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 typedef uint64_t inp_gen_t; 74 75 /* 76 * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet. 77 * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing 78 * the following structure. 79 */ 80 struct in_addr_4in6 { 81 u_int32_t ia46_pad32[3]; 82 struct in_addr ia46_addr4; 83 }; 84 85 union in_dependaddr { 86 struct in_addr_4in6 id46_addr; 87 struct in6_addr id6_addr; 88 }; 89 90 /* 91 * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553. in_conninfo has 92 * some extra padding to accomplish this. 93 * NOTE 2: tcp_syncache.c uses first 5 32-bit words, which identify fport, 94 * lport, faddr to generate hash, so these fields shouldn't be moved. 95 */ 96 struct in_endpoints { 97 u_int16_t ie_fport; /* foreign port */ 98 u_int16_t ie_lport; /* local port */ 99 /* protocol dependent part, local and foreign addr */ 100 union in_dependaddr ie_dependfaddr; /* foreign host table entry */ 101 union in_dependaddr ie_dependladdr; /* local host table entry */ 102 #define ie_faddr ie_dependfaddr.id46_addr.ia46_addr4 103 #define ie_laddr ie_dependladdr.id46_addr.ia46_addr4 104 #define ie6_faddr ie_dependfaddr.id6_addr 105 #define ie6_laddr ie_dependladdr.id6_addr 106 u_int32_t ie6_zoneid; /* scope zone id */ 107 }; 108 109 /* 110 * XXX The defines for inc_* are hacks and should be changed to direct 111 * references. 112 */ 113 struct in_conninfo { 114 u_int8_t inc_flags; 115 u_int8_t inc_len; 116 u_int16_t inc_fibnum; /* XXX was pad, 16 bits is plenty */ 117 /* protocol dependent part */ 118 struct in_endpoints inc_ie; 119 }; 120 121 /* 122 * Flags for inc_flags. 123 */ 124 #define INC_ISIPV6 0x01 125 126 #define inc_isipv6 inc_flags /* temp compatibility */ 127 #define inc_fport inc_ie.ie_fport 128 #define inc_lport inc_ie.ie_lport 129 #define inc_faddr inc_ie.ie_faddr 130 #define inc_laddr inc_ie.ie_laddr 131 #define inc6_faddr inc_ie.ie6_faddr 132 #define inc6_laddr inc_ie.ie6_laddr 133 #define inc6_zoneid inc_ie.ie6_zoneid 134 135 #if defined(_KERNEL) || defined(_WANT_INPCB) 136 /* 137 * struct inpcb captures the network layer state for TCP, UDP, and raw IPv4 and 138 * IPv6 sockets. In the case of TCP and UDP, further per-connection state is 139 * hung off of inp_ppcb most of the time. Almost all fields of struct inpcb 140 * are static after creation or protected by a per-inpcb rwlock, inp_lock. A 141 * few fields are protected by multiple locks as indicated in the locking notes 142 * below. For these fields, all of the listed locks must be write-locked for 143 * any modifications. However, these fields can be safely read while any one of 144 * the listed locks are read-locked. This model can permit greater concurrency 145 * for read operations. For example, connections can be looked up while only 146 * holding a read lock on the global pcblist lock. This is important for 147 * performance when attempting to find the connection for a packet given its IP 148 * and port tuple. 149 * 150 * One noteworthy exception is that the global pcbinfo lock follows a different 151 * set of rules in relation to the inp_list field. Rather than being 152 * write-locked for modifications and read-locked for list iterations, it must 153 * be read-locked during modifications and write-locked during list iterations. 154 * This ensures that the relatively rare global list iterations safely walk a 155 * stable snapshot of connections while allowing more common list modifications 156 * to safely grab the pcblist lock just while adding or removing a connection 157 * from the global list. 158 * 159 * Key: 160 * (b) - Protected by the hpts lock. 161 * (c) - Constant after initialization 162 * (e) - Protected by the net_epoch_prempt epoch 163 * (g) - Protected by the pcbgroup lock 164 * (i) - Protected by the inpcb lock 165 * (p) - Protected by the pcbinfo lock for the inpcb 166 * (l) - Protected by the pcblist lock for the inpcb 167 * (h) - Protected by the pcbhash lock for the inpcb 168 * (s) - Protected by another subsystem's locks 169 * (x) - Undefined locking 170 * 171 * Notes on the tcp_hpts: 172 * 173 * First Hpts lock order is 174 * 1) INP_WLOCK() 175 * 2) HPTS_LOCK() i.e. hpts->pmtx 176 * 177 * To insert a TCB on the hpts you *must* be holding the INP_WLOCK(). 178 * You may check the inp->inp_in_hpts flag without the hpts lock. 179 * The hpts is the only one that will clear this flag holding 180 * only the hpts lock. This means that in your tcp_output() 181 * routine when you test for the inp_in_hpts flag to be 1 182 * it may be transitioning to 0 (by the hpts). 183 * That's ok since that will just mean an extra call to tcp_output 184 * that most likely will find the call you executed 185 * (when the mis-match occured) will have put the TCB back 186 * on the hpts and it will return. If your 187 * call did not add the inp back to the hpts then you will either 188 * over-send or the cwnd will block you from sending more. 189 * 190 * Note you should also be holding the INP_WLOCK() when you 191 * call the remove from the hpts as well. Though usually 192 * you are either doing this from a timer, where you need and have 193 * the INP_WLOCK() or from destroying your TCB where again 194 * you should already have the INP_WLOCK(). 195 * 196 * The inp_hpts_cpu, inp_hpts_cpu_set, inp_input_cpu and 197 * inp_input_cpu_set fields are controlled completely by 198 * the hpts. Do not ever set these. The inp_hpts_cpu_set 199 * and inp_input_cpu_set fields indicate if the hpts has 200 * setup the respective cpu field. It is advised if this 201 * field is 0, to enqueue the packet with the appropriate 202 * hpts_immediate() call. If the _set field is 1, then 203 * you may compare the inp_*_cpu field to the curcpu and 204 * may want to again insert onto the hpts if these fields 205 * are not equal (i.e. you are not on the expected CPU). 206 * 207 * A note on inp_hpts_calls and inp_input_calls, these 208 * flags are set when the hpts calls either the output 209 * or do_segment routines respectively. If the routine 210 * being called wants to use this, then it needs to 211 * clear the flag before returning. The hpts will not 212 * clear the flag. The flags can be used to tell if 213 * the hpts is the function calling the respective 214 * routine. 215 * 216 * A few other notes: 217 * 218 * When a read lock is held, stability of the field is guaranteed; to write 219 * to a field, a write lock must generally be held. 220 * 221 * netinet/netinet6-layer code should not assume that the inp_socket pointer 222 * is safe to dereference without inp_lock being held, even for protocols 223 * other than TCP (where the inpcb persists during TIMEWAIT even after the 224 * socket has been freed), or there may be close(2)-related races. 225 * 226 * The inp_vflag field is overloaded, and would otherwise ideally be (c). 227 * 228 * TODO: Currently only the TCP stack is leveraging the global pcbinfo lock 229 * read-lock usage during modification, this model can be applied to other 230 * protocols (especially SCTP). 231 */ 232 struct icmp6_filter; 233 struct inpcbpolicy; 234 struct m_snd_tag; 235 struct inpcb { 236 /* Cache line #1 (amd64) */ 237 CK_LIST_ENTRY(inpcb) inp_hash; /* [w](h/i) [r](e/i) hash list */ 238 CK_LIST_ENTRY(inpcb) inp_pcbgrouphash; /* (g/i) hash list */ 239 struct rwlock inp_lock; 240 /* Cache line #2 (amd64) */ 241 #define inp_start_zero inp_hpts 242 #define inp_zero_size (sizeof(struct inpcb) - \ 243 offsetof(struct inpcb, inp_start_zero)) 244 TAILQ_ENTRY(inpcb) inp_hpts; /* pacing out queue next lock(b) */ 245 246 uint32_t inp_hpts_request; /* Current hpts request, zero if 247 * fits in the pacing window (i&b). */ 248 /* 249 * Note the next fields are protected by a 250 * different lock (hpts-lock). This means that 251 * they must correspond in size to the smallest 252 * protectable bit field (uint8_t on x86, and 253 * other platfomrs potentially uint32_t?). Also 254 * since CPU switches can occur at different times the two 255 * fields can *not* be collapsed into a signal bit field. 256 */ 257 #if defined(__amd64__) || defined(__i386__) 258 volatile uint8_t inp_in_hpts; /* on output hpts (lock b) */ 259 volatile uint8_t inp_in_input; /* on input hpts (lock b) */ 260 #else 261 volatile uint32_t inp_in_hpts; /* on output hpts (lock b) */ 262 volatile uint32_t inp_in_input; /* on input hpts (lock b) */ 263 #endif 264 volatile uint16_t inp_hpts_cpu; /* Lock (i) */ 265 u_int inp_refcount; /* (i) refcount */ 266 int inp_flags; /* (i) generic IP/datagram flags */ 267 int inp_flags2; /* (i) generic IP/datagram flags #2*/ 268 volatile uint16_t inp_input_cpu; /* Lock (i) */ 269 volatile uint8_t inp_hpts_cpu_set :1, /* on output hpts (i) */ 270 inp_input_cpu_set : 1, /* on input hpts (i) */ 271 inp_hpts_calls :1, /* (i) from output hpts */ 272 inp_input_calls :1, /* (i) from input hpts */ 273 inp_spare_bits2 : 4; 274 uint8_t inp_spare_byte; /* Compiler hole */ 275 void *inp_ppcb; /* (i) pointer to per-protocol pcb */ 276 struct socket *inp_socket; /* (i) back pointer to socket */ 277 uint32_t inp_hptsslot; /* Hpts wheel slot this tcb is Lock(i&b) */ 278 uint32_t inp_hpts_drop_reas; /* reason we are dropping the PCB (lock i&b) */ 279 TAILQ_ENTRY(inpcb) inp_input; /* pacing in queue next lock(b) */ 280 struct inpcbinfo *inp_pcbinfo; /* (c) PCB list info */ 281 struct inpcