 * Bucket version has been changed since last lookup, 1736 * do lookup again to be sure that state does not exist. 1737 */ 1738 if (dyn_lookup_ipv6_state_locked(pkt, zoneid, ulp, pktlen, 1739 bucket, kidx) != 0) { 1740 DYN_BUCKET_UNLOCK(bucket); 1741 return (EEXIST); 1742 } 1743 } 1744 1745 data = dyn_alloc_dyndata(parent, ruleid, rulenum, pkt, ulp, 1746 pktlen, hashval, fibnum); 1747 if (data == NULL) { 1748 DYN_BUCKET_UNLOCK(bucket); 1749 return (ENOMEM); 1750 } 1751 1752 s = dyn_alloc_ipv6_state(pkt, zoneid, kidx, type); 1753 if (s == NULL) { 1754 DYN_BUCKET_UNLOCK(bucket); 1755 uma_zfree(V_dyn_data_zone, data); 1756 return (ENOMEM); 1757 } 1758 1759 s->data = data; 1760 CK_SLIST_INSERT_HEAD(&V_dyn_ipv6[bucket], s, entry); 1761 DYN_COUNT_INC(dyn_count); 1762 DYN_BUCKET_VERSION_BUMP(bucket, ipv6_add); 1763 DYN_BUCKET_UNLOCK(bucket); 1764 return (0); 1765 } 1766 #endif /* INET6 */ 1767 1768 static void * 1769 dyn_get_parent_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, 1770 struct ip_fw *rule, uint32_t hashval, uint32_t limit, uint16_t kidx) 1771 { 1772 char sbuf[24]; 1773 struct dyn_parent *p; 1774 void *ret; 1775 uint32_t bucket, version; 1776 1777 p = NULL; 1778 ret = NULL; 1779 bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); 1780 DYNSTATE_CRITICAL_ENTER(); 1781 if (IS_IP4_FLOW_ID(pkt)) { 1782 struct dyn_ipv4_state *s; 1783 1784 version = DYN_BUCKET_VERSION(bucket, ipv4_parent_add); 1785 s = dyn_lookup_ipv4_parent(pkt, rule, rule->id, 1786 rule->rulenum, bucket); 1787 if (s == NULL) { 1788 /* 1789 * Exit from critical section because dyn_add_parent() 1790 * will acquire bucket lock. 1791 */ 1792 DYNSTATE_CRITICAL_EXIT(); 1793 1794 s = dyn_add_ipv4_parent(rule, rule->id, 1795 rule->rulenum, pkt, hashval, version, kidx); 1796 if (s == NULL) 1797 return (NULL); 1798 /* Now we are in critical section again. */ 1799 } 1800 ret = s; 1801 p = s->limit; 1802 } 1803 #ifdef INET6 1804 else if (IS_IP6_FLOW_ID(pkt)) { 1805 struct dyn_ipv6_state *s; 1806 1807 version = DYN_BUCKET_VERSION(bucket, ipv6_parent_add); 1808 s = dyn_lookup_ipv6_parent(pkt, zoneid, rule, rule->id, 1809 rule->rulenum, bucket); 1810 if (s == NULL) { 1811 /* 1812 * Exit from critical section because dyn_add_parent() 1813 * can acquire bucket mutex. 1814 */ 1815 DYNSTATE_CRITICAL_EXIT(); 1816 1817 s = dyn_add_ipv6_parent(rule, rule->id, 1818 rule->rulenum, pkt, zoneid, hashval, version, 1819 kidx); 1820 if (s == NULL) 1821 return (NULL); 1822 /* Now we are in critical section again. */ 1823 } 1824 ret = s; 1825 p = s->limit; 1826 } 1827 #endif 1828 else { 1829 DYNSTATE_CRITICAL_EXIT(); 1830 return (NULL); 1831 } 1832 1833 /* Check the limit */ 1834 if (DPARENT_COUNT(p) >= limit) { 1835 DYNSTATE_CRITICAL_EXIT(); 1836 if (V_fw_verbose && last_log != time_uptime) { 1837 last_log = time_uptime; 1838 snprintf(sbuf, sizeof(sbuf), "%u drop session", 1839 rule->rulenum); 1840 print_dyn_rule_flags(pkt, O_LIMIT, 1841 LOG_SECURITY | LOG_DEBUG, sbuf, 1842 "too many entries"); 1843 } 1844 return (NULL); 1845 } 1846 1847 /* Take new session into account. */ 1848 DPARENT_COUNT_INC(p); 1849 /* 1850 * We must exit from critical section because the following code 1851 * can acquire bucket mutex. 1852 * We rely on the 'count' field. The state will not expire 1853 * until it has some child states, i.e. 'count' field is not zero. 1854 * Return state pointer, it will be used by child states as parent. 1855 */ 1856 DYNSTATE_CRITICAL_EXIT(); 1857 return (ret); 1858 } 1859 1860 static int 1861 dyn_install_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, 1862 uint16_t fibnum, const void *ulp, int pktlen, struct ip_fw *rule, 1863 struct ipfw_dyn_info *info, uint32_t limit, uint16_t limit_mask, 1864 uint16_t kidx, uint8_t type) 1865 { 1866 struct ipfw_flow_id id; 1867 uint32_t hashval, parent_hashval, ruleid, rulenum; 1868 int ret; 1869 1870 MPASS(type == O_LIMIT || type == O_KEEP_STATE); 1871 1872 ruleid = rule->id; 1873 rulenum = rule->rulenum; 1874 if (type == O_LIMIT) { 1875 /* Create masked flow id and calculate bucket */ 1876 id.addr_type = pkt->addr_type; 1877 id.proto = pkt->proto; 1878 id.fib = fibnum; /* unused */ 1879 id.src_port = (limit_mask & DYN_SRC_PORT) ? 1880 pkt->src_port: 0; 1881 id.dst_port = (limit_mask & DYN_DST_PORT) ? 1882 pkt->dst_port: 0; 1883 if (IS_IP4_FLOW_ID(pkt)) { 1884 id.src_ip = (limit_mask & DYN_SRC_ADDR) ? 1885 pkt->src_ip: 0; 1886 id.dst_ip = (limit_mask & DYN_DST_ADDR) ? 1887 pkt->dst_ip: 0; 1888 } 1889 #ifdef INET6 1890 else if (IS_IP6_FLOW_ID(pkt)) { 1891 if (limit_mask & DYN_SRC_ADDR) 1892 id.src_ip6 = pkt->src_ip6; 1893 else 1894 memset(&id.src_ip6, 0, sizeof(id.src_ip6)); 1895 if (limit_mask & DYN_DST_ADDR) 1896 id.dst_ip6 = pkt->dst_ip6; 1897 else 1898 memset(&id.dst_ip6, 0, sizeof(id.dst_ip6)); 1899 } 1900 #endif 1901 else 1902 return (EAFNOSUPPORT); 1903 1904 parent_hashval = hash_parent(&id, rule); 1905 rule = dyn_get_parent_state(&id, zoneid, rule, parent_hashval, 1906 limit, kidx); 1907 if (rule == NULL) { 1908 #if 0 1909 if (V_fw_verbose && last_log != time_uptime) { 1910 last_log = time_uptime; 1911 snprintf(sbuf, sizeof(sbuf), 1912 "%u drop session", rule->rulenum); 1913 print_dyn_rule_flags(pkt, O_LIMIT, 1914 LOG_SECURITY | LOG_DEBUG, sbuf, 1915 "too many entries"); 1916 } 1917 #endif 1918 return (EACCES); 1919 } 1920 /* 1921 * Limit is not reached, create new state. 1922 * Now rule points to parent state. 1923 */ 1924 } 1925 1926 hashval = hash_packet(pkt); 1927 if (IS_IP4_FLOW_ID(pkt)) 1928 ret = dyn_add_ipv4_state(rule, ruleid, rulenum, pkt, 1929 ulp, pktlen, hashval, info, fibnum, kidx, type); 1930 #ifdef INET6 1931 else if (IS_IP6_FLOW_ID(pkt)) 1932 ret = dyn_add_ipv6_state(rule, ruleid, rulenum, pkt, 1933 zoneid, ulp, pktlen, hashval, info, fibnum, kidx, type); 1934 #endif /* INET6 */ 1935 else 1936 ret = EAFNOSUPPORT; 1937 1938 if (type == O_LIMIT) { 1939 if (ret != 0) { 1940 /* 1941 * We failed to create child state for O_LIMIT 1942 * opcode. Since we already counted it in the parent, 1943 * we must revert counter back. The 'rule' points to 1944 * parent state, use it to get dyn_parent. 