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 <net/if.h> 55 #include <net/if_var.h> 56 #include <vm/uma.h> 57 #endif 58 #include <sys/ck.h> 59 60 #define in6pcb inpcb /* for KAME src sync over BSD*'s */ 61 #define in6p_sp inp_sp /* for KAME src sync over BSD*'s */ 62 63 /* 64 * struct inpcb is the common protocol control block structure used in most 65 * IP transport protocols. 66 * 67 * Pointers to local and foreign host table entries, local and foreign socket 68 * numbers, and pointers up (to a socket structure) and down (to a 69 * protocol-specific control block) are stored here. 70 */ 71 CK_LIST_HEAD(inpcbhead, inpcb); 72 CK_LIST_HEAD(inpcbporthead, inpcbport); 73 CK_LIST_HEAD(inpcblbgrouphead, inpcblbgroup); 74 typedef uint64_t inp_gen_t; 75 76 /* 77 * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet. 78 * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing 79 * the following structure. 80 */ 81 struct in_addr_4in6 { 82 u_int32_t ia46_pad32[3]; 83 struct in_addr ia46_addr4; 84 }; 85 86 union in_dependaddr { 87 struct in_addr_4in6 id46_addr; 88 struct in6_addr id6_addr; 89 }; 90 91 /* 92 * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553. in_conninfo has 93 * some extra padding to accomplish this. 94 * NOTE 2: tcp_syncache.c uses first 5 32-bit words, which identify fport, 95 * lport, faddr to generate hash, so these fields shouldn't be moved. 96 */ 97 struct in_endpoints { 98 u_int16_t ie_fport; /* foreign port */ 99 u_int16_t ie_lport; /* local port */ 100 /* protocol dependent part, local and foreign addr */ 101 union in_dependaddr ie_dependfaddr; /* foreign host table entry */ 102 union in_dependaddr ie_dependladdr; /* local host table entry */ 103 #define ie_faddr ie_dependfaddr.id46_addr.ia46_addr4 104 #define ie_laddr ie_dependladdr.id46_addr.ia46_addr4 105 #define ie6_faddr ie_dependfaddr.id6_addr 106 #define ie6_laddr ie_dependladdr.id6_addr 107 u_int32_t ie6_zoneid; /* scope zone id */ 108 }; 109 110 /* 111 * XXX The defines for inc_* are hacks and should be changed to direct 112 * references. 113 */ 114 struct in_conninfo { 115 u_int8_t inc_flags; 116 u_int8_t inc_len; 117 u_int16_t inc_fibnum; /* XXX was pad, 16 bits is plenty */ 118 /* protocol dependent part */ 119 struct in_endpoints inc_ie; 120 }; 121 122 /* 123 * Flags for inc_flags. 124 */ 125 #define INC_ISIPV6 0x01 126 #define INC_IPV6MINMTU 0x02 127 128 #define inc_fport inc_ie.ie_fport 129 #define inc_lport inc_ie.ie_lport 130 #define inc_faddr inc_ie.ie_faddr 131 #define inc_laddr inc_ie.ie_laddr 132 #define inc6_faddr inc_ie.ie6_faddr 133 #define inc6_laddr inc_ie.ie6_laddr 134 #define inc6_zoneid inc_ie.ie6_zoneid 135 136 #if defined(_KERNEL) || defined(_WANT_INPCB) 137 /* 138 * struct inpcb captures the network layer state for TCP, UDP, and raw IPv4 and 139 * IPv6 sockets. In the case of TCP and UDP, further per-connection state is 140 * hung off of inp_ppcb most of the time. Almost all fields of struct inpcb 141 * are static after creation or protected by a per-inpcb rwlock, inp_lock. A 142 * few fields are protected by multiple locks as indicated in the locking notes 143 * below. For these fields, all of the listed locks must be write-locked for 144 * any modifications. However, these fields can be safely read while any one of 145 * the listed locks are read-locked. This model can permit greater concurrency 146 * for read operations. For example, connections can be looked up while only 147 * holding a read lock on the global pcblist lock. This is important for 148 * performance when attempting to find the connection for a packet given its IP 149 * and port tuple. 150 * 151 * One noteworthy exception is that the global pcbinfo lock follows a different 152 * set of rules in relation to the inp_list field. Rather than being 153 * write-locked for modifications and read-locked for list iterations, it must 154 * be read-locked during modifications and write-locked during list iterations. 155 * This ensures that the relatively rare global list iterations safely walk a 156 * stable snapshot of connections while allowing more common list modifications 157 * to safely grab the pcblist lock just while adding or removing a connection 158 * from the global list. 159 * 160 * Key: 161 * (b) - Protected by the hpts lock. 162 * (c) - Constant after initialization 163 * (e) - Protected by the net_epoch_prempt epoch 164 * (g) - Protected by the pcbgroup lock 165 * (i) - Protected by the inpcb lock 166 * (p) - Protected by the pcbinfo lock for the inpcb 167 * (l) - Protected by the pcblist lock for the inpcb 168 * (h) - Protected by the pcbhash lock for the inpcb 169 * (s) - Protected by another subsystem's locks 170 * (x) - Undefined locking 171 * 172 * Notes on the tcp_hpts: 173 * 174 * First Hpts lock order is 175 * 1) INP_WLOCK() 176 * 2) HPTS_LOCK() i.e. hpts->pmtx 177 * 178 * To insert a TCB on the hpts you *must* be holding the INP_WLOCK(). 179 * You may check the inp->inp_in_hpts flag without the hpts lock. 180 * The hpts is the only one that will clear this flag holding 181 * only the hpts lock. This means that in your tcp_output() 182 * routine when you test for the inp_in_hpts flag to be 1 183 * it may be transitioning to 0 (by the hpts). 184 * That's ok since that will just mean an extra call to tcp_output 185 * that most likely will find the call you executed 186 * (when the mis-match occured) will have put the TCB back 187 * on the hpts and it will return. If your 188 * call did not add the inp back to the hpts then you will either 189 * over-send or the cwnd will block you from sending more. 190 * 191 * Note you should also be holding the INP_WLOCK() when you 192 * call the remove from the hpts as well. Though usually 193 * you are either doing this from a timer, where you need and have 194 * the INP_WLOCK() or from destroying your TCB where again 195 * you should already have the INP_WLOCK(). 196 * 197 * The inp_hpts_cpu, inp_hpts_cpu_set, inp_input_cpu and 198 * inp_input_cpu_set fields are controlled completely by 199 * the hpts. Do not ever set these. The inp_hpts_cpu_set 200 * and inp_input_cpu_set fields indicate if the hpts has 201 * setup the respective cpu field. It is advised if this 202 * field is 0, to enqueue the packet with the appropriate 203 * hpts_immediate() call. If the _set field is 1, then 204 * you may compare the inp_*_cpu field to the curcpu and 205 * may want to again insert onto the hpts if these fields 206 * are not equal (i.e. you are not on the expected CPU). 207 * 208 * A note on inp_hpts_calls and inp_input_calls, these 209 * flags are set when the hpts calls either the output 210 * or do_segment routines respectively. If the routine 211 * being called wants to use this, then it needs to 212 * clear the flag before returning. The hpts will not 213 * clear the flag. The flags can be used to tell if 214 * the hpts is the function calling the respective 215 * routine. 216 * 217 * A few other notes: 218 * 219 * When a read lock is held, stability of the field is guaranteed; to write 220 * to a field, a write lock must generally be held. 221 * 222 * netinet/netinet6-layer code should not assume that the inp_socket pointer 223 * is safe to dereference without inp_lock being held, even for protocols 224 * other than TCP (where the inpcb persists during TIMEWAIT even after the 225 * socket has been freed), or there may be close(2)-related races. 226 * 227 * The inp_vflag field is overloaded, and would otherwise ideally be (c). 228 * 229 * TODO: Currently only the TCP stack is leveraging the global pcbinfo lock 230 * read-lock usage during modification, this model can be applied to other 231 * protocols (especially SCTP). 232 */ 233 struct icmp6_filter; 234 struct inpcbpolicy; 235 struct m_snd_tag; 236 struct inpcb { 237 /* Cache line #1 (amd64) */ 238 CK_LIST_ENTRY(inpcb) inp_hash; /* [w](h/i) [r](e/i) hash list */ 239 CK_LIST_ENTRY(inpcb) inp_pcbgrouphash; /* (g/i) hash list */ 240 struct rwlock inp_lock; 241 /* Cache line #2 (amd64) */ 242 #define inp_start_zero inp_hpts 243 #define inp_zero_size (sizeof(struct inpcb) - \ 244 offsetof(struct inpcb, inp_start_zero)) 245 TAILQ_ENTRY(inpcb) inp_hpts; /* pacing out queue next lock(b) */ 246 247 uint32_t inp_hpts_request; /* Current hpts request, zero if 248 * fits in the pacing window (i&b). */ 249 /* 250 * Note the next fields are protected by a 251 * different lock (hpts-lock). This means that 252 * they must correspond in size to the smallest 253 * protectable bit field (uint8_t on x86, and 254 * other platfomrs potentially uint32_t?). Also 255 * since CPU switches can occur at different times the two 256 * fields can *not* be collapsed into a signal bit field. 257 */ 258 #if defined(__amd64__) || defined(__i386__) 259 volatile uint8_t inp_in_hpts; /* on output hpts (lock b) */ 260 volatile uint8_t inp_in_input; /* on input hpts (lock b) */ 261 #else 262 volatile uint32_t inp_in_hpts; /* on output hpts (lock b) */ 263 volatile uint32_t inp_in_input; /* on input hpts (lock b) */ 264 #endif 265 volatile uint16_t inp_hpts_cpu; /* Lock (i) */ 266 u_int inp_refcount; /* (i) refcount */ 267 int inp_flags; /* (i) generic IP/datagram flags */ 268 int inp_flags2; /* (i) generic IP/datagram flags #2*/ 269 volatile uint16_t inp_input_cpu; /* Lock (i) */ 270 volatile uint8_t inp_hpts_cpu_set :1, /* on output hpts (i) */ 271 inp_input_cpu_set : 1, /* on input hpts (i) */ 272 inp_hpts_calls :1, /* (i) from output hpts */ 273 inp_input_calls :1, /* (i) from input hpts */ 274 inp_spare_bits2 : 4; 275 uint8_t inp_spare_byte; /* Compiler hole */ 276 void *inp_ppcb; /* (i) pointer to per-protocol pcb */ 277 struct socket *inp_socket; /* (i) back pointer to socket */ 278 uint32_t inp_hptsslot; /* Hpts wheel slot this tcb is Lock(i&b) */ 279 uint32_t inp_hpts_drop_reas; /* reason we are dropping the PCB (lock i&b) */ 280 TAILQ_ENTRY(inpcb) inp_input; /* pacing in queue next lock(b) */ 281 struct inpcbinfo *inp_pcbinfo; /* (c) PCB list info */ 282 struct inpcbgroup *inp_pcbgroup; /* (g/i) PCB group list */ 283 CK_LIST_ENTRY(inpcb) inp_pcbgroup_wild; /* (g/i/h) group wildcard entry */ 284 struct ucred *inp_cred; /* (c) cache of socket cred */ 285 u_int32_t inp_flow; /* (i) IPv6 flow information */ 286 u_char inp_vflag; /* (i) IP version flag (v4/v6) */ 287 u_char inp_ip_ttl; /* (i) time to live proto */ 288 u_char inp_ip_p; /* (c) protocol proto */ 289 u_char inp_ip_minttl; /* (i) minimum TTL or drop */ 290 uint32_t inp_flowid; /* (x) flow id / queue id */ 291 struct m_snd_tag *inp_snd_tag; /* (i) send tag for outgoing mbufs */ 292 uint32_t inp_flowtype; /* (x) M_HASHTYPE value */ 293 uint32_t inp_rss_listen_bucket; /* (x) overridden RSS listen bucket */ 294 295 /* Local and foreign ports, local and foreign addr. */ 296 struct in_conninfo inp_inc; /* (i) list for PCB's local port */ 297 298 /* MAC and IPSEC policy information. */ 299 struct label *inp_label; /* (i) MAC label */ 300 struct inpcbpolicy *inp_sp; /* (s) for IPSEC */ 301 302 /* Protocol-dependent part; options. */ 303 struct { 304 u_char inp_ip_tos; /* (i) type of service proto */ 305 struct mbuf *inp_options; /* (i) IP options */ 306 struct ip_moptions *inp_moptions; /* (i) mcast options */ 307 }; 308 struct { 309 /* (i) IP options */ 310 struct mbuf *in6p_options; 311 /* (i) IP6 options for outgoing packets */ 312 struct ip6_pktopts *in6p_outputopts; 313 /* (i) IP multicast options */ 314 struct ip6_moptions *in6p_moptions; 315 /* (i) ICMPv6 code type filter */ 316 struct icmp6_filter *in6p_icmp6filt; 317 /* (i) IPV6_CHECKSUM setsockopt */ 318 int in6p_cksum; 319 short in6p_hops; 320 }; 321 CK_LIST_ENTRY(inpcb) inp_portlist; /* (i/h) */ 322 struct inpcbport *inp_phd; /* (i/h) head of this list */ 323 inp_gen_t inp_gencnt; /* (c) generation count */ 324 void *spare_ptr; /* Spare pointer. */ 325 rt_gen_t inp_rt_cookie; /* generation for route entry */ 326 union { /* cached L3 information */ 327 struct route inp_route; 328 struct route_in6 inp_route6; 329 }; 330 CK_LIST_ENTRY(inpcb) inp_list; /* (p/l) list for all PCBs for proto */ 331 /* (e[r]) for list iteration */ 332 /* (p[w]/l) for addition/removal */ 333 struct epoch_context inp_epoch_ctx; 334 }; 335 #endif /* _KERNEL */ 336 337 #define inp_fport inp_inc.inc_fport 338 #define inp_lport inp_inc.inc_lport 339 #define inp_faddr inp_inc.inc_faddr 340 #define inp_laddr inp_inc.inc_laddr 341 342 #define in6p_faddr inp_inc.inc6_faddr 343 #define in6p_laddr inp_inc.inc6_laddr 344 #define in6p_zoneid inp_inc.inc6_zoneid 345 #define in6p_flowinfo inp_flow 346 347 #define inp_vnet inp_pcbinfo->ipi_vnet 348 349 /* 350 * The range of the generation count, as used in this implementation, is 9e19. 351 * We would have to create 300 billion connections per second for this number 352 * to roll over in a year. This seems sufficiently unlikely that we simply 353 * don't concern ourselves with that possibility. 354 */ 355 356 /* 357 * Interface exported to userland by various protocols which use inpcbs. Hack 358 * alert -- only define if struct xsocket is in scope. 359 * Fields prefixed with "xi_" are unique to this structure, and the rest 360 * match fields in the struct inpcb, to ease coding and porting. 361 * 362 * Legend: 363 * (s) - used by userland utilities in src 364 * (p) - used by utilities in ports 365 * (3) - is known to be used by third party software not in ports 366 * (n) - no known usage 367 */ 368 #ifdef _SYS_SOCKETVAR_H_ 369 struct xinpcb { 370 ksize_t xi_len; /* length of this structure */ 371 struct xsocket xi_socket; /* (s,p) */ 372 struct in_conninfo inp_inc; /* (s,p) */ 373 uint64_t inp_gencnt; /* (s,p) */ 374 kvaddr_t inp_ppcb; /* (s) netstat(1) */ 375 int64_t inp_spare64[4]; 376 uint32_t inp_flow; /* (s) */ 377 uint32_t inp_flowid; /* (s) */ 378 uint32_t inp_flowtype; /* (s) */ 379 int32_t inp_flags; /* (s,p) */ 380 int32_t inp_flags2; /* (s) */ 381 int32_t inp_rss_listen_bucket; /* (n) */ 382 int32_t in6p_cksum; /* (n) */ 383 int32_t inp_spare32[4]; 384 uint16_t in6p_hops; /* (n) */ 385 uint8_t inp_ip_tos; /* (n) */ 386 int8_t pad8; 387 uint8_t inp_vflag; /* (s,p) */ 388 uint8_t inp_ip_ttl; /* (n) */ 389 uint8_t inp_ip_p; /* (n) */ 390 uint8_t inp_ip_minttl; /* (n) */ 391 int8_t inp_spare8[4]; 392 } __aligned(8); 393 394 struct xinpgen { 395 ksize_t xig_len; /* length of this structure */ 396 u_int xig_count; /* number of PCBs at this time */ 397 uint32_t _xig_spare32; 398 inp_gen_t xig_gen; /* generation count at this time */ 399 so_gen_t xig_sogen; /* socket generation count this time */ 400 uint64_t _xig_spare64[4]; 401 } __aligned(8); 402 #ifdef _KERNEL 403 void in_pcbtoxinpcb(const struct inpcb *, struct xinpcb *); 404 #endif 405 #endif /* _SYS_SOCKETVAR_H_ */ 406 407 struct inpcbport { 408 struct epoch_context phd_epoch_ctx; 409 CK_LIST_ENTRY(inpcbport) phd_hash; 410 struct inpcbhead phd_pcblist; 411 u_short phd_port; 412 }; 413 414 struct in_pcblist { 415 int il_count; 416 struct epoch_context il_epoch_ctx; 417 struct inpcbinfo *il_pcbinfo; 418 struct inpcb *il_inp_list[0]; 419 }; 420 421 /*- 422 * Global data structure for each high-level protocol (UDP, TCP, ...) in both 423 * IPv4 and IPv6. Holds inpcb lists and information for managing them. 424 * 425 * Each pcbinfo is protected by three locks: ipi_lock, ipi_hash_lock and 426 * ipi_list_lock: 427 * - ipi_lock covering the global pcb list stability during loop iteration, 428 * - ipi_hash_lock covering the hashed lookup tables, 429 * - ipi_list_lock covering mutable global fields (such as the global 430 * pcb list) 431 * 432 * The lock order is: 433 * 434 * ipi_lock (before) 435 * inpcb locks (before) 436 * ipi_list locks (before) 437 * {ipi_hash_lock, pcbgroup locks} 438 * 439 * Locking key: 440 * 441 * (c) Constant or nearly constant after initialisation 442 * (e) - Protected by the net_epoch_prempt epoch 443 * (g) Locked by ipi_lock 444 * (l) Locked by ipi_list_lock 445 * (h) Read using either net_epoch_preempt or inpcb lock; write requires both ipi_hash_lock and inpcb lock 446 * (p) Protected by one or more pcbgroup locks 447 * (x) Synchronisation properties poorly defined 448 */ 449 struct inpcbinfo { 450 /* 451 * Global lock protecting inpcb list modification 452 */ 453 struct mtx ipi_lock; 454 455 /* 456 * Global list of inpcbs on the protocol. 457 */ 458 struct inpcbhead *ipi_listhead; /* [r](e) [w](g/l) */ 459 u_int ipi_count; /* (l) */ 460 461 /* 462 * Generation count -- incremented each time a connection is allocated 463 * or freed. 464 */ 465 u_quad_t ipi_gencnt; /* (l) */ 466 467 /* 468 * Fields associated with port lookup and allocation. 469 */ 470 u_short ipi_lastport; /* (x) */ 471 u_short ipi_lastlow; /* (x) */ 472 u_short ipi_lasthi; /* (x) */ 473 474 /* 475 * UMA zone from which inpcbs are allocated for this protocol. 476 */ 477 struct uma_zone *ipi_zone; /* (c) */ 478 479 /* 480 * Connection groups associated with this protocol. These fields are 481 * constant, but pcbgroup structures themselves are protected by 482 * per-pcbgroup locks. 483 */ 484 struct inpcbgroup *ipi_pcbgroups; /* (c) */ 485 u_int ipi_npcbgroups; /* (c) */ 486 u_int ipi_hashfields; /* (c) */ 487 488 /* 489 * Global lock protecting modification non-pcbgroup hash lookup tables. 490 */ 491 struct mtx ipi_hash_lock; 492 493 /* 494 * Global hash of inpcbs, hashed by local and foreign addresses and 495 * port numbers. 496 */ 497 struct inpcbhead *ipi_hashbase; /* (h) */ 498 u_long ipi_hashmask; /* (h) */ 499 500 /* 501 * Global hash of inpcbs, hashed by only local port number. 502 */ 503 struct inpcbporthead *ipi_porthashbase; /* (h) */ 504 u_long ipi_porthashmask; /* (h) */ 505 506 /* 507 * List of wildcard inpcbs for use with pcbgroups. In the past, was 508 * per-pcbgroup but is now global. All pcbgroup locks must be held 509 * to modify the list, so any is sufficient to read it. 510 */ 511 struct inpcbhead *ipi_wildbase; /* (p) */ 512 u_long ipi_wildmask; /* (p) */ 513 514 /* 515 * Load balance groups used for the SO_REUSEPORT_LB option, 516 * hashed by local port. 517 */ 518 struct inpcblbgrouphead *ipi_lbgrouphashbase; /* (h) */ 519 u_long ipi_lbgrouphashmask; /* (h) */ 520 521 /* 522 * Pointer to network stack instance 523 */ 524 struct vnet *ipi_vnet; /* (c) */ 525 526 /* 527 * general use 2 528 */ 529 void *ipi_pspare[2]; 530 531 /* 532 * Global lock protecting global inpcb list, inpcb count, etc. 533 */ 534 struct rwlock ipi_list_lock; 535 }; 536 537 #ifdef _KERNEL 538 /* 539 * Connection groups hold sets of connections that have similar CPU/thread 540 * affinity. Each connection belongs to exactly one connection group. 541 */ 542 struct inpcbgroup { 543 /* 544 * Per-connection group hash of inpcbs, hashed by local and foreign 545 * addresses and port numbers. 546 */ 547 struct inpcbhead *ipg_hashbase; /* (c) */ 548 u_long ipg_hashmask; /* (c) */ 549 550 /* 551 * Notional affinity of this pcbgroup. 552 */ 553 u_int ipg_cpu; /* (p) */ 554 555 /* 556 * Per-connection group lock, not to be confused with ipi_lock. 557 * Protects the hash table hung off the group, but also the global 558 * wildcard list in inpcbinfo. 559 */ 560 struct mtx ipg_lock; 561 } __aligned(CACHE_LINE_SIZE); 562 563 /* 564 * Load balance groups used for the SO_REUSEPORT_LB socket option. Each group 565 * (or unique address:port combination) can be re-used at most 566 * INPCBLBGROUP_SIZMAX (256) times. The inpcbs are stored in il_inp which 567 * is dynamically resized as processes bind/unbind to that specific group. 568 */ 569 struct inpcblbgroup { 570 CK_LIST_ENTRY(inpcblbgroup) il_list; 571 struct epoch_context il_epoch_ctx; 572 uint16_t il_lport; /* (c) */ 573 u_char il_vflag; /* (c) */ 574 u_char il_pad; 575 uint32_t il_pad2; 576 union in_dependaddr il_dependladdr; /* (c) */ 577 #define il_laddr il_dependladdr.id46_addr.ia46_addr4 578 #define il6_laddr il_dependladdr.id6_addr 579 uint32_t il_inpsiz; /* max count in il_inp[] (h) */ 580 uint32_t il_inpcnt; /* cur count in il_inp[] (h) */ 581 struct inpcb *il_inp[]; /* (h) */ 582 }; 583 584 #define INP_LOCK_INIT(inp, d, t) \ 585 rw_init_flags(&(inp)->inp_lock, (t), RW_RECURSE | RW_DUPOK) 586 #define INP_LOCK_DESTROY(inp) rw_destroy(&(inp)->inp_lock) 587 #define INP_RLOCK(inp) rw_rlock(&(inp)->inp_lock) 588 #define INP_WLOCK(inp) rw_wlock(&(inp)->inp_lock) 589 #define INP_TRY_RLOCK(inp) rw_try_rlock(&(inp)->inp_lock) 590 #define INP_TRY_WLOCK(inp) rw_try_wlock(&(inp)->inp_lock) 591 #define INP_RUNLOCK(inp) rw_runlock(&(inp)->inp_lock) 592 #define INP_WUNLOCK(inp) rw_wunlock(&(inp)->inp_lock) 593 #define INP_TRY_UPGRADE(inp) rw_try_upgrade(&(inp)->inp_lock) 594 #define INP_DOWNGRADE(inp) rw_downgrade(&(inp)->inp_lock) 595 #define INP_WLOCKED(inp) rw_wowned(&(inp)->inp_lock) 596 #define INP_LOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_LOCKED) 597 #define INP_RLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_RLOCKED) 598 #define INP_WLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_WLOCKED) 599 #define INP_UNLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_UNLOCKED) 600 601 /* 602 * These locking functions are for inpcb consumers outside of sys/netinet, 603 * more specifically, they were added for the benefit of TOE drivers. The 604 * macros are reserved for use by the stack. 605 */ 606 void inp_wlock(struct inpcb *); 607 void inp_wunlock(struct inpcb *); 608 void inp_rlock(struct inpcb *); 609 void inp_runlock(struct inpcb *); 610 611 #ifdef INVARIANT_SUPPORT 612 void inp_lock_assert(struct inpcb *); 613 void inp_unlock_assert(struct inpcb *); 614 #else 615 #define inp_lock_assert(inp) do {} while (0) 616 #define inp_unlock_assert(inp) do {} while (0) 617 #endif 618 619 void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg); 620 int inp_ip_tos_get(const struct inpcb *inp); 621 void inp_ip_tos_set(struct inpcb *inp, int val); 622 struct socket * 623 inp_inpcbtosocket(struct inpcb *inp); 624 struct tcpcb * 625 inp_inpcbtotcpcb(struct inpcb *inp); 626 void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 627 uint32_t *faddr, uint16_t *fp); 628 int inp_so_options(const struct inpcb *inp); 629 630 #endif /* _KERNEL */ 631 632 #define INP_INFO_LOCK_INIT(ipi, d) \ 633 mtx_init(&(ipi)->ipi_lock, (d), NULL, MTX_DEF| MTX_RECURSE) 634 #define INP_INFO_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_lock) 635 #define INP_INFO_RLOCK_ET(ipi, et) NET_EPOCH_ENTER((et)) 636 #define INP_INFO_WLOCK(ipi) mtx_lock(&(ipi)->ipi_lock) 637 #define INP_INFO_TRY_WLOCK(ipi) mtx_trylock(&(ipi)->ipi_lock) 638 #define INP_INFO_WLOCKED(ipi) mtx_owned(&(ipi)->ipi_lock) 639 #define INP_INFO_RUNLOCK_ET(ipi, et) NET_EPOCH_EXIT((et)) 640 #define INP_INFO_RUNLOCK_TP(ipi, tp) NET_EPOCH_EXIT(*(tp)->t_inpcb->inp_et) 641 #define INP_INFO_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_lock) 642 #define INP_INFO_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_lock)) 643 #define INP_INFO_RLOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt)) 644 #define INP_INFO_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_lock, MA_OWNED) 645 #define INP_INFO_WUNLOCK_ASSERT(ipi) \ 646 mtx_assert(&(ipi)->ipi_lock, MA_NOTOWNED) 647 #define INP_INFO_UNLOCK_ASSERT(ipi) MPASS(!in_epoch(net_epoch_preempt) && !mtx_owned(&(ipi)->ipi_lock)) 648 649 #define INP_LIST_LOCK_INIT(ipi, d) \ 650 rw_init_flags(&(ipi)->ipi_list_lock, (d), 0) 651 #define INP_LIST_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_list_lock) 652 #define INP_LIST_RLOCK(ipi) rw_rlock(&(ipi)->ipi_list_lock) 653 #define INP_LIST_WLOCK(ipi) rw_wlock(&(ipi)->ipi_list_lock) 654 #define INP_LIST_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_list_lock) 655 #define INP_LIST_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_list_lock) 656 #define INP_LIST_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_list_lock) 657 #define INP_LIST_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_list_lock) 658 #define INP_LIST_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_list_lock) 659 #define INP_LIST_LOCK_ASSERT(ipi) \ 660 rw_assert(&(ipi)->ipi_list_lock, RA_LOCKED) 661 #define INP_LIST_RLOCK_ASSERT(ipi) \ 662 rw_assert(&(ipi)->ipi_list_lock, RA_RLOCKED) 663 #define INP_LIST_WLOCK_ASSERT(ipi) \ 664 rw_assert(&(ipi)->ipi_list_lock, RA_WLOCKED) 665 #define INP_LIST_UNLOCK_ASSERT(ipi) \ 666 rw_assert(&(ipi)->ipi_list_lock, RA_UNLOCKED) 667 668 #define INP_HASH_LOCK_INIT(ipi, d) mtx_init(&(ipi)->ipi_hash_lock, (d), NULL, MTX_DEF) 669 #define INP_HASH_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_hash_lock) 670 #define INP_HASH_RLOCK(ipi) struct epoch_tracker inp_hash_et; epoch_enter_preempt(net_epoch_preempt, &inp_hash_et) 671 #define INP_HASH_RLOCK_ET(ipi, et) epoch_enter_preempt(net_epoch_preempt, &(et)) 672 #define INP_HASH_WLOCK(ipi) mtx_lock(&(ipi)->ipi_hash_lock) 673 #define INP_HASH_RUNLOCK(ipi) NET_EPOCH_EXIT(inp_hash_et) 674 #define INP_HASH_RUNLOCK_ET(ipi, et) NET_EPOCH_EXIT((et)) 675 #define INP_HASH_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_hash_lock) 676 #define INP_HASH_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_hash_lock)) 677 #define INP_HASH_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_hash_lock, MA_OWNED); 678 679 #define INP_GROUP_LOCK_INIT(ipg, d) mtx_init(&(ipg)->ipg_lock, (d), NULL, \ 680 MTX_DEF | MTX_DUPOK) 681 #define INP_GROUP_LOCK_DESTROY(ipg) mtx_destroy(&(ipg)->ipg_lock) 682 683 #define INP_GROUP_LOCK(ipg) mtx_lock(&(ipg)->ipg_lock) 684 #define INP_GROUP_LOCK_ASSERT(ipg) mtx_assert(&(ipg)->ipg_lock, MA_OWNED) 685 #define INP_GROUP_UNLOCK(ipg) mtx_unlock(&(ipg)->ipg_lock) 686 687 #define INP_PCBHASH(faddr, lport, fport, mask) \ 688 (((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask)) 689 #define INP_PCBPORTHASH(lport, mask) \ 690 (ntohs((lport)) & (mask)) 691 #define INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) \ 692 ((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) 693 #define INP6_PCBHASHKEY(faddr) ((faddr)->s6_addr32[3]) 694 695 /* 696 * Flags for inp_vflags -- historically version flags only 697 */ 698 #define INP_IPV4 0x1 699 #define INP_IPV6 0x2 700 #define INP_IPV6PROTO 0x4 /* opened under IPv6 protocol */ 701 702 /* 703 * Flags for inp_flags. 704 */ 705 #define INP_RECVOPTS 0x00000001 /* receive incoming IP options */ 706 #define INP_RECVRETOPTS 0x00000002 /* receive IP options for reply */ 707 #define INP_RECVDSTADDR 0x00000004 /* receive IP dst address */ 708 #define INP_HDRINCL 0x00000008 /* user supplies entire IP header */ 709 #define INP_HIGHPORT 0x00000010 /* user wants "high" port binding */ 710 #define INP_LOWPORT 0x00000020 /* user wants "low" port binding */ 711 #define INP_ANONPORT 0x00000040 /* port chosen for user */ 712 #define INP_RECVIF 0x00000080 /* receive incoming interface */ 713 #define INP_MTUDISC 0x00000100 /* user can do MTU discovery */ 714 /* 0x000200 unused: was INP_FAITH */ 715 #define INP_RECVTTL 0x00000400 /* receive incoming IP TTL */ 716 #define INP_DONTFRAG 0x00000800 /* don't fragment packet */ 717 #define INP_BINDANY 0x00001000 /* allow bind to any address */ 718 #define INP_INHASHLIST 0x00002000 /* in_pcbinshash() has been called */ 719 #define INP_RECVTOS 0x00004000 /* receive incoming IP TOS */ 720 #define IN6P_IPV6_V6ONLY 0x00008000 /* restrict AF_INET6 socket for v6 */ 721 #define IN6P_PKTINFO 0x00010000 /* receive IP6 dst and I/F */ 722 #define IN6P_HOPLIMIT 0x00020000 /* receive hoplimit */ 723 #define IN6P_HOPOPTS 0x00040000 /* receive hop-by-hop options */ 724 #define IN6P_DSTOPTS 0x00080000 /* receive dst options after rthdr */ 725 #define IN6P_RTHDR 0x00100000 /* receive routing header */ 726 #define IN6P_RTHDRDSTOPTS 0x00200000 /* receive dstoptions before rthdr */ 727 #define IN6P_TCLASS 0x00400000 /* receive traffic class value */ 728 #define IN6P_AUTOFLOWLABEL 0x00800000 /* attach flowlabel automatically */ 729 #define INP_TIMEWAIT 0x01000000 /* in TIMEWAIT, ppcb is tcptw */ 730 #define INP_ONESBCAST 0x02000000 /* send all-ones broadcast */ 731 #define INP_DROPPED 0x04000000 /* protocol drop flag */ 732 #define INP_SOCKREF 0x08000000 /* strong socket reference */ 733 #define INP_RESERVED_0 0x10000000 /* reserved field */ 734 #define INP_RESERVED_1 0x20000000 /* reserved field */ 735 #define IN6P_RFC2292 0x40000000 /* used RFC2292 API on the socket */ 736 #define IN6P_MTU 0x80000000 /* receive path MTU */ 737 738 #define INP_CONTROLOPTS (INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\ 739 INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\ 740 IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\ 741 IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\ 742 IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\ 743 IN6P_MTU) 744 745 /* 746 * Flags for inp_flags2. 747 */ 748 #define INP_2UNUSED1 0x00000001 749 #define INP_2UNUSED2 0x00000002 750 #define INP_PCBGROUPWILD 0x00000004 /* in pcbgroup wildcard list */ 751 #define INP_REUSEPORT 0x00000008 /* SO_REUSEPORT option is set */ 752 #define INP_FREED 0x00000010 /* inp itself is not valid */ 753 #define INP_REUSEADDR 0x00000020 /* SO_REUSEADDR option is set */ 754 #define INP_BINDMULTI 0x00000040 /* IP_BINDMULTI option is set */ 755 #define INP_RSS_BUCKET_SET 0x00000080 /* IP_RSS_LISTEN_BUCKET is set */ 756 #define INP_RECVFLOWID 0x00000100 /* populate recv datagram with flow info */ 757 #define INP_RECVRSSBUCKETID 0x00000200 /* populate recv datagram with bucket id */ 758 #define INP_RATE_LIMIT_CHANGED 0x00000400 /* rate limit needs attention */ 759 #define INP_ORIGDSTADDR 0x00000800 /* receive IP dst address/port */ 760 #define INP_CANNOT_DO_ECN 0x00001000 /* The stack does not do ECN */ 761 #define INP_REUSEPORT_LB 0x00002000 /* SO_REUSEPORT_LB option is set */ 762 763 /* 764 * Flags passed to in_pcblookup*() functions. 765 */ 766 #define INPLOOKUP_WILDCARD 0x00000001 /* Allow wildcard sockets. */ 767 #define INPLOOKUP_RLOCKPCB 0x00000002 /* Return inpcb read-locked. */ 768 #define INPLOOKUP_WLOCKPCB 0x00000004 /* Return inpcb write-locked. */ 769 770 #define INPLOOKUP_MASK (INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB | \ 771 INPLOOKUP_WLOCKPCB) 772 773 #define sotoinpcb(so) ((struct inpcb *)(so)->so_pcb) 774 #define sotoin6pcb(so) sotoinpcb(so) /* for KAME src sync over BSD*'s */ 775 776 #define INP_SOCKAF(so) so->so_proto->pr_domain->dom_family 777 778 #define INP_CHECK_SOCKAF(so, af) (INP_SOCKAF(so) == af) 779 780 /* 781 * Constants for pcbinfo.ipi_hashfields. 782 */ 783 #define IPI_HASHFIELDS_NONE 0 784 #define IPI_HASHFIELDS_2TUPLE 1 785 #define IPI_HASHFIELDS_4TUPLE 2 786 787 #ifdef _KERNEL 788 VNET_DECLARE(int, ipport_reservedhigh); 789 VNET_DECLARE(int, ipport_reservedlow); 790 VNET_DECLARE(int, ipport_lowfirstauto); 791 VNET_DECLARE(int, ipport_lowlastauto); 792 VNET_DECLARE(int, ipport_firstauto); 793 VNET_DECLARE(int, ipport_lastauto); 794 VNET_DECLARE(int, ipport_hifirstauto); 795 VNET_DECLARE(int, ipport_hilastauto); 796 VNET_DECLARE(int, ipport_randomized); 797 VNET_DECLARE(int, ipport_randomcps); 798 VNET_DECLARE(int, ipport_randomtime); 799 VNET_DECLARE(int, ipport_stoprandom); 800 VNET_DECLARE(int, ipport_tcpallocs); 801 802 #define V_ipport_reservedhigh VNET(ipport_reservedhigh) 803 #define V_ipport_reservedlow VNET(ipport_reservedlow) 804 #define V_ipport_lowfirstauto VNET(ipport_lowfirstauto) 805 #define V_ipport_lowlastauto VNET(ipport_lowlastauto) 806 #define V_ipport_firstauto VNET(ipport_firstauto) 807 #define V_ipport_lastauto VNET(ipport_lastauto) 808 #define V_ipport_hifirstauto VNET(ipport_hifirstauto) 809 #define V_ipport_hilastauto VNET(ipport_hilastauto) 810 #define V_ipport_randomized VNET(ipport_randomized) 811 #define V_ipport_randomcps VNET(ipport_randomcps) 812 #define V_ipport_randomtime VNET(ipport_randomtime) 813 #define V_ipport_stoprandom VNET(ipport_stoprandom) 814 #define V_ipport_tcpallocs VNET(ipport_tcpallocs) 815 816 void in_pcbinfo_destroy(struct inpcbinfo *); 817 void in_pcbinfo_init(struct inpcbinfo *, const char *, struct inpcbhead *, 818 int, int, char *, uma_init, u_int); 819 820 int in_pcbbind_check_bindmulti(const struct inpcb *ni, 821 const struct inpcb *oi); 822 823 struct inpcbgroup * 824 in_pcbgroup_byhash(struct inpcbinfo *, u_int, uint32_t); 825 struct inpcbgroup * 826 in_pcbgroup_byinpcb(struct inpcb *); 827 struct inpcbgroup * 828 in_pcbgroup_bytuple(struct inpcbinfo *, struct in_addr, u_short, 829 struct in_addr, u_short); 830 void in_pcbgroup_destroy(struct inpcbinfo *); 831 int in_pcbgroup_enabled(struct inpcbinfo *); 832 void in_pcbgroup_init(struct inpcbinfo *, u_int, int); 833 void in_pcbgroup_remove(struct inpcb *); 834 void in_pcbgroup_update(struct inpcb *); 835 void in_pcbgroup_update_mbuf(struct inpcb *, struct mbuf *); 836 837 void in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *); 838 int in_pcballoc(struct socket *, struct inpcbinfo *); 839 int in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *); 840 int in_pcb_lport(struct inpcb *, struct in_addr *, u_short *, 841 struct ucred *, int); 842 int in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *, 843 u_short *, struct ucred *); 844 int in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *); 845 int in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *, 846 struct mbuf *); 847 int in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *, 848 u_short *, in_addr_t *, u_short *, struct inpcb **, 849 struct ucred *); 850 void in_pcbdetach(struct inpcb *); 851 void in_pcbdisconnect(struct inpcb *); 852 void in_pcbdrop(struct inpcb *); 853 void in_pcbfree(struct inpcb *); 854 int in_pcbinshash(struct inpcb *); 855 int in_pcbinshash_nopcbgroup(struct inpcb *); 856 int in_pcbladdr(struct inpcb *, struct in_addr *, struct in_addr *, 857 struct ucred *); 858 struct inpcb * 859 in_pcblookup_local(struct inpcbinfo *, 860 struct in_addr, u_short, int, struct ucred *); 861 struct inpcb * 862 in_pcblookup(struct inpcbinfo *, struct in_addr, u_int, 863 struct in_addr, u_int, int, struct ifnet *); 864 struct inpcb * 865 in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int, 866 struct in_addr, u_int, int, struct ifnet *, struct mbuf *); 867 void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr, 868 int, struct inpcb *(*)(struct inpcb *, int)); 869 void in_pcbref(struct inpcb *); 870 void in_pcbrehash(struct inpcb *); 871 void in_pcbrehash_mbuf(struct inpcb *, struct mbuf *); 872 int in_pcbrele(struct inpcb *); 873 int in_pcbrele_rlocked(struct inpcb *); 874 int in_pcbrele_wlocked(struct inpcb *); 875 void in_pcblist_rele_rlocked(epoch_context_t ctx); 876 void in_losing(struct inpcb *); 877 void in_pcbsetsolabel(struct socket *so); 878 int in_getpeeraddr(struct socket *so, struct sockaddr **nam); 879 int in_getsockaddr(struct socket *so, struct sockaddr **nam); 880 struct sockaddr * 881 in_sockaddr(in_port_t port, struct in_addr *addr); 882 void in_pcbsosetlabel(struct socket *so); 883 #ifdef RATELIMIT 884 int in_pcbattach_txrtlmt(struct inpcb *, struct ifnet *, uint32_t, uint32_t, uint32_t); 885 void in_pcbdetach_txrtlmt(struct inpcb *); 886 int in_pcbmodify_txrtlmt(struct inpcb *, uint32_t); 887 int in_pcbquery_txrtlmt(struct inpcb *, uint32_t *); 888 int in_pcbquery_txrlevel(struct inpcb *, uint32_t *); 889 void in_pcboutput_txrtlmt(struct inpcb *, struct ifnet *, struct mbuf *); 890 void in_pcboutput_eagain(struct inpcb *); 891 #endif 892 #endif /* _KERNEL */ 893 894 #endif /* !_NETINET_IN_PCB_H_ */ 895