1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1990, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2010-2011 Juniper Networks, Inc. 7 * All rights reserved. 8 * 9 * Portions of this software were developed by Robert N. M. Watson under 10 * contract to Juniper Networks, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)in_pcb.h 8.1 (Berkeley) 6/10/93 37 * $FreeBSD$ 38 */ 39 40 #ifndef _NETINET_IN_PCB_H_ 41 #define _NETINET_IN_PCB_H_ 42 43 #include <sys/queue.h> 44 #include <sys/epoch.h> 45 #include <sys/_lock.h> 46 #include <sys/_mutex.h> 47 #include <sys/_rwlock.h> 48 #include <net/route.h> 49 50 #ifdef _KERNEL 51 #include <sys/lock.h> 52 #include <sys/rwlock.h> 53 #include <net/vnet.h> 54 #include <vm/uma.h> 55 #endif 56 #include <sys/ck.h> 57 58 #define in6pcb inpcb /* for KAME src sync over BSD*'s */ 59 #define in6p_sp inp_sp /* for KAME src sync over BSD*'s */ 60 61 /* 62 * struct inpcb is the common protocol control block structure used in most 63 * IP transport protocols. 64 * 65 * Pointers to local and foreign host table entries, local and foreign socket 66 * numbers, and pointers up (to a socket structure) and down (to a 67 * protocol-specific control block) are stored here. 68 */ 69 CK_LIST_HEAD(inpcbhead, inpcb); 70 CK_LIST_HEAD(inpcbporthead, inpcbport); 71 typedef uint64_t inp_gen_t; 72 73 /* 74 * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet. 75 * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing 76 * the following structure. 77 */ 78 struct in_addr_4in6 { 79 u_int32_t ia46_pad32[3]; 80 struct in_addr ia46_addr4; 81 }; 82 83 union in_dependaddr { 84 struct in_addr_4in6 id46_addr; 85 struct in6_addr id6_addr; 86 }; 87 88 /* 89 * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553. in_conninfo has 90 * some extra padding to accomplish this. 91 * NOTE 2: tcp_syncache.c uses first 5 32-bit words, which identify fport, 92 * lport, faddr to generate hash, so these fields shouldn't be moved. 93 */ 94 struct in_endpoints { 95 u_int16_t ie_fport; /* foreign port */ 96 u_int16_t ie_lport; /* local port */ 97 /* protocol dependent part, local and foreign addr */ 98 union in_dependaddr ie_dependfaddr; /* foreign host table entry */ 99 union in_dependaddr ie_dependladdr; /* local host table entry */ 100 #define ie_faddr ie_dependfaddr.id46_addr.ia46_addr4 101 #define ie_laddr ie_dependladdr.id46_addr.ia46_addr4 102 #define ie6_faddr ie_dependfaddr.id6_addr 103 #define ie6_laddr ie_dependladdr.id6_addr 104 u_int32_t ie6_zoneid; /* scope zone id */ 105 }; 106 107 /* 108 * XXX The defines for inc_* are hacks and should be changed to direct 109 * references. 110 */ 111 struct in_conninfo { 112 u_int8_t inc_flags; 113 u_int8_t inc_len; 114 u_int16_t inc_fibnum; /* XXX was pad, 16 bits is plenty */ 115 /* protocol dependent part */ 116 struct in_endpoints inc_ie; 117 }; 118 119 /* 120 * Flags for inc_flags. 121 */ 122 #define INC_ISIPV6 0x01 123 124 #define inc_isipv6 inc_flags /* temp compatibility */ 125 #define inc_fport inc_ie.ie_fport 126 #define inc_lport inc_ie.ie_lport 127 #define inc_faddr inc_ie.ie_faddr 128 #define inc_laddr inc_ie.ie_laddr 129 #define inc6_faddr inc_ie.ie6_faddr 130 #define inc6_laddr inc_ie.ie6_laddr 131 #define inc6_zoneid inc_ie.ie6_zoneid 132 133 #if defined(_KERNEL) || defined(_WANT_INPCB) 134 /* 135 * struct inpcb captures the network layer state for TCP, UDP, and raw IPv4 and 136 * IPv6 sockets. In the case of TCP and UDP, further per-connection state is 137 * hung off of inp_ppcb most of the time. Almost all fields of struct inpcb 138 * are static after creation or protected by a per-inpcb rwlock, inp_lock. A 139 * few fields are protected by multiple locks as indicated in the locking notes 140 * below. For these fields, all of the listed locks must be write-locked for 141 * any modifications. However, these fields can be safely read while any one of 142 * the listed locks are read-locked. This model can permit greater concurrency 143 * for read operations. For example, connections can be looked up while only 144 * holding a read lock on the global pcblist lock. This is important for 145 * performance when attempting to find the connection for a packet given its IP 146 * and port tuple. 147 * 148 * One noteworthy exception is that the global pcbinfo lock follows a different 149 * set of rules in relation to the inp_list field. Rather than being 150 * write-locked for modifications and read-locked for list iterations, it must 151 * be read-locked during modifications and write-locked during list iterations. 152 * This ensures that the relatively rare global list iterations safely walk a 153 * stable snapshot of connections while allowing more common list modifications 154 * to safely grab the pcblist lock just while adding or removing a connection 155 * from the global list. 156 * 157 * Key: 158 * (b) - Protected by the hpts lock. 159 * (c) - Constant after initialization 160 * (g) - Protected by the pcbgroup lock 161 * (i) - Protected by the inpcb lock 162 * (p) - Protected by the pcbinfo lock for the inpcb 163 * (l) - Protected by the pcblist lock for the inpcb 164 * (h) - Protected by the pcbhash lock for the inpcb 165 * (s) - Protected by another subsystem's locks 166 * (x) - Undefined locking 167 * 168 * Notes on the tcp_hpts: 169 * 170 * First Hpts lock order is 171 * 1) INP_WLOCK() 172 * 2) HPTS_LOCK() i.e. hpts->pmtx 173 * 174 * To insert a TCB on the hpts you *must* be holding the INP_WLOCK(). 175 * You may check the inp->inp_in_hpts flag without the hpts lock. 176 * The hpts is the only one that will clear this flag holding 177 * only the hpts lock. This means that in your tcp_output() 178 * routine when you test for the inp_in_hpts flag to be 1 179 * it may be transitioning to 0 (by the hpts). 180 * That's ok since that will just mean an extra call to tcp_output 181 * that most likely will find the call you executed 182 * (when the mis-match occured) will have put the TCB back 183 * on the hpts and it will return. If your 184 * call did not add the inp back to the hpts then you will either 185 * over-send or the cwnd will block you from sending more. 186 * 187 * Note you should also be holding the INP_WLOCK() when you 188 * call the remove from the hpts as well. Though usually 189 * you are either doing this from a timer, where you need and have 190 * the INP_WLOCK() or from destroying your TCB where again 191 * you should already have the INP_WLOCK(). 192 * 193 * The inp_hpts_cpu, inp_hpts_cpu_set, inp_input_cpu and 194 * inp_input_cpu_set fields are controlled completely by 195 * the hpts. Do not ever set these. The inp_hpts_cpu_set 196 * and inp_input_cpu_set fields indicate if the hpts has 197 * setup the respective cpu field. It is advised if this 198 * field is 0, to enqueue the packet with the appropriate 199 * hpts_immediate() call. If the _set field is 1, then 200 * you may compare the inp_*_cpu field to the curcpu and 201 * may want to again insert onto the hpts if these fields 202 * are not equal (i.e. you are not on the expected CPU). 203 * 204 * A note on inp_hpts_calls and inp_input_calls, these 205 * flags are set when the hpts calls either the output 206 * or do_segment routines respectively. If the routine 207 * being called wants to use this, then it needs to 208 * clear the flag before returning. The hpts will not 209 * clear the flag. The flags can be used to tell if 210 * the hpts is the function calling the respective 211 * routine. 212 * 213 * A few other notes: 214 * 215 * When a read lock is held, stability of the field is guaranteed; to write 216 * to a field, a write lock must generally be held. 217 * 218 * netinet/netinet6-layer code should not assume that the inp_socket pointer 219 * is safe to dereference without inp_lock being held, even for protocols 220 * other than TCP (where the inpcb persists during TIMEWAIT even after the 221 * socket has been freed), or there may be close(2)-related races. 222 * 223 * The inp_vflag field is overloaded, and would otherwise ideally be (c). 224 * 225 * TODO: Currently only the TCP stack is leveraging the global pcbinfo lock 226 * read-lock usage during modification, this model can be applied to other 227 * protocols (especially SCTP). 228 */ 229 struct icmp6_filter; 230 struct inpcbpolicy; 231 struct m_snd_tag; 232 struct inpcb { 233 /* Cache line #1 (amd64) */ 234 CK_LIST_ENTRY(inpcb) inp_hash; /* (h/i) hash list */ 235 CK_LIST_ENTRY(inpcb) inp_pcbgrouphash; /* (g/i) hash list */ 236 struct rwlock inp_lock; 237 /* Cache line #2 (amd64) */ 238 #define inp_start_zero inp_hpts 239 #define inp_zero_size (sizeof(struct inpcb) - \ 240 offsetof(struct inpcb, inp_start_zero)) 241 TAILQ_ENTRY(inpcb) inp_hpts; /* pacing out queue next lock(b) */ 242 243 uint32_t inp_hpts_request; /* Current hpts request, zero if 244 * fits in the pacing window (i&b). */ 245 /* 246 * Note the next fields are protected by a 247 * different lock (hpts-lock). This means that 248 * they must correspond in size to the smallest 249 * protectable bit field (uint8_t on x86, and 250 * other platfomrs potentially uint32_t?). Also 251 * since CPU switches can occur at different times the two 252 * fields can *not* be collapsed into a signal bit field. 253 */ 254 #if defined(__amd64__) || defined(__i386__) 255 volatile uint8_t inp_in_hpts; /* on output hpts (lock b) */ 256 volatile uint8_t inp_in_input; /* on input hpts (lock b) */ 257 #else 258 volatile uint32_t inp_in_hpts; /* on output hpts (lock b) */ 259 volatile uint32_t inp_in_input; /* on input hpts (lock b) */ 260 #endif 261 volatile uint16_t inp_hpts_cpu; /* Lock (i) */ 262 u_int inp_refcount; /* (i) refcount */ 263 int inp_flags; /* (i) generic IP/datagram flags */ 264 int inp_flags2; /* (i) generic IP/datagram flags #2*/ 265 volatile uint16_t inp_input_cpu; /* Lock (i) */ 266 volatile uint8_t inp_hpts_cpu_set :1, /* on output hpts (i) */ 267 inp_input_cpu_set : 1, /* on input hpts (i) */ 268 inp_hpts_calls :1, /* (i) from output hpts */ 269 inp_input_calls :1, /* (i) from input hpts */ 270 inp_spare_bits2 : 4; 271 uint8_t inp_spare_byte; /* Compiler hole */ 272 void *inp_ppcb; /* (i) pointer to per-protocol pcb */ 273 struct socket *inp_socket; /* (i) back pointer to socket */ 274 uint32_t inp_hptsslot; /* Hpts wheel slot this tcb is Lock(i&b) */ 275 uint32_t inp_hpts_drop_reas; /* reason we are dropping the PCB (lock i&b) */ 276 TAILQ_ENTRY(inpcb) inp_input; /* pacing in queue next lock(b) */ 277 struct inpcbinfo *inp_pcbinfo; /* (c) PCB list info */ 278 struct inpcbgroup *inp_pcbgroup; /* (g/i) PCB group list */ 279 CK_LIST_ENTRY(inpcb) inp_pcbgroup_wild; /* (g/i/h) group wildcard entry */ 280 struct ucred *inp_cred; /* (c) cache of socket cred */ 281 u_int32_t inp_flow; /* (i) IPv6 flow information */ 282 u_char inp_vflag; /* (i) IP version flag (v4/v6) */ 283 u_char inp_ip_ttl; /* (i) time to live proto */ 284 u_char inp_ip_p; /* (c) protocol proto */ 285 u_char inp_ip_minttl; /* (i) minimum TTL or drop */ 286 uint32_t inp_flowid; /* (x) flow id / queue id */ 287 struct m_snd_tag *inp_snd_tag; /* (i) send tag for outgoing mbufs */ 288 uint32_t inp_flowtype; /* (x) M_HASHTYPE value */ 289 uint32_t inp_rss_listen_bucket; /* (x) overridden RSS listen bucket */ 290 291 /* Local and foreign ports, local and foreign addr. */ 292 struct in_conninfo inp_inc; /* (i) list for PCB's local port */ 293 294 /* MAC and IPSEC policy information. */ 295 struct label *inp_label; /* (i) MAC label */ 296 struct inpcbpolicy *inp_sp; /* (s) for IPSEC */ 297 298 /* Protocol-dependent part; options. */ 299 struct { 300 u_char inp_ip_tos; /* (i) type of service proto */ 301 struct mbuf *inp_options; /* (i) IP options */ 302 struct ip_moptions *inp_moptions; /* (i) mcast options */ 303 }; 304 struct { 305 /* (i) IP options */ 306 struct mbuf *in6p_options; 307 /* (i) IP6 options for outgoing packets */ 308 struct ip6_pktopts *in6p_outputopts; 309 /* (i) IP multicast options */ 310 struct ip6_moptions *in6p_moptions; 311 /* (i) ICMPv6 code type filter */ 312 struct icmp6_filter *in6p_icmp6filt; 313 /* (i) IPV6_CHECKSUM setsockopt */ 314 int in6p_cksum; 315 short in6p_hops; 316 }; 317 CK_LIST_ENTRY(inpcb) inp_portlist; /* (i/h) */ 318 struct inpcbport *inp_phd; /* (i/h) head of this list */ 319 inp_gen_t inp_gencnt; /* (c) generation count */ 320 struct llentry *inp_lle; /* cached L2 information */ 321 rt_gen_t inp_rt_cookie; /* generation for route entry */ 322 union { /* cached L3 information */ 323 struct route inp_route; 324 struct route_in6 inp_route6; 325 }; 326 CK_LIST_ENTRY(inpcb) inp_list; /* (p/l) list for all PCBs for proto */ 327 /* (p[w]) for list iteration */ 328 /* (p[r]/l) for addition/removal */ 329 struct epoch_context inp_epoch_ctx; 330 }; 331 #endif /* _KERNEL */ 332 333 #define inp_fport inp_inc.inc_fport 334 #define inp_lport inp_inc.inc_lport 335 #define inp_faddr inp_inc.inc_faddr 336 #define inp_laddr inp_inc.inc_laddr 337 338 #define in6p_faddr inp_inc.inc6_faddr 339 #define in6p_laddr inp_inc.inc6_laddr 340 #define in6p_zoneid inp_inc.inc6_zoneid 341 #define in6p_flowinfo inp_flow 342 343 #define inp_vnet inp_pcbinfo->ipi_vnet 344 345 /* 346 * The range of the generation count, as used in this implementation, is 9e19. 347 * We would have to create 300 billion connections per second for this number 348 * to roll over in a year. This seems sufficiently unlikely that we simply 349 * don't concern ourselves with that possibility. 350 */ 351 352 /* 353 * Interface exported to userland by various protocols which use inpcbs. Hack 354 * alert -- only define if struct xsocket is in scope. 355 * Fields prefixed with "xi_" are unique to this structure, and the rest 356 * match fields in the struct inpcb, to ease coding and porting. 357 * 358 * Legend: 359 * (s) - used by userland utilities in src 360 * (p) - used by utilities in ports 361 * (3) - is known to be used by third party software not in ports 362 * (n) - no known usage 363 */ 364 #ifdef _SYS_SOCKETVAR_H_ 365 struct xinpcb { 366 size_t xi_len; /* length of this structure */ 367 struct xsocket xi_socket; /* (s,p) */ 368 struct in_conninfo inp_inc; /* (s,p) */ 369 uint64_t inp_gencnt; /* (s,p) */ 370 union { 371 void *inp_ppcb; /* (s) netstat(1) */ 372 int64_t ph_ppcb; 373 }; 374 int64_t inp_spare64[4]; 375 uint32_t inp_flow; /* (s) */ 376 uint32_t inp_flowid; /* (s) */ 377 uint32_t inp_flowtype; /* (s) */ 378 int32_t inp_flags; /* (s,p) */ 379 int32_t inp_flags2; /* (s) */ 380 int32_t inp_rss_listen_bucket; /* (n) */ 381 int32_t in6p_cksum; /* (n) */ 382 int32_t inp_spare32[4]; 383 uint16_t in6p_hops; /* (n) */ 384 uint8_t inp_ip_tos; /* (n) */ 385 int8_t pad8; 386 uint8_t inp_vflag; /* (s,p) */ 387 uint8_t inp_ip_ttl; /* (n) */ 388 uint8_t inp_ip_p; /* (n) */ 389 uint8_t inp_ip_minttl; /* (n) */ 390 int8_t inp_spare8[4]; 391 } __aligned(8); 392 393 struct xinpgen { 394 size_t xig_len; /* length of this structure */ 395 u_int xig_count; /* number of PCBs at this time */ 396 inp_gen_t xig_gen; /* generation count at this time */ 397 so_gen_t xig_sogen; /* socket generation count this time */ 398 } __aligned(8); 399 #ifdef _KERNEL 400 void in_pcbtoxinpcb(const struct inpcb *, struct xinpcb *); 401 #endif 402 #endif /* _SYS_SOCKETVAR_H_ */ 403 404 struct inpcbport { 405 struct epoch_context phd_epoch_ctx; 406 CK_LIST_ENTRY(inpcbport) phd_hash; 407 struct inpcbhead phd_pcblist; 408 u_short phd_port; 409 }; 410 411 struct in_pcblist { 412 int il_count; 413 struct epoch_context il_epoch_ctx; 414 struct inpcbinfo *il_pcbinfo; 415 struct inpcb *il_inp_list[0]; 416 }; 417 418 /*- 419 * Global data structure for each high-level protocol (UDP, TCP, ...) in both 420 * IPv4 and IPv6. Holds inpcb lists and information for managing them. 421 * 422 * Each pcbinfo is protected by three locks: ipi_lock, ipi_hash_lock and 423 * ipi_list_lock: 424 * - ipi_lock covering the global pcb list stability during loop iteration, 425 * - ipi_hash_lock covering the hashed lookup tables, 426 * - ipi_list_lock covering mutable global fields (such as the global 427 * pcb list) 428 * 429 * The lock order is: 430 * 431 * ipi_lock (before) 432 * inpcb locks (before) 433 * ipi_list locks (before) 434 * {ipi_hash_lock, pcbgroup locks} 435 * 436 * Locking key: 437 * 438 * (c) Constant or nearly constant after initialisation 439 * (g) Locked by ipi_lock 440 * (l) Locked by ipi_list_lock 441 * (h) Read using either ipi_hash_lock or inpcb lock; write requires both 442 * (p) Protected by one or more pcbgroup locks 443 * (x) Synchronisation properties poorly defined 444 */ 445 struct inpcbinfo { 446 /* 447 * Global lock protecting full inpcb list traversal 448 */ 449 struct rwlock ipi_lock; 450 451 /* 452 * Global list of inpcbs on the protocol. 453 */ 454 struct inpcbhead *ipi_listhead; /* (g/l) */ 455 u_int ipi_count; /* (l) */ 456 457 /* 458 * Generation count -- incremented each time a connection is allocated 459 * or freed. 460 */ 461 u_quad_t ipi_gencnt; /* (l) */ 462 463 /* 464 * Fields associated with port lookup and allocation. 465 */ 466 u_short ipi_lastport; /* (x) */ 467 u_short ipi_lastlow; /* (x) */ 468 u_short ipi_lasthi; /* (x) */ 469 470 /* 471 * UMA zone from which inpcbs are allocated for this protocol. 472 */ 473 struct uma_zone *ipi_zone; /* (c) */ 474 475 /* 476 * Connection groups associated with this protocol. These fields are 477 * constant, but pcbgroup structures themselves are protected by 478 * per-pcbgroup locks. 479 */ 480 struct inpcbgroup *ipi_pcbgroups; /* (c) */ 481 u_int ipi_npcbgroups; /* (c) */ 482 u_int ipi_hashfields; /* (c) */ 483 484 /* 485 * Global lock protecting non-pcbgroup hash lookup tables. 486 */ 487 struct rwlock ipi_hash_lock; 488 489 /* 490 * Global hash of inpcbs, hashed by local and foreign addresses and 491 * port numbers. 492 */ 493 struct inpcbhead *ipi_hashbase; /* (h) */ 494 u_long ipi_hashmask; /* (h) */ 495 496 /* 497 * Global hash of inpcbs, hashed by only local port number. 498 */ 499 struct inpcbporthead *ipi_porthashbase; /* (h) */ 500 u_long ipi_porthashmask; /* (h) */ 501 502 /* 503 * List of wildcard inpcbs for use with pcbgroups. In the past, was 504 * per-pcbgroup but is now global. All pcbgroup locks must be held 505 * to modify the list, so any is sufficient to read it. 506 */ 507 struct inpcbhead *ipi_wildbase; /* (p) */ 508 u_long ipi_wildmask; /* (p) */ 509 510 /* 511 * Load balance groups used for the SO_REUSEPORT_LB option, 512 * hashed by local port. 513 */ 514 struct inpcblbgrouphead *ipi_lbgrouphashbase; /* (h) */ 515 u_long ipi_lbgrouphashmask; /* (h) */ 516 517 /* 518 * Pointer to network stack instance 519 */ 520 struct vnet *ipi_vnet; /* (c) */ 521 522 /* 523 * general use 2 524 */ 525 void *ipi_pspare[2]; 526 527 /* 528 * Global lock protecting global inpcb list, inpcb count, etc. 529 */ 530 struct rwlock ipi_list_lock; 531 }; 532 533 #ifdef _KERNEL 534 /* 535 * Connection groups hold sets of connections that have similar CPU/thread 536 * affinity. Each connection belongs to exactly one connection group. 537 */ 538 struct inpcbgroup { 539 /* 540 * Per-connection group hash of inpcbs, hashed by local and foreign 541 * addresses and port numbers. 542 */ 543 struct inpcbhead *ipg_hashbase; /* (c) */ 544 u_long ipg_hashmask; /* (c) */ 545 546 /* 547 * Notional affinity of this pcbgroup. 548 */ 549 u_int ipg_cpu; /* (p) */ 550 551 /* 552 * Per-connection group lock, not to be confused with ipi_lock. 553 * Protects the hash table hung off the group, but also the global 554 * wildcard list in inpcbinfo. 555 */ 556 struct mtx ipg_lock; 557 } __aligned(CACHE_LINE_SIZE); 558 559 /* 560 * Load balance groups used for the SO_REUSEPORT_LB socket option. Each group 561 * (or unique address:port combination) can be re-used at most 562 * INPCBLBGROUP_SIZMAX (256) times. The inpcbs are stored in il_inp which 563 * is dynamically resized as processes bind/unbind to that specific group. 564 */ 565 struct inpcblbgroup { 566 LIST_ENTRY(inpcblbgroup) il_list; 567 uint16_t il_lport; /* (c) */ 568 u_char il_vflag; /* (c) */ 569 u_char il_pad; 570 uint32_t il_pad2; 571 union in_dependaddr il_dependladdr; /* (c) */ 572 #define il_laddr il_dependladdr.id46_addr.ia46_addr4 573 #define il6_laddr il_dependladdr.id6_addr 574 uint32_t il_inpsiz; /* max count in il_inp[] (h) */ 575 uint32_t il_inpcnt; /* cur count in il_inp[] (h) */ 576 struct inpcb *il_inp[]; /* (h) */ 577 }; 578 LIST_HEAD(inpcblbgrouphead, inpcblbgroup); 579 580 #define INP_LOCK_INIT(inp, d, t) \ 581 rw_init_flags(&(inp)->inp_lock, (t), RW_RECURSE | RW_DUPOK) 582 #define INP_LOCK_DESTROY(inp) rw_destroy(&(inp)->inp_lock) 583 #define INP_RLOCK(inp) rw_rlock(&(inp)->inp_lock) 584 #define INP_WLOCK(inp) rw_wlock(&(inp)->inp_lock) 585 #define INP_TRY_RLOCK(inp) rw_try_rlock(&(inp)->inp_lock) 586 #define INP_TRY_WLOCK(inp) rw_try_wlock(&(inp)->inp_lock) 587 #define INP_RUNLOCK(inp) rw_runlock(&(inp)->inp_lock) 588 #define INP_WUNLOCK(inp) rw_wunlock(&(inp)->inp_lock) 589 #define INP_TRY_UPGRADE(inp) rw_try_upgrade(&(inp)->inp_lock) 590 #define INP_DOWNGRADE(inp) rw_downgrade(&(inp)->inp_lock) 591 #define INP_WLOCKED(inp) rw_wowned(&(inp)->inp_lock) 592 #define INP_LOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_LOCKED) 593 #define INP_RLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_RLOCKED) 594 #define INP_WLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_WLOCKED) 595 #define INP_UNLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_UNLOCKED) 596 597 /* 598 * These locking functions are for inpcb consumers outside of sys/netinet, 599 * more specifically, they were added for the benefit of TOE drivers. The 600 * macros are reserved for use by the stack. 601 */ 602 void inp_wlock(struct inpcb *); 603 void inp_wunlock(struct inpcb *); 604 void inp_rlock(struct inpcb *); 605 void inp_runlock(struct inpcb *); 606 607 #ifdef INVARIANT_SUPPORT 608 void inp_lock_assert(struct inpcb *); 609 void inp_unlock_assert(struct inpcb *); 610 #else 611 #define inp_lock_assert(inp) do {} while (0) 612 #define inp_unlock_assert(inp) do {} while (0) 613 #endif 614 615 void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg); 616 int inp_ip_tos_get(const struct inpcb *inp); 617 void inp_ip_tos_set(struct inpcb *inp, int val); 618 struct socket * 619 inp_inpcbtosocket(struct inpcb *inp); 620 struct tcpcb * 621 inp_inpcbtotcpcb(struct inpcb *inp); 622 void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 623 uint32_t *faddr, uint16_t *fp); 624 int inp_so_options(const struct inpcb *inp); 625 626 #endif /* _KERNEL */ 627 628 #define INP_INFO_LOCK_INIT(ipi, d) \ 629 rw_init_flags(&(ipi)->ipi_lock, (d), RW_RECURSE) 630 #define INP_INFO_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_lock) 631 #define INP_INFO_RLOCK(ipi) rw_rlock(&(ipi)->ipi_lock) 632 #define INP_INFO_WLOCK(ipi) rw_wlock(&(ipi)->ipi_lock) 633 #define INP_INFO_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_lock) 634 #define INP_INFO_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_lock) 635 #define INP_INFO_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_lock) 636 #define INP_INFO_WLOCKED(ipi) rw_wowned(&(ipi)->ipi_lock) 637 #define INP_INFO_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_lock) 638 #define INP_INFO_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_lock) 639 #define INP_INFO_LOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_LOCKED) 640 #define INP_INFO_RLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_RLOCKED) 641 #define INP_INFO_WLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_WLOCKED) 642 #define INP_INFO_UNLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_UNLOCKED) 643 644 #define INP_LIST_LOCK_INIT(ipi, d) \ 645 rw_init_flags(&(ipi)->ipi_list_lock, (d), 0) 646 #define INP_LIST_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_list_lock) 647 #define INP_LIST_RLOCK(ipi) rw_rlock(&(ipi)->ipi_list_lock) 648 #define INP_LIST_WLOCK(ipi) rw_wlock(&(ipi)->ipi_list_lock) 649 #define INP_LIST_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_list_lock) 650 #define INP_LIST_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_list_lock) 651 #define INP_LIST_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_list_lock) 652 #define INP_LIST_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_list_lock) 653 #define INP_LIST_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_list_lock) 654 #define INP_LIST_LOCK_ASSERT(ipi) \ 655 rw_assert(&(ipi)->ipi_list_lock, RA_LOCKED) 656 #define INP_LIST_RLOCK_ASSERT(ipi) \ 657 rw_assert(&(ipi)->ipi_list_lock, RA_RLOCKED) 658 #define INP_LIST_WLOCK_ASSERT(ipi) \ 659 rw_assert(&(ipi)->ipi_list_lock, RA_WLOCKED) 660 #define INP_LIST_UNLOCK_ASSERT(ipi) \ 661 rw_assert(&(ipi)->ipi_list_lock, RA_UNLOCKED) 662 663 #define INP_HASH_LOCK_INIT(ipi, d) \ 664 rw_init_flags(&(ipi)->ipi_hash_lock, (d), 0) 665 #define INP_HASH_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_hash_lock) 666 #define INP_HASH_RLOCK(ipi) rw_rlock(&(ipi)->ipi_hash_lock) 667 #define INP_HASH_WLOCK(ipi) rw_wlock(&(ipi)->ipi_hash_lock) 668 #define INP_HASH_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_hash_lock) 669 #define INP_HASH_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_hash_lock) 670 #define INP_HASH_LOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_hash_lock, \ 671 RA_LOCKED) 672 #define INP_HASH_WLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_hash_lock, \ 673 RA_WLOCKED) 674 675 #define INP_GROUP_LOCK_INIT(ipg, d) mtx_init(&(ipg)->ipg_lock, (d), NULL, \ 676 MTX_DEF | MTX_DUPOK) 677 #define INP_GROUP_LOCK_DESTROY(ipg) mtx_destroy(&(ipg)->ipg_lock) 678 679 #define INP_GROUP_LOCK(ipg) mtx_lock(&(ipg)->ipg_lock) 680 #define INP_GROUP_LOCK_ASSERT(ipg) mtx_assert(&(ipg)->ipg_lock, MA_OWNED) 681 #define INP_GROUP_UNLOCK(ipg) mtx_unlock(&(ipg)->ipg_lock) 682 683 #define INP_PCBHASH(faddr, lport, fport, mask) \ 684 (((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask)) 685 #define INP_PCBPORTHASH(lport, mask) \ 686 (ntohs((lport)) & (mask)) 687 #define INP_PCBLBGROUP_PORTHASH(lport, mask) \ 688 (ntohs((lport)) & (mask)) 689 #define INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) \ 690 ((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) 691 #define INP6_PCBHASHKEY(faddr) ((faddr)->s6_addr32[3]) 692 693 /* 694 * Flags for inp_vflags -- historically version flags only 695 */ 696 #define INP_IPV4 0x1 697 #define INP_IPV6 0x2 698 #define INP_IPV6PROTO 0x4 /* opened under IPv6 protocol */ 699 700 /* 701 * Flags for inp_flags. 702 */ 703 #define INP_RECVOPTS 0x00000001 /* receive incoming IP options */ 704 #define INP_RECVRETOPTS 0x00000002 /* receive IP options for reply */ 705 #define INP_RECVDSTADDR 0x00000004 /* receive IP dst address */ 706 #define INP_HDRINCL 0x00000008 /* user supplies entire IP header */ 707 #define INP_HIGHPORT 0x00000010 /* user wants "high" port binding */ 708 #define INP_LOWPORT 0x00000020 /* user wants "low" port binding */ 709 #define INP_ANONPORT 0x00000040 /* port chosen for user */ 710 #define INP_RECVIF 0x00000080 /* receive incoming interface */ 711 #define INP_MTUDISC 0x00000100 /* user can do MTU discovery */ 712 /* 0x000200 unused: was INP_FAITH */ 713 #define INP_RECVTTL 0x00000400 /* receive incoming IP TTL */ 714 #define INP_DONTFRAG 0x00000800 /* don't fragment packet */ 715 #define INP_BINDANY 0x00001000 /* allow bind to any address */ 716 #define INP_INHASHLIST 0x00002000 /* in_pcbinshash() has been called */ 717 #define INP_RECVTOS 0x00004000 /* receive incoming IP TOS */ 718 #define IN6P_IPV6_V6ONLY 0x00008000 /* restrict AF_INET6 socket for v6 */ 719 #define IN6P_PKTINFO 0x00010000 /* receive IP6 dst and I/F */ 720 #define IN6P_HOPLIMIT 0x00020000 /* receive hoplimit */ 721 #define IN6P_HOPOPTS 0x00040000 /* receive hop-by-hop options */ 722 #define IN6P_DSTOPTS 0x00080000 /* receive dst options after rthdr */ 723 #define IN6P_RTHDR 0x00100000 /* receive routing header */ 724 #define IN6P_RTHDRDSTOPTS 0x00200000 /* receive dstoptions before rthdr */ 725 #define IN6P_TCLASS 0x00400000 /* receive traffic class value */ 726 #define IN6P_AUTOFLOWLABEL 0x00800000 /* attach flowlabel automatically */ 727 #define INP_TIMEWAIT 0x01000000 /* in TIMEWAIT, ppcb is tcptw */ 728 #define INP_ONESBCAST 0x02000000 /* send all-ones broadcast */ 729 #define INP_DROPPED 0x04000000 /* protocol drop flag */ 730 #define INP_SOCKREF 0x08000000 /* strong socket reference */ 731 #define INP_RESERVED_0 0x10000000 /* reserved field */ 732 #define INP_RESERVED_1 0x20000000 /* reserved field */ 733 #define IN6P_RFC2292 0x40000000 /* used RFC2292 API on the socket */ 734 #define IN6P_MTU 0x80000000 /* receive path MTU */ 735 736 #define INP_CONTROLOPTS (INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\ 737 INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\ 738 IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\ 739 IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\ 740 IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\ 741 IN6P_MTU) 742 743 /* 744 * Flags for inp_flags2. 745 */ 746 #define INP_LLE_VALID 0x00000001 /* cached lle is valid */ 747 #define INP_RT_VALID 0x00000002 /* cached rtentry is valid */ 748 #define INP_PCBGROUPWILD 0x00000004 /* in pcbgroup wildcard list */ 749 #define INP_REUSEPORT 0x00000008 /* SO_REUSEPORT option is set */ 750 #define INP_FREED 0x00000010 /* inp itself is not valid */ 751 #define INP_REUSEADDR 0x00000020 /* SO_REUSEADDR option is set */ 752 #define INP_BINDMULTI 0x00000040 /* IP_BINDMULTI option is set */ 753 #define INP_RSS_BUCKET_SET 0x00000080 /* IP_RSS_LISTEN_BUCKET is set */ 754 #define INP_RECVFLOWID 0x00000100 /* populate recv datagram with flow info */ 755 #define INP_RECVRSSBUCKETID 0x00000200 /* populate recv datagram with bucket id */ 756 #define INP_RATE_LIMIT_CHANGED 0x00000400 /* rate limit needs attention */ 757 #define INP_ORIGDSTADDR 0x00000800 /* receive IP dst address/port */ 758 #define INP_CANNOT_DO_ECN 0x00001000 /* The stack does not do ECN */ 759 #define INP_REUSEPORT_LB 0x00002000 /* SO_REUSEPORT_LB option is set */ 760 761 /* 762 * Flags passed to in_pcblookup*() functions. 763 */ 764 #define INPLOOKUP_WILDCARD 0x00000001 /* Allow wildcard sockets. */ 765 #define INPLOOKUP_RLOCKPCB 0x00000002 /* Return inpcb read-locked. */ 766 #define INPLOOKUP_WLOCKPCB 0x00000004 /* Return inpcb write-locked. */ 767 768 #define INPLOOKUP_MASK (INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB | \ 769 INPLOOKUP_WLOCKPCB) 770 771 #define sotoinpcb(so) ((struct inpcb *)(so)->so_pcb) 772 #define sotoin6pcb(so) sotoinpcb(so) /* for KAME src sync over BSD*'s */ 773 774 #define INP_SOCKAF(so) so->so_proto->pr_domain->dom_family 775 776 #define INP_CHECK_SOCKAF(so, af) (INP_SOCKAF(so) == af) 777 778 /* 779 * Constants for pcbinfo.ipi_hashfields. 780 */ 781 #define IPI_HASHFIELDS_NONE 0 782 #define IPI_HASHFIELDS_2TUPLE 1 783 #define IPI_HASHFIELDS_4TUPLE 2 784 785 #ifdef _KERNEL 786 VNET_DECLARE(int, ipport_reservedhigh); 787 VNET_DECLARE(int, ipport_reservedlow); 788 VNET_DECLARE(int, ipport_lowfirstauto); 789 VNET_DECLARE(int, ipport_lowlastauto); 790 VNET_DECLARE(int, ipport_firstauto); 791 VNET_DECLARE(int, ipport_lastauto); 792 VNET_DECLARE(int, ipport_hifirstauto); 793 VNET_DECLARE(int, ipport_hilastauto); 794 VNET_DECLARE(int, ipport_randomized); 795 VNET_DECLARE(int, ipport_randomcps); 796 VNET_DECLARE(int, ipport_randomtime); 797 VNET_DECLARE(int, ipport_stoprandom); 798 VNET_DECLARE(int, ipport_tcpallocs); 799 800 #define V_ipport_reservedhigh VNET(ipport_reservedhigh) 801 #define V_ipport_reservedlow VNET(ipport_reservedlow) 802 #define V_ipport_lowfirstauto VNET(ipport_lowfirstauto) 803 #define V_ipport_lowlastauto VNET(ipport_lowlastauto) 804 #define V_ipport_firstauto VNET(ipport_firstauto) 805 #define V_ipport_lastauto VNET(ipport_lastauto) 806 #define V_ipport_hifirstauto VNET(ipport_hifirstauto) 807 #define V_ipport_hilastauto VNET(ipport_hilastauto) 808 #define V_ipport_randomized VNET(ipport_randomized) 809 #define V_ipport_randomcps VNET(ipport_randomcps) 810 #define V_ipport_randomtime VNET(ipport_randomtime) 811 #define V_ipport_stoprandom VNET(ipport_stoprandom) 812 #define V_ipport_tcpallocs VNET(ipport_tcpallocs) 813 814 void in_pcbinfo_destroy(struct inpcbinfo *); 815 void in_pcbinfo_init(struct inpcbinfo *, const char *, struct inpcbhead *, 816 int, int, char *, uma_init, u_int); 817 818 int in_pcbbind_check_bindmulti(const struct inpcb *ni, 819 const struct inpcb *oi); 820 821 struct inpcbgroup * 822 in_pcbgroup_byhash(struct inpcbinfo *, u_int, uint32_t); 823 struct inpcbgroup * 824 in_pcbgroup_byinpcb(struct inpcb *); 825 struct inpcbgroup * 826 in_pcbgroup_bytuple(struct inpcbinfo *, struct in_addr, u_short, 827 struct in_addr, u_short); 828 void in_pcbgroup_destroy(struct inpcbinfo *); 829 int in_pcbgroup_enabled(struct inpcbinfo *); 830 void in_pcbgroup_init(struct inpcbinfo *, u_int, int); 831 void in_pcbgroup_remove(struct inpcb *); 832 void in_pcbgroup_update(struct inpcb *); 833 void in_pcbgroup_update_mbuf(struct inpcb *, struct mbuf *); 834 835 void in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *); 836 int in_pcballoc(struct socket *, struct inpcbinfo *); 837 int in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *); 838 int in_pcb_lport(struct inpcb *, struct in_addr *, u_short *, 839 struct ucred *, int); 840 int in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *, 841 u_short *, struct ucred *); 842 int in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *); 843 int in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *, 844 struct mbuf *); 845 int in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *, 846 u_short *, in_addr_t *, u_short *, struct inpcb **, 847 struct ucred *); 848 void in_pcbdetach(struct inpcb *); 849 void in_pcbdisconnect(struct inpcb *); 850 void in_pcbdrop(struct inpcb *); 851 void in_pcbfree(struct inpcb *); 852 int in_pcbinshash(struct inpcb *); 853 int in_pcbinshash_nopcbgroup(struct inpcb *); 854 int in_pcbladdr(struct inpcb *, struct in_addr *, struct in_addr *, 855 struct ucred *); 856 struct inpcb * 857 in_pcblookup_local(struct inpcbinfo *, 858 struct in_addr, u_short, int, struct ucred *); 859 struct inpcb * 860 in_pcblookup(struct inpcbinfo *, struct in_addr, u_int, 861 struct in_addr, u_int, int, struct ifnet *); 862 struct inpcb * 863 in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int, 864 struct in_addr, u_int, int, struct ifnet *, struct mbuf *); 865 void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr, 866 int, struct inpcb *(*)(struct inpcb *, int)); 867 void in_pcbref(struct inpcb *); 868 void in_pcbrehash(struct inpcb *); 869 void in_pcbrehash_mbuf(struct inpcb *, struct mbuf *); 870 int in_pcbrele(struct inpcb *); 871 int in_pcbrele_rlocked(struct inpcb *); 872 int in_pcbrele_wlocked(struct inpcb *); 873 void in_pcblist_rele_rlocked(epoch_context_t ctx); 874 void in_losing(struct inpcb *); 875 void in_pcbsetsolabel(struct socket *so); 876 int in_getpeeraddr(struct socket *so, struct sockaddr **nam); 877 int in_getsockaddr(struct socket *so, struct sockaddr **nam); 878 struct sockaddr * 879 in_sockaddr(in_port_t port, struct in_addr *addr); 880 void in_pcbsosetlabel(struct socket *so); 881 #ifdef RATELIMIT 882 int in_pcbattach_txrtlmt(struct inpcb *, struct ifnet *, uint32_t, uint32_t, uint32_t); 883 void in_pcbdetach_txrtlmt(struct inpcb *); 884 int in_pcbmodify_txrtlmt(struct inpcb *, uint32_t); 885 int in_pcbquery_txrtlmt(struct inpcb *, uint32_t *); 886 int in_pcbquery_txrlevel(struct inpcb *, uint32_t *); 887 void in_pcboutput_txrtlmt(struct inpcb *, struct ifnet *, struct mbuf *); 888 void in_pcboutput_eagain(struct inpcb *); 889 #endif 890 #endif /* _KERNEL */ 891 892 #endif /* !_NETINET_IN_PCB_H_ */ 893