1945 * 1946 * XXXAE: it should be safe to use 'rule' pointer 1947 * without extra lookup, parent state is referenced 1948 * and should not be freed. 1949 */ 1950 if (IS_IP4_FLOW_ID(&id)) 1951 DPARENT_COUNT_DEC( 1952 ((struct dyn_ipv4_state *)rule)->limit); 1953 #ifdef INET6 1954 else if (IS_IP6_FLOW_ID(&id)) 1955 DPARENT_COUNT_DEC( 1956 ((struct dyn_ipv6_state *)rule)->limit); 1957 #endif 1958 } 1959 } 1960 /* 1961 * EEXIST means that simultaneous thread has created this 1962 * state. Consider this as success. 1963 * 1964 * XXXAE: should we invalidate 'info' content here? 1965 */ 1966 if (ret == EEXIST) 1967 return (0); 1968 return (ret); 1969 } 1970 1971 /* 1972 * Install dynamic state. 1973 * chain - ipfw's instance; 1974 * rule - the parent rule that installs the state; 1975 * cmd - opcode that installs the state; 1976 * args - ipfw arguments; 1977 * ulp - upper level protocol header; 1978 * pktlen - packet length; 1979 * info - dynamic state lookup info; 1980 * tablearg - tablearg id. 1981 * 1982 * Returns non-zero value (failure) if state is not installed because 1983 * of errors or because session limitations are enforced. 1984 */ 1985 int 1986 ipfw_dyn_install_state(struct ip_fw_chain *chain, struct ip_fw *rule, 1987 const ipfw_insn_limit *cmd, const struct ip_fw_args *args, 1988 const void *ulp, int pktlen, struct ipfw_dyn_info *info, 1989 uint32_t tablearg) 1990 { 1991 uint32_t limit; 1992 uint16_t limit_mask; 1993 1994 if (cmd->o.opcode == O_LIMIT) { 1995 limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit); 1996 limit_mask = cmd->limit_mask; 1997 } else { 1998 limit = 0; 1999 limit_mask = 0; 2000 } 2001 return (dyn_install_state(&args->f_id, 2002 #ifdef INET6 2003 IS_IP6_FLOW_ID(&args->f_id) ? dyn_getscopeid(args): 2004 #endif 2005 0, M_GETFIB(args->m), ulp, pktlen, rule, info, limit, 2006 limit_mask, cmd->o.arg1, cmd->o.opcode)); 2007 } 2008 2009 /* 2010 * Free safe to remove state entries from expired lists. 2011 */ 2012 static void 2013 dyn_free_states(struct ip_fw_chain *chain) 2014 { 2015 struct dyn_ipv4_state *s4, *s4n; 2016 #ifdef INET6 2017 struct dyn_ipv6_state *s6, *s6n; 2018 #endif 2019 int cached_count, i; 2020 2021 /* 2022 * We keep pointers to objects that are in use on each CPU 2023 * in the per-cpu dyn_hp pointer. When object is going to be 2024 * removed, first of it is unlinked from the corresponding 2025 * list. This leads to changing of dyn_bucket_xxx_delver version. 2026 * Unlinked objects is placed into corresponding dyn_expired_xxx 2027 * list. Reader that is going to dereference object pointer checks 2028 * dyn_bucket_xxx_delver version before and after storing pointer 2029 * into dyn_hp. If version is the same, the object is protected 2030 * from freeing and it is safe to dereference. Othervise reader 2031 * tries to iterate list again from the beginning, but this object 2032 * now unlinked and thus will not be accessible. 2033 * 2034 * Copy dyn_hp pointers for each CPU into dyn_hp_cache array. 2035 * It does not matter that some pointer can be changed in 2036 * time while we are copying. We need to check, that objects 2037 * removed in the previous pass are not in use. And if dyn_hp 2038 * pointer does not contain it in the time when we are copying, 2039 * it will not appear there, because it is already unlinked. 2040 * And for new pointers we will not free objects that will be 2041 * unlinked in this pass. 2042 */ 2043 cached_count = 0; 2044 CPU_FOREACH(i) { 2045 dyn_hp_cache[cached_count] = DYNSTATE_GET(i); 2046 if (dyn_hp_cache[cached_count] != NULL) 2047 cached_count++; 2048 } 2049 2050 /* 2051 * Free expired states that are safe to free. 2052 * Check each entry from previous pass in the dyn_expired_xxx 2053 * list, if pointer to the object is in the dyn_hp_cache array, 2054 * keep it until next pass. Otherwise it is safe to free the 2055 * object. 2056 * 2057 * XXXAE: optimize this to use SLIST_REMOVE_AFTER. 2058 */ 2059 #define DYN_FREE_STATES(s, next, name) do { \ 2060 s = SLIST_FIRST(&V_dyn_expired_ ## name); \ 2061 while (s != NULL) { \ 2062 next = SLIST_NEXT(s, expired); \ 2063 for (i = 0; i < cached_count; i++) \ 2064 if (dyn_hp_cache[i] == s) \ 2065 break; \ 2066 if (i == cached_count) { \ 2067 if (s->type == O_LIMIT_PARENT && \ 2068 s->limit->count != 0) { \ 2069 s = next; \ 2070 continue; \ 2071 } \ 2072 SLIST_REMOVE(&V_dyn_expired_ ## name, \ 2073 s, dyn_ ## name ## _state, expired); \ 2074 if (s->type == O_LIMIT_PARENT) \ 2075 uma_zfree(V_dyn_parent_zone, s->limit); \ 2076 else \ 2077 uma_zfree(V_dyn_data_zone, s->data); \ 2078 uma_zfree(V_dyn_ ## name ## _zone, s); \ 2079 } \ 2080 s = next; \ 2081 } \ 2082 } while (0) 2083 2084 /* 2085 * Protect access to expired lists with DYN_EXPIRED_LOCK. 2086 * Userland can invoke ipfw_expire_dyn_states() to delete 2087 * specific states, this will lead to modification of expired 2088 * lists. 2089 * 2090 * XXXAE: do we need DYN_EXPIRED_LOCK? We can just use 2091 * IPFW_UH_WLOCK to protect access to these lists. 2092 */ 2093 DYN_EXPIRED_LOCK(); 2094 DYN_FREE_STATES(s4, s4n, ipv4); 2095 #ifdef INET6 2096 DYN_FREE_STATES(s6, s6n, ipv6); 2097 #endif 2098 DYN_EXPIRED_UNLOCK(); 2099 #undef DYN_FREE_STATES 2100 } 2101 2102 /* 2103 * Returns: 2104 * 0 when state is not matched by specified range; 2105 * 1 when state is matched by specified range; 2106 * 2 when state is matched by specified range and requested deletion of 2107 * dynamic states. 2108 */ 2109 static int 2110 dyn_match_range(uint16_t rulenum, uint8_t set, const ipfw_range_tlv *rt) 2111 { 2112 2113 MPASS(rt != NULL); 2114 /* flush all states */ 2115 if (rt->flags & IPFW_RCFLAG_ALL) { 2116 if (rt->flags & IPFW_RCFLAG_DYNAMIC) 2117 return (2); /* forced */ 2118 return (1); 2119 } 2120 if ((rt->flags & IPFW_RCFLAG_SET) != 0 && set != rt->set) 2121 return (0); 2122 if ((rt->flags & IPFW_RCFLAG_RANGE) != 0 && 2123 (rulenum < rt->start_rule || rulenum > rt->end_rule)) 2124 return (0); 2125 if (rt->flags & IPFW_RCFLAG_DYNAMIC) 2126 return (2); 2127 return (1); 2128 } 2129 2130 static void 2131 dyn_acquire_rule(struct ip_fw_chain *ch, struct dyn_data *data, 2132 struct ip_fw *rule, uint16_t kidx) 2133 { 2134 struct dyn_state_obj *obj; 2135 2136 /* 2137 * Do not acquire reference twice. 2138 * This can happen when rule deletion executed for 2139 * the same range, but different ruleset id. 2140 */ 2141 if (data->flags & DYN_REFERENCED) 2142 return; 2143 2144 IPFW_UH_WLOCK_ASSERT(ch); 2145 MPASS(kidx != 0); 2146 2147 data->flags |= DYN_REFERENCED; 2148 /* Reference the named object */ 2149 obj = SRV_OBJECT(ch, kidx); 2150 obj->no.refcnt++; 2151 MPASS(obj->no.etlv == IPFW_TLV_STATE_NAME); 2152 2153 /* Reference the parent rule */ 2154 rule->refcnt++; 2155 } 2156 2157 static void 2158 dyn_release_rule(struct ip_fw_chain *ch, struct dyn_data *data, 2159 struct ip_fw *rule, uint16_t kidx) 2160 { 2161 struct dyn_state_obj *obj; 2162 2163 IPFW_UH_WLOCK_ASSERT(ch); 2164 MPASS(kidx != 0); 2165 2166 obj = SRV_OBJECT(ch, kidx); 2167 if (obj->no.refcnt == 1) 2168 dyn_destroy(ch, &obj->no); 2169 else 2170 obj->no.refcnt--; 2171 2172 if (--rule->refcnt == 1) 2173 ipfw_free_rule(rule); 2174 } 2175 2176 /* 2177 * We do not keep O_LIMIT_PARENT states when V_dyn_keep_states is enabled. 2178 * O_LIMIT state is created when new connection is going to be established 2179 * and there is no matching state. So, since the old parent rule was deleted 2180 * we can't create new states with old parent, and thus we can not account 2181 * new connections with already established connections, and can not do 2182 * proper limiting. 2183 */ 2184 static int 2185 dyn_match_ipv4_state(struct ip_fw_chain *ch, struct dyn_ipv4_state *s, 2186 const ipfw_range_tlv *rt) 2187 { 2188 struct ip_fw *rule; 2189 int ret; 2190 2191 if (s->type == O_LIMIT_PARENT) { 2192 rule = s->limit->parent; 2193 return (dyn_match_range(s->limit->rulenum, rule->set, rt)); 2194 } 2195 2196 rule = s->data->parent; 2197 if (s->type == O_LIMIT) 2198 rule = ((struct dyn_ipv4_state *)rule)->limit->parent; 2199 2200 ret = dyn_match_range(s->data->rulenum, rule->set, rt); 2201 if (ret == 0 || V_dyn_keep_states == 0 || ret > 1) 2202 return (ret); 2203 2204 dyn_acquire_rule(ch, s->data, rule, s->kidx); 2205 return (0); 2206 } 2207 2208 #ifdef INET6 2209 static int 2210 dyn_match_ipv6_state(struct ip_fw_chain *ch, struct dyn_ipv6_state *s, 2211 const ipfw_range_tlv *rt) 2212 { 2213 struct ip_fw *rule; 2214 int ret; 2215 2216 if (s->type == O_LIMIT_PARENT) { 2217 rule = s->limit->parent; 2218 return (dyn_match_range(s->limit->rulenum, rule->set, rt)); 2219 } 2220 2221 rule = s->data->parent; 2222 if (s->type == O_LIMIT) 2223 rule = ((struct dyn_ipv6_state *)rule)->limit->parent; 2224 2225 ret = dyn_match_range(s->data->rulenum, rule->set, rt); 2226 if (ret == 0 || V_dyn_keep_states == 0 || ret > 1) 2227 return (ret); 2228 2229 dyn_acquire_rule(ch, s->data, rule, s->kidx); 2230 return (0); 2231 } 2232 #endif 2233 2234 /* 2235 * Unlink expired entries from states lists. 2236 * @rt can be used to specify the range of states for deletion. 2237 */ 2238 static void 2239 dyn_expire_states(struct ip_fw_chain *ch, ipfw_range_tlv *rt) 2240 { 2241 struct dyn_ipv4_slist expired_ipv4; 2242 #ifdef INET6 2243 struct dyn_ipv6_slist expired_ipv6; 2244 struct dyn_ipv6_state *s6, *s6n, *s6p; 2245 #endif 2246 struct dyn_ipv4_state *s4, *s4n, *s4p; 2247 void *rule; 2248 int bucket, removed, length, max_length; 2249 2250 IPFW_UH_WLOCK_ASSERT(ch); 2251 2252 /* 2253 * Unlink expired states from each bucket. 2254 * With acquired bucket lock iterate entries of each lists: 2255 * ipv4, ipv4_parent, ipv6, and ipv6_parent. Check expired time 2256 * and unlink entry from the list, link entry into temporary 2257 * expired_xxx lists then bump "del" bucket version. 2258 * 2259 * When an entry is removed, corresponding states counter is 2260 * decremented. If entry has O_LIMIT type, parent's reference 2261 * counter is decremented. 2262 * 2263 * NOTE: this function can be called from userspace context 2264 * when user deletes rules. In this case all matched states 2265 * will be forcedly unlinked. O_LIMIT_PARENT states will be kept 2266 * in the expired lists until reference counter become zero. 2267 */ 2268 #define DYN_UNLINK_STATES(s, prev, next, exp, af, name, extra) do { \ 2269 length = 0; \ 2270 removed = 0; \ 2271 prev = NULL; \ 2272 s = CK_SLIST_FIRST(&V_dyn_ ## name [bucket]); \ 2273 while (s != NULL) { \ 2274 next = CK_SLIST_NEXT(s, entry); \ 2275 if ((TIME_LEQ((s)->exp, time_uptime) && extra) || \ 2276 (rt != NULL && \ 2277 dyn_match_ ## af ## _state(ch, s, rt))) { \ 2278 if (prev != NULL) \ 2279 CK_SLIST_REMOVE_AFTER(prev, entry); \ 2280 else \ 2281 CK_SLIST_REMOVE_HEAD( \ 2282 &V_dyn_ ## name [bucket], entry); \ 2283 removed++; \ 2284 SLIST_INSERT_HEAD(&expired_ ## af, s, expired); \ 2285 if (s->type == O_LIMIT_PARENT) \ 2286 DYN_COUNT_DEC(dyn_parent_count); \ 2287 else { \ 2288 DYN_COUNT_DEC(dyn_count); \ 2289 if (s->data->flags & DYN_REFERENCED) { \ 2290 rule = s->data->parent; \ 2291 if (s->type == O_LIMIT) \ 2292 rule = ((__typeof(s)) \ 2293 rule)->limit->parent;\ 2294 dyn_release_rule(ch, s->data, \ 2295 rule, s->kidx); \ 2296 } \ 2297 if (s->type == O_LIMIT) { \ 2298 s = s->data->parent; \ 2299 DPARENT_COUNT_DEC(s->limit); \ 2300 } \ 2301 } \ 2302 } else { \ 2303 prev = s; \ 2304 length++; \ 2305 } \ 2306 s = next; \ 2307 } \ 2308 if (removed != 0) \ 2309 DYN_BUCKET_VERSION_BUMP(bucket, name ## _del); \ 2310 if (length > max_length) \ 2311 max_length = length; \ 2312 } while (0) 2313 2314 SLIST_INIT(&expired_ipv4); 2315 #ifdef INET6 2316 SLIST_INIT(&expired_ipv6); 2317 #endif 2318 max_length = 0; 2319 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 2320 DYN_BUCKET_LOCK(bucket); 2321 DYN_UNLINK_STATES(s4, s4p, s4n, data->expire, ipv4, ipv4, 1); 2322 DYN_UNLINK_STATES(s4, s4p, s4n, limit->expire, ipv4, 2323 ipv4_parent, (s4->limit->count == 0)); 2324 #ifdef INET6 2325 DYN_UNLINK_STATES(s6, s6p, s6n, data->expire, ipv6, ipv6, 1); 2326 DYN_UNLINK_STATES(s6, s6p, s6n, limit->expire, ipv6, 2327 ipv6_parent, (s6->limit->count == 0)); 2328 #endif 2329 DYN_BUCKET_UNLOCK(bucket); 2330 } 2331 /* Update curr_max_length for statistics. */ 2332 V_curr_max_length = max_length; 2333 /* 2334 * Concatenate temporary lists with global expired lists. 2335 */ 2336 DYN_EXPIRED_LOCK(); 2337 SLIST_CONCAT(&V_dyn_expired_ipv4, &expired_ipv4, 2338 dyn_ipv4_state, expired); 2339 #ifdef INET6 2340 SLIST_CONCAT(&V_dyn_expired_ipv6, &expired_ipv6, 2341 dyn_ipv6_state, expired); 2342 #endif 2343 DYN_EXPIRED_UNLOCK(); 2344 #undef DYN_UNLINK_STATES 2345 #undef DYN_UNREF_STATES 2346 } 2347 2348 static struct mbuf * 2349 dyn_mgethdr(int len, uint16_t fibnum) 2350 { 2351 struct mbuf *m; 2352 2353 m = m_gethdr(M_NOWAIT, MT_DATA); 2354 if (m == NULL) 2355 return (NULL); 2356 #ifdef MAC 2357 mac_netinet_firewall_send(m); 2358 #endif 2359 M_SETFIB(m, fibnum); 2360 m->m_data += max_linkhdr; 2361 m->m_flags |= M_SKIP_FIREWALL; 2362 m->m_len = m->m_pkthdr.len = len; 2363 bzero(m->m_data, len); 2364 return (m); 2365 } 2366 2367 static void 2368 dyn_make_keepalive_ipv4(struct mbuf *m, in_addr_t src, in_addr_t dst, 2369 uint32_t seq, uint32_t ack, uint16_t sport, uint16_t dport) 2370 { 2371 struct tcphdr *tcp; 2372 struct ip *ip; 2373 2374 ip = mtod(m, struct ip *); 2375 ip->ip_v = 4; 2376 ip->ip_hl = sizeof(*ip) >> 2; 2377 ip->ip_tos = IPTOS_LOWDELAY; 2378 ip->ip_len = htons(m->m_len); 2379 ip->ip_off |= htons(IP_DF); 2380 ip->ip_ttl = V_ip_defttl; 2381 ip->ip_p = IPPROTO_TCP; 2382 ip->ip_src.s_addr = htonl(src); 2383 ip->ip_dst.s_addr = htonl(dst); 2384 2385 tcp = mtodo(m, sizeof(struct ip)); 2386 tcp->th_sport = htons(sport); 2387 tcp->th_dport = htons(dport); 2388 tcp->th_off = sizeof(struct tcphdr) >> 2; 2389 tcp->th_seq = htonl(seq); 2390 tcp->th_ack = htonl(ack); 2391 tcp->th_flags = TH_ACK; 2392 tcp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2393 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 2394 2395 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2396 m->m_pkthdr.csum_flags = CSUM_TCP; 2397 } 2398 2399 static void 2400 dyn_enqueue_keepalive_ipv4(struct mbufq *q, const struct dyn_ipv4_state *s) 2401 { 2402 struct mbuf *m; 2403 2404 if ((s->data->state & ACK_FWD) == 0 && s->data->ack_fwd > 0) { 2405 m = dyn_mgethdr(sizeof(struct ip) + sizeof(struct tcphdr), 2406 s->data->fibnum); 2407 if (m != NULL) { 2408 dyn_make_keepalive_ipv4(m, s->dst, s->src, 2409 s->data->ack_fwd - 1, s->data->ack_rev, 2410 s->dport, s->sport); 2411 if (mbufq_enqueue(q, m)) { 2412 m_freem(m); 2413 log(LOG_DEBUG, "ipfw: limit for IPv4 " 2414 "keepalive queue is reached.\n"); 2415 return; 2416 } 2417 } 2418 } 2419 2420 if ((s->data->state & ACK_REV) == 0 && s->data->ack_rev > 0) { 2421 m = dyn_mgethdr(sizeof(struct ip) + sizeof(struct tcphdr), 2422 s->data->fibnum); 2423 if (m != NULL) { 2424 dyn_make_keepalive_ipv4(m, s->src, s->dst, 2425 s->data->ack_rev - 1, s->data->ack_fwd, 2426 s->sport, s->dport); 2427 if (mbufq_enqueue(q, m)) { 2428 m_freem(m); 2429 log(LOG_DEBUG, "ipfw: limit for IPv4 " 2430 "keepalive queue is reached.\n"); 2431 return; 2432 } 2433 } 2434 } 2435 } 2436 2437 /* 2438 * Prepare and send keep-alive packets. 2439 */ 2440 static void 2441 dyn_send_keepalive_ipv4(struct ip_fw_chain *chain) 2442 { 2443 struct mbufq q; 2444 struct mbuf *m; 2445 struct dyn_ipv4_state *s; 2446 uint32_t bucket; 2447 2448 mbufq_init(&q, INT_MAX); 2449 IPFW_UH_RLOCK(chain); 2450 /* 2451 * It is safe to not use hazard pointer and just do lockless 2452 * access to the lists, because states entries can not be deleted 2453 * while we hold IPFW_UH_RLOCK. 2454 */ 2455 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 2456 CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) { 2457 /* 2458 * Only established TCP connections that will 2459 * become expired within dyn_keepalive_interval. 2460 */ 2461 if (s->proto != IPPROTO_TCP || 2462 (s->data->state & BOTH_SYN) != BOTH_SYN || 2463 TIME_LEQ(time_uptime + V_dyn_keepalive_interval, 2464 s->data->expire)) 2465 continue; 2466 dyn_enqueue_keepalive_ipv4(&q, s); 2467 } 2468 } 2469 IPFW_UH_RUNLOCK(chain); 2470 while ((m = mbufq_dequeue(&q)) != NULL) 2471 ip_output(m, NULL, NULL, 0, NULL, NULL); 2472 } 2473 2474 #ifdef INET6 2475 static void 2476 dyn_make_keepalive_ipv6(struct mbuf *m, const struct in6_addr *src, 2477 const struct in6_addr *dst, uint32_t zoneid, uint32_t seq, uint32_t ack, 2478 uint16_t sport, uint16_t dport) 2479 { 2480 struct tcphdr *tcp; 2481 struct ip6_hdr *ip6; 2482 2483 ip6 = mtod(m, struct ip6_hdr *); 2484 ip6->ip6_vfc |= IPV6_VERSION; 2485 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 2486 ip6->ip6_nxt = IPPROTO_TCP; 2487 ip6->ip6_hlim = IPV6_DEFHLIM; 2488 ip6->ip6_src = *src; 2489 if (IN6_IS_ADDR_LINKLOCAL(src)) 2490 ip6->ip6_src.s6_addr16[1] = htons(zoneid & 0xffff); 2491 ip6->ip6_dst = *dst; 2492 if (IN6_IS_ADDR_LINKLOCAL(dst)) 2493 ip6->ip6_dst.s6_addr16[1] = htons(zoneid & 0xffff); 2494 2495 tcp = mtodo(m, sizeof(struct ip6_hdr)); 2496 tcp->th_sport = htons(sport); 2497 tcp->th_dport = htons(dport); 2498 tcp->th_off = sizeof(struct tcphdr) >> 2; 2499 tcp->th_seq = htonl(seq); 2500 tcp->th_ack = htonl(ack); 2501 tcp->th_flags = TH_ACK; 2502 tcp->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr), 2503 IPPROTO_TCP, 0); 2504 2505 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2506 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2507 } 2508 2509 static void 2510 dyn_enqueue_keepalive_ipv6(struct mbufq *q, const struct dyn_ipv6_state *s) 2511 { 2512 struct mbuf *m; 2513 2514 if ((s->data->state & ACK_FWD) == 0 && s->data->ack_fwd > 0) { 2515 m = dyn_mgethdr(sizeof(struct ip6_hdr) + 2516 sizeof(struct tcphdr), s->data->fibnum); 2517 if (m != NULL) { 2518 dyn_make_keepalive_ipv6(m, &s->dst, &s->src, 2519 s->zoneid, s->data->ack_fwd - 1, s->data->ack_rev, 2520 s->dport, s->sport); 2521 if (mbufq_enqueue(q, m)) { 2522 m_freem(m); 2523 log(LOG_DEBUG, "ipfw: limit for IPv6 " 2524 "keepalive queue is reached.\n"); 2525 return; 2526 } 2527 } 2528 } 2529 2530 if ((s->data->state & ACK_REV) == 0 && s->data->ack_rev > 0) { 2531 m = dyn_mgethdr(sizeof(struct ip6_hdr) + 2532 sizeof(struct tcphdr), s->data->fibnum); 2533 if (m != NULL) { 2534 dyn_make_keepalive_ipv6(m, &s->src, &s->dst, 2535 s->zoneid, s->data->ack_rev - 1, s->data->ack_fwd, 2536 s->sport, s->dport); 2537 if (mbufq_enqueue(q, m)) { 2538 m_freem(m); 2539 log(LOG_DEBUG, "ipfw: limit for IPv6 " 2540 "keepalive queue is reached.\n"); 2541 return; 2542 } 2543 } 2544 } 2545 } 2546 2547 static void 2548 dyn_send_keepalive_ipv6(struct ip_fw_chain *chain) 2549 { 2550 struct mbufq q; 2551 struct mbuf *m; 2552 struct dyn_ipv6_state *s; 2553 uint32_t bucket; 2554 2555 mbufq_init(&q, INT_MAX); 2556 IPFW_UH_RLOCK(chain); 2557 /* 2558 * It is safe to not use hazard pointer and just do lockless 2559 * access to the lists, because states entries can not be deleted 2560 * while we hold IPFW_UH_RLOCK. 2561 */ 2562 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 2563 CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) { 2564 /* 2565 * Only established TCP connections that will 2566 * become expired within dyn_keepalive_interval. 2567 */ 2568 if (s->proto != IPPROTO_TCP || 2569 (s->data->state & BOTH_SYN) != BOTH_SYN || 2570 TIME_LEQ(time_uptime + V_dyn_keepalive_interval, 2571 s->data->expire)) 2572 continue; 2573 dyn_enqueue_keepalive_ipv6(&q, s); 2574 } 2575 } 2576 IPFW_UH_RUNLOCK(chain); 2577 while ((m = mbufq_dequeue(&q)) != NULL) 2578 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 2579 } 2580 #endif /* INET6 */ 2581 2582 static void 2583 dyn_grow_hashtable(struct ip_fw_chain *chain, uint32_t new, int flags) 2584 { 2585 #ifdef INET6 2586 struct dyn_ipv6ck_slist *ipv6, *ipv6_parent; 2587 uint32_t *ipv6_add, *ipv6_del, *ipv6_parent_add, *ipv6_parent_del; 2588 struct dyn_ipv6_state *s6; 2589 #endif 2590 struct dyn_ipv4ck_slist *ipv4, *ipv4_parent; 2591 uint32_t *ipv4_add, *ipv4_del, *ipv4_parent_add, *ipv4_parent_del; 2592 struct dyn_ipv4_state *s4; 2593 struct mtx *bucket_lock; 2594 void *tmp; 2595 uint32_t bucket; 2596 2597 MPASS(powerof2(new)); 2598 DYN_DEBUG("grow hash size %u -> %u", V_curr_dyn_buckets, new); 2599 /* 2600 * Allocate and initialize new lists. 2601 */ 2602 bucket_lock = malloc(new * sizeof(struct mtx), M_IPFW, 2603 flags | M_ZERO); 2604 if (bucket_lock == NULL) 2605 return; 2606 2607 ipv4 = ipv4_parent = NULL; 2608 ipv4_add = ipv4_del = ipv4_parent_add = ipv4_parent_del = NULL; 2609 #ifdef INET6 2610 ipv6 = ipv6_parent = NULL; 2611 ipv6_add = ipv6_del = ipv6_parent_add = ipv6_parent_del = NULL; 2612 #endif 2613 2614 ipv4 = malloc(new * sizeof(struct dyn_ipv4ck_slist), M_IPFW, 2615 flags | M_ZERO); 2616 if (ipv4 == NULL) 2617 goto bad; 2618 ipv4_parent = malloc(new * sizeof(struct dyn_ipv4ck_slist), M_IPFW, 2619 flags | M_ZERO); 2620 if (ipv4_parent == NULL) 2621 goto bad; 2622 ipv4_add = malloc(new * sizeof(uint32_t), M_IPFW, flags | M_ZERO); 2623 if (ipv4_add == NULL) 2624 goto bad; 2625 ipv4_del = malloc(new * sizeof(uint32_t), M_IPFW, flags | M_ZERO); 2626 if (ipv4_del == NULL) 2627 goto bad; 2628 ipv4_parent_add = malloc(new * sizeof(uint32_t), M_IPFW, 2629 flags | M_ZERO); 2630 if (ipv4_parent_add == NULL) 2631 goto bad; 2632 ipv4_parent_del = malloc(new * sizeof(uint32_t), M_IPFW, 2633 flags | M_ZERO); 2634 if (ipv4_parent_del == NULL) 2635 goto bad; 2636 #ifdef INET6 2637 ipv6 = malloc(new * sizeof(struct dyn_ipv6ck_slist), M_IPFW, 2638 flags | M_ZERO); 2639 if (ipv6 == NULL) 2640 goto bad; 2641 ipv6_parent = malloc(new * sizeof(struct dyn_ipv6ck_slist), M_IPFW, 2642 flags | M_ZERO); 2643 if (ipv6_parent == NULL) 2644 goto bad; 2645 ipv6_add = malloc(new * sizeof(uint32_t), M_IPFW, flags | M_ZERO); 2646 if (ipv6_add == NULL) 2647 goto bad; 2648 ipv6_del = malloc(new * sizeof(uint32_t), M_IPFW, flags | M_ZERO); 2649 if (ipv6_del == NULL) 2650 goto bad; 2651 ipv6_parent_add = malloc(new * sizeof(uint32_t), M_IPFW, 2652 flags | M_ZERO); 2653 if (ipv6_parent_add == NULL) 2654 goto bad; 2655 ipv6_parent_del = malloc(new * sizeof(uint32_t), M_IPFW, 2656 flags | M_ZERO); 2657 if (ipv6_parent_del == NULL) 2658 goto bad; 2659 #endif 2660 for (bucket = 0; bucket < new; bucket++) { 2661 DYN_BUCKET_LOCK_INIT(bucket_lock, bucket); 2662 CK_SLIST_INIT(&ipv4[bucket]); 2663 CK_SLIST_INIT(&ipv4_parent[bucket]); 2664 #ifdef INET6 2665 CK_SLIST_INIT(&ipv6[bucket]); 2666 CK_SLIST_INIT(&ipv6_parent[bucket]); 2667 #endif 2668 } 2669 2670 #define DYN_RELINK_STATES(s, hval, i, head, ohead) do { \ 2671 while ((s = CK_SLIST_FIRST(&V_dyn_ ## ohead[i])) != NULL) { \ 2672 CK_SLIST_REMOVE_HEAD(&V_dyn_ ## ohead[i], entry); \ 2673 CK_SLIST_INSERT_HEAD(&head[DYN_BUCKET(s->hval, new)], \ 2674 s, entry); \ 2675 } \ 2676 } while (0) 2677 /* 2678 * Prevent rules changing from userland. 2679 */ 2680 IPFW_UH_WLOCK(chain); 2681 /* 2682 * Hold traffic processing until we finish resize to 2683 * prevent access to states lists. 2684 */ 2685 IPFW_WLOCK(chain); 2686 /* Re-link all dynamic states */ 2687 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 2688 DYN_RELINK_STATES(s4, data->hashval, bucket, ipv4, ipv4); 2689 DYN_RELINK_STATES(s4, limit->hashval, bucket, ipv4_parent, 2690 ipv4_parent); 2691 #ifdef INET6 2692 DYN_RELINK_STATES(s6, data->hashval, bucket, ipv6, ipv6); 2693 DYN_RELINK_STATES(s6, limit->hashval, bucket, ipv6_parent, 2694 ipv6_parent); 2695 #endif 2696 } 2697 2698 #define DYN_SWAP_PTR(old, new, tmp) do { \ 2699 tmp = old; \ 2700 old = new; \ 2701 new = tmp; \ 2702 } while (0) 2703 /* Swap pointers */ 2704 DYN_SWAP_PTR(V_dyn_bucket_lock, bucket_lock, tmp); 2705 DYN_SWAP_PTR(V_dyn_ipv4, ipv4, tmp); 2706 DYN_SWAP_PTR(V_dyn_ipv4_parent, ipv4_parent, tmp); 2707 DYN_SWAP_PTR(V_dyn_ipv4_add, ipv4_add, tmp); 2708 DYN_SWAP_PTR(V_dyn_ipv4_parent_add, ipv4_parent_add, tmp); 2709 DYN_SWAP_PTR(V_dyn_ipv4_del, ipv4_del, tmp); 2710 DYN_SWAP_PTR(V_dyn_ipv4_parent_del, ipv4_parent_del, tmp); 2711 2712 #ifdef INET6 2713 DYN_SWAP_PTR(V_dyn_ipv6, ipv6, tmp); 2714 DYN_SWAP_PTR(V_dyn_ipv6_parent, ipv6_parent, tmp); 2715 DYN_SWAP_PTR(V_dyn_ipv6_add, ipv6_add, tmp); 2716 DYN_SWAP_PTR(V_dyn_ipv6_parent_add, ipv6_parent_add, tmp); 2717 DYN_SWAP_PTR(V_dyn_ipv6_del, ipv6_del, tmp); 2718 DYN_SWAP_PTR(V_dyn_ipv6_parent_del, ipv6_parent_del, tmp); 2719 #endif 2720 bucket = V_curr_dyn_buckets; 2721 V_curr_dyn_buckets = new; 2722 2723 IPFW_WUNLOCK(chain); 2724 IPFW_UH_WUNLOCK(chain); 2725 2726 /* Release old resources */ 2727 while (bucket-- != 0) 2728 DYN_BUCKET_LOCK_DESTROY(bucket_lock, bucket); 2729 bad: 2730 free(bucket_lock, M_IPFW); 2731 free(ipv4, M_IPFW); 2732 free(ipv4_parent, M_IPFW); 2733 free(ipv4_add, M_IPFW); 2734 free(ipv4_parent_add, M_IPFW); 2735 free(ipv4_del, M_IPFW); 2736 free(ipv4_parent_del, M_IPFW); 2737 #ifdef INET6 2738 free(ipv6, M_IPFW); 2739 free(ipv6_parent, M_IPFW); 2740 free(ipv6_add, M_IPFW); 2741 free(ipv6_parent_add, M_IPFW); 2742 free(ipv6_del, M_IPFW); 2743 free(ipv6_parent_del, M_IPFW); 2744 #endif 2745 } 2746 2747 /* 2748 * This function is used to perform various maintenance 2749 * on dynamic hash lists. Currently it is called every second. 2750 */ 2751 static void 2752 dyn_tick(void *vnetx) 2753 { 2754 struct epoch_tracker et; 2755 uint32_t buckets; 2756 2757 CURVNET_SET((struct vnet *)vnetx); 2758 /* 2759 * First free states unlinked in previous passes. 2760 */ 2761 dyn_free_states(&V_layer3_chain); 2762 /* 2763 * Now unlink others expired states. 2764 * We use IPFW_UH_WLOCK to avoid concurrent call of 2765 * dyn_expire_states(). It is the only function that does 2766 * deletion of state entries from states lists. 2767 */ 2768 IPFW_UH_WLOCK(&V_layer3_chain); 2769 dyn_expire_states(&V_layer3_chain, NULL); 2770 IPFW_UH_WUNLOCK(&V_layer3_chain); 2771 /* 2772 * Send keepalives if they are enabled and the time has come. 2773 */ 2774 if (V_dyn_keepalive != 0 && 2775 V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) { 2776 V_dyn_keepalive_last = time_uptime; 2777 NET_EPOCH_ENTER(et); 2778 dyn_send_keepalive_ipv4(&V_layer3_chain); 2779 #ifdef INET6 2780 dyn_send_keepalive_ipv6(&V_layer3_chain); 2781 #endif 2782 NET_EPOCH_EXIT(et); 2783 } 2784 /* 2785 * Check if we need to resize the hash: 2786 * if current number of states exceeds number of buckets in hash, 2787 * and dyn_buckets_max permits to grow the number of buckets, then 2788 * do it. Grow hash size to the minimum power of 2 which is bigger 2789 * than current states count. 2790 */ 2791 if (V_curr_dyn_buckets < V_dyn_buckets_max && 2792 (V_curr_dyn_buckets < V_dyn_count / 2 || ( 2793 V_curr_dyn_buckets < V_dyn_count && V_curr_max_length > 8))) { 2794 buckets = 1 << fls(V_dyn_count); 2795 if (buckets > V_dyn_buckets_max) 2796 buckets = V_dyn_buckets_max; 2797 dyn_grow_hashtable(&V_layer3_chain, buckets, M_NOWAIT); 2798 } 2799 2800 callout_reset_on(&V_dyn_timeout, hz, dyn_tick, vnetx, 0); 2801 CURVNET_RESTORE(); 2802 } 2803 2804 void 2805 ipfw_expire_dyn_states(struct ip_fw_chain *chain, ipfw_range_tlv *rt) 2806 { 2807 /* 2808 * Do not perform any checks if we currently have no dynamic states 2809 */ 2810 if (V_dyn_count == 0) 2811 return; 2812 2813 IPFW_UH_WLOCK_ASSERT(chain); 2814 dyn_expire_states(chain, rt); 2815 } 2816 2817 /* 2818 * Pass through all states and reset eaction for orphaned rules. 2819 */ 2820 void 2821 ipfw_dyn_reset_eaction(struct ip_fw_chain *ch, uint16_t eaction_id, 2822 uint16_t default_id, uint16_t instance_id) 2823 { 2824 #ifdef INET6 2825 struct dyn_ipv6_state *s6; 2826 #endif 2827 struct dyn_ipv4_state *s4; 2828 struct ip_fw *rule; 2829 uint32_t bucket; 2830 2831 #define DYN_RESET_EACTION(s, h, b) \ 2832 CK_SLIST_FOREACH(s, &V_dyn_ ## h[b], entry) { \ 2833 if ((s->data->flags & DYN_REFERENCED) == 0) \ 2834 continue; \ 2835 rule = s->data->parent; \ 2836 if (s->type == O_LIMIT) \ 2837 rule = ((__typeof(s))rule)->limit->parent; \ 2838 ipfw_reset_eaction(ch, rule, eaction_id, \ 2839 default_id, instance_id); \ 2840 } 2841 2842 IPFW_UH_WLOCK_ASSERT(ch); 2843 if (V_dyn_count == 0) 2844 return; 2845 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 2846 DYN_RESET_EACTION(s4, ipv4, bucket); 2847 #ifdef INET6 2848 DYN_RESET_EACTION(s6, ipv6, bucket); 2849 #endif 2850 } 2851 } 2852 2853 /* 2854 * Returns size of dynamic states in legacy format 2855 */ 2856 int 2857 ipfw_dyn_len(void) 2858 { 2859 2860 return ((V_dyn_count + V_dyn_parent_count) * sizeof(ipfw_dyn_rule)); 2861 } 2862 2863 /* 2864 * Returns number of dynamic states. 2865 * Marks every named object index used by dynamic states with bit in @bmask. 2866 * Returns number of named objects accounted in bmask via @nocnt. 2867 * Used by dump format v1 (current). 2868 */ 2869 uint32_t 2870 ipfw_dyn_get_count(uint32_t *bmask, int *nocnt) 2871 { 2872 #ifdef INET6 2873 struct dyn_ipv6_state *s6; 2874 #endif 2875 struct dyn_ipv4_state *s4; 2876 uint32_t bucket; 2877 2878 #define DYN_COUNT_OBJECTS(s, h, b) \ 2879 CK_SLIST_FOREACH(s, &V_dyn_ ## h[b], entry) { \ 2880 MPASS(s->kidx != 0); \ 2881 if (ipfw_mark_object_kidx(bmask, IPFW_TLV_STATE_NAME, \ 2882 s->kidx) != 0) \ 2883 (*nocnt)++; \ 2884 } 2885 2886 IPFW_UH_RLOCK_ASSERT(&V_layer3_chain); 2887 2888 /* No need to pass through all the buckets. */ 2889 *nocnt = 0; 2890 if (V_dyn_count + V_dyn_parent_count == 0) 2891 return (0); 2892 2893 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 2894 DYN_COUNT_OBJECTS(s4, ipv4, bucket); 2895 #ifdef INET6 2896 DYN_COUNT_OBJECTS(s6, ipv6, bucket); 2897 #endif 2898 } 2899 2900 return (V_dyn_count + V_dyn_parent_count); 2901 } 2902 2903 /* 2904 * Check if rule contains at least one dynamic opcode. 2905 * 2906 * Returns 1 if such opcode is found, 0 otherwise. 2907 */ 2908 int 2909 ipfw_is_dyn_rule(struct ip_fw *rule) 2910 { 2911 int cmdlen, l; 2912 ipfw_insn *cmd; 2913 2914 l = rule->cmd_len; 2915 cmd = rule->cmd; 2916 cmdlen = 0; 2917 for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { 2918 cmdlen = F_LEN(cmd); 2919 2920 switch (cmd->opcode) { 2921 case O_LIMIT: 2922 case O_KEEP_STATE: 2923 case O_PROBE_STATE: 2924 case O_CHECK_STATE: 2925 return (1); 2926 } 2927 } 2928 2929 return (0); 2930 } 2931 2932 static void 2933 dyn_export_parent(const struct dyn_parent *p, uint16_t kidx, uint8_t set, 2934 ipfw_dyn_rule *dst) 2935 { 2936 2937 dst->dyn_type = O_LIMIT_PARENT; 2938 dst->kidx = kidx; 2939 dst->count = (uint16_t)DPARENT_COUNT(p); 2940 dst->expire = TIME_LEQ(p->expire, time_uptime) ? 0: 2941 p->expire - time_uptime; 2942 2943 /* 'rule' is used to pass up the rule number and set */ 2944 memcpy(&dst->rule, &p->rulenum, sizeof(p->rulenum)); 2945 2946 /* store set number into high word of dst->rule pointer. */ 2947 memcpy((char *)&dst->rule + sizeof(p->rulenum), &set, sizeof(set)); 2948 2949 /* unused fields */ 2950 dst->pcnt = 0; 2951 dst->bcnt = 0; 2952 dst->parent = NULL; 2953 dst->state = 0; 2954 dst->ack_fwd = 0; 2955 dst->ack_rev = 0; 2956 dst->bucket = p->hashval; 2957 /* 2958 * The legacy userland code will interpret a NULL here as a marker 2959 * for the last dynamic rule. 2960 */ 2961 dst->next = (ipfw_dyn_rule *)1; 2962 } 2963 2964 static void 2965 dyn_export_data(const struct dyn_data *data, uint16_t kidx, uint8_t type, 2966 uint8_t set, ipfw_dyn_rule *dst) 2967 { 2968 2969 dst->dyn_type = type; 2970 dst->kidx = kidx; 2971 dst->pcnt = data->pcnt_fwd + data->pcnt_rev; 2972 dst->bcnt = data->bcnt_fwd + data->bcnt_rev; 2973 dst->expire = TIME_LEQ(data->expire, time_uptime) ? 0: 2974 data->expire - time_uptime; 2975 2976 /* 'rule' is used to pass up the rule number and set */ 2977 memcpy(&dst->rule, &data->rulenum, sizeof(data->rulenum)); 2978 2979 /* store set number into high word of dst->rule pointer. */ 2980 memcpy((char *)&dst->rule + sizeof(data->rulenum), &set, sizeof(set)); 2981 2982 dst->state = data->state; 2983 if (data->flags & DYN_REFERENCED) 2984 dst->state |= IPFW_DYN_ORPHANED; 2985 2986 /* unused fields */ 2987 dst->parent = NULL; 2988 dst->ack_fwd = data->ack_fwd; 2989 dst->ack_rev = data->ack_rev; 2990 dst->count = 0; 2991 dst->bucket = data->hashval; 2992 /* 2993 * The legacy userland code will interpret a NULL here as a marker 2994 * for the last dynamic rule. 2995 */ 2996 dst->next = (ipfw_dyn_rule *)1; 2997 } 2998 2999 static void 3000 dyn_export_ipv4_state(const struct dyn_ipv4_state *s, ipfw_dyn_rule *dst) 3001 { 3002 struct ip_fw *rule; 3003 3004 switch (s->type) { 3005 case O_LIMIT_PARENT: 3006 rule = s->limit->parent; 3007 dyn_export_parent(s->limit, s->kidx, rule->set, dst); 3008 break; 3009 default: 3010 rule = s->data->parent; 3011 if (s->type == O_LIMIT) 3012 rule = ((struct dyn_ipv4_state *)rule)->limit->parent; 3013 dyn_export_data(s->data, s->kidx, s->type, rule->set, dst); 3014 } 3015 3016 dst->id.dst_ip = s->dst; 3017 dst->id.src_ip = s->src; 3018 dst->id.dst_port = s->dport; 3019 dst->id.src_port = s->sport; 3020 dst->id.fib = s->data->fibnum; 3021 dst->id.proto = s->proto; 3022 dst->id._flags = 0; 3023 dst->id.addr_type = 4; 3024 3025 memset(&dst->id.dst_ip6, 0, sizeof(dst->id.dst_ip6)); 3026 memset(&dst->id.src_ip6, 0, sizeof(dst->id.src_ip6)); 3027 dst->id.flow_id6 = dst->id.extra = 0; 3028 } 3029 3030 #ifdef INET6 3031 static void 3032 dyn_export_ipv6_state(const struct dyn_ipv6_state *s, ipfw_dyn_rule *dst) 3033 { 3034 struct ip_fw *rule; 3035 3036 switch (s->type) { 3037 case O_LIMIT_PARENT: 3038 rule = s->limit->parent; 3039 dyn_export_parent(s->limit, s->kidx, rule->set, dst); 3040 break; 3041 default: 3042 rule = s->data->parent; 3043 if (s->type == O_LIMIT) 3044 rule = ((struct dyn_ipv6_state *)rule)->limit->parent; 3045 dyn_export_data(s->data, s->kidx, s->type, rule->set, dst); 3046 } 3047 3048 dst->id.src_ip6 = s->src; 3049 dst->id.dst_ip6 = s->dst; 3050 dst->id.dst_port = s->dport; 3051 dst->id.src_port = s->sport; 3052 dst->id.fib = s->data->fibnum; 3053 dst->id.proto = s->proto; 3054 dst->id._flags = 0; 3055 dst->id.addr_type = 6; 3056 3057 dst->id.dst_ip = dst->id.src_ip = 0; 3058 dst->id.flow_id6 = dst->id.extra = 0; 3059 } 3060 #endif /* INET6 */ 3061 3062 /* 3063 * Fills the buffer given by @sd with dynamic states. 3064 * Used by dump format v1 (current). 3065 * 3066 * Returns 0 on success. 3067 */ 3068 int 3069 ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd) 3070 { 3071 #ifdef INET6 3072 struct dyn_ipv6_state *s6; 3073 #endif 3074 struct dyn_ipv4_state *s4; 3075 ipfw_obj_dyntlv *dst, *last; 3076 ipfw_obj_ctlv *ctlv; 3077 uint32_t bucket; 3078 3079 if (V_dyn_count == 0) 3080 return (0); 3081 3082 /* 3083 * IPFW_UH_RLOCK garantees that another userland request 3084 * and callout thread will not delete entries from states 3085 * lists. 3086 */ 3087 IPFW_UH_RLOCK_ASSERT(chain); 3088 3089 ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); 3090 if (ctlv == NULL) 3091 return (ENOMEM); 3092 ctlv->head.type = IPFW_TLV_DYNSTATE_LIST; 3093 ctlv->objsize = sizeof(ipfw_obj_dyntlv); 3094 last = NULL; 3095 3096 #define DYN_EXPORT_STATES(s, af, h, b) \ 3097 CK_SLIST_FOREACH(s, &V_dyn_ ## h[b], entry) { \ 3098 dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, \ 3099 sizeof(ipfw_obj_dyntlv)); \ 3100 if (dst == NULL) \ 3101 return (ENOMEM); \ 3102 dyn_export_ ## af ## _state(s, &dst->state); \ 3103 dst->head.length = sizeof(ipfw_obj_dyntlv); \ 3104 dst->head.type = IPFW_TLV_DYN_ENT; \ 3105 last = dst; \ 3106 } 3107 3108 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 3109 DYN_EXPORT_STATES(s4, ipv4, ipv4_parent, bucket); 3110 DYN_EXPORT_STATES(s4, ipv4, ipv4, bucket); 3111 #ifdef INET6 3112 DYN_EXPORT_STATES(s6, ipv6, ipv6_parent, bucket); 3113 DYN_EXPORT_STATES(s6, ipv6, ipv6, bucket); 3114 #endif /* INET6 */ 3115 } 3116 3117 /* mark last dynamic rule */ 3118 if (last != NULL) 3119 last->head.flags = IPFW_DF_LAST; /* XXX: unused */ 3120 return (0); 3121 #undef DYN_EXPORT_STATES 3122 } 3123 3124 /* 3125 * Fill given buffer with dynamic states (legacy format). 3126 * IPFW_UH_RLOCK has to be held while calling. 3127 */ 3128 void 3129 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep) 3130 { 3131 #ifdef INET6 3132 struct dyn_ipv6_state *s6; 3133 #endif 3134 struct dyn_ipv4_state *s4; 3135 ipfw_dyn_rule *p, *last = NULL; 3136 char *bp; 3137 uint32_t bucket; 3138 3139 if (V_dyn_count == 0) 3140 return; 3141 bp = *pbp; 3142 3143 IPFW_UH_RLOCK_ASSERT(chain); 3144 3145 #define DYN_EXPORT_STATES(s, af, head, b) \ 3146 CK_SLIST_FOREACH(s, &V_dyn_ ## head[b], entry) { \ 3147 if (bp + sizeof(*p) > ep) \ 3148 break; \ 3149 p = (ipfw_dyn_rule *)bp; \ 3150 dyn_export_ ## af ## _state(s, p); \ 3151 last = p; \ 3152 bp += sizeof(*p); \ 3153 } 3154 3155 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 3156 DYN_EXPORT_STATES(s4, ipv4, ipv4_parent, bucket); 3157 DYN_EXPORT_STATES(s4, ipv4, ipv4, bucket); 3158 #ifdef INET6 3159 DYN_EXPORT_STATES(s6, ipv6, ipv6_parent, bucket); 3160 DYN_EXPORT_STATES(s6, ipv6, ipv6, bucket); 3161 #endif /* INET6 */ 3162 } 3163 3164 if (last != NULL) /* mark last dynamic rule */ 3165 last->next = NULL; 3166 *pbp = bp; 3167 #undef DYN_EXPORT_STATES 3168 } 3169 3170 void 3171 ipfw_dyn_init(struct ip_fw_chain *chain) 3172 { 3173 3174 #ifdef IPFIREWALL_JENKINSHASH 3175 V_dyn_hashseed = arc4random(); 3176 #endif 3177 V_dyn_max = 16384; /* max # of states */ 3178 V_dyn_parent_max = 4096; /* max # of parent states */ 3179 V_dyn_buckets_max = 8192; /* must be power of 2 */ 3180 3181 V_dyn_ack_lifetime = 300; 3182 V_dyn_syn_lifetime = 20; 3183 V_dyn_fin_lifetime = 1; 3184 V_dyn_rst_lifetime = 1; 3185 V_dyn_udp_lifetime = 10; 3186 V_dyn_short_lifetime = 5; 3187 3188 V_dyn_keepalive_interval = 20; 3189 V_dyn_keepalive_period = 5; 3190 V_dyn_keepalive = 1; /* send keepalives */ 3191 V_dyn_keepalive_last = time_uptime; 3192 3193 V_dyn_data_zone = uma_zcreate("IPFW dynamic states data", 3194 sizeof(struct dyn_data), NULL, NULL, NULL, NULL, 3195 UMA_ALIGN_PTR, 0); 3196 uma_zone_set_max(V_dyn_data_zone, V_dyn_max); 3197 3198 V_dyn_parent_zone = uma_zcreate("IPFW parent dynamic states", 3199 sizeof(struct dyn_parent), NULL, NULL, NULL, NULL, 3200 UMA_ALIGN_PTR, 0); 3201 uma_zone_set_max(V_dyn_parent_zone, V_dyn_parent_max); 3202 3203 SLIST_INIT(&V_dyn_expired_ipv4); 3204 V_dyn_ipv4 = NULL; 3205 V_dyn_ipv4_parent = NULL; 3206 V_dyn_ipv4_zone = uma_zcreate("IPFW IPv4 dynamic states", 3207 sizeof(struct dyn_ipv4_state), NULL, NULL, NULL, NULL, 3208 UMA_ALIGN_PTR, 0); 3209 3210 #ifdef INET6 3211 SLIST_INIT(&V_dyn_expired_ipv6); 3212 V_dyn_ipv6 = NULL; 3213 V_dyn_ipv6_parent = NULL; 3214 V_dyn_ipv6_zone = uma_zcreate("IPFW IPv6 dynamic states", 3215 sizeof(struct dyn_ipv6_state), NULL, NULL, NULL, NULL, 3216 UMA_ALIGN_PTR, 0); 3217 #endif 3218 3219 /* Initialize buckets. */ 3220 V_curr_dyn_buckets = 0; 3221 V_dyn_bucket_lock = NULL; 3222 dyn_grow_hashtable(chain, 256, M_WAITOK); 3223 3224 if (IS_DEFAULT_VNET(curvnet)) 3225 dyn_hp_cache = malloc(mp_ncpus * sizeof(void *), M_IPFW, 3226 M_WAITOK | M_ZERO); 3227 3228 DYN_EXPIRED_LOCK_INIT(); 3229 callout_init(&V_dyn_timeout, 1); 3230 callout_reset(&V_dyn_timeout, hz, dyn_tick, curvnet); 3231 IPFW_ADD_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); 3232 } 3233 3234 void 3235 ipfw_dyn_uninit(int pass) 3236 { 3237 #ifdef INET6 3238 struct dyn_ipv6_state *s6; 3239 #endif 3240 struct dyn_ipv4_state *s4; 3241 int bucket; 3242 3243 if (pass == 0) { 3244 callout_drain(&V_dyn_timeout); 3245 return; 3246 } 3247 IPFW_DEL_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); 3248 DYN_EXPIRED_LOCK_DESTROY(); 3249 3250 #define DYN_FREE_STATES_FORCED(CK, s, af, name, en) do { \ 3251 while ((s = CK ## SLIST_FIRST(&V_dyn_ ## name)) != NULL) { \ 3252 CK ## SLIST_REMOVE_HEAD(&V_dyn_ ## name, en); \ 3253 if (s->type == O_LIMIT_PARENT) \ 3254 uma_zfree(V_dyn_parent_zone, s->limit); \ 3255 else \ 3256 uma_zfree(V_dyn_data_zone, s->data); \ 3257 uma_zfree(V_dyn_ ## af ## _zone, s); \ 3258 } \ 3259 } while (0) 3260 for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { 3261 DYN_BUCKET_LOCK_DESTROY(V_dyn_bucket_lock, bucket); 3262 3263 DYN_FREE_STATES_FORCED(CK_, s4, ipv4, ipv4[bucket], entry); 3264 DYN_FREE_STATES_FORCED(CK_, s4, ipv4, ipv4_parent[bucket], 3265 entry); 3266 #ifdef INET6 3267 DYN_FREE_STATES_FORCED(CK_, s6, ipv6, ipv6[bucket], entry); 3268 DYN_FREE_STATES_FORCED(CK_, s6, ipv6, ipv6_parent[bucket], 3269 entry); 3270 #endif /* INET6 */ 3271 } 3272 DYN_FREE_STATES_FORCED(, s4, ipv4, expired_ipv4, expired); 3273 #ifdef INET6 3274 DYN_FREE_STATES_FORCED(, s6, ipv6, expired_ipv6, expired); 3275 #endif 3276 #undef DYN_FREE_STATES_FORCED 3277 3278 uma_zdestroy(V_dyn_ipv4_zone); 3279 uma_zdestroy(V_dyn_data_zone); 3280 uma_zdestroy(V_dyn_parent_zone); 3281 #ifdef INET6 3282 uma_zdestroy(V_dyn_ipv6_zone); 3283 free(V_dyn_ipv6, M_IPFW); 3284 free(V_dyn_ipv6_parent, M_IPFW); 3285 free(V_dyn_ipv6_add, M_IPFW); 3286 free(V_dyn_ipv6_parent_add, M_IPFW); 3287 free(V_dyn_ipv6_del, M_IPFW); 3288 free(V_dyn_ipv6_parent_del, M_IPFW); 3289 #endif 3290 free(V_dyn_bucket_lock, M_IPFW); 3291 free(V_dyn_ipv4, M_IPFW); 3292 free(V_dyn_ipv4_parent, M_IPFW); 3293 free(V_dyn_ipv4_add, M_IPFW); 3294 free(V_dyn_ipv4_parent_add, M_IPFW); 3295 free(V_dyn_ipv4_del, M_IPFW); 3296 free(V_dyn_ipv4_parent_del, M_IPFW); 3297 if (IS_DEFAULT_VNET(curvnet)) 3298 free(dyn_hp_cache, M_IPFW); 3299 } 3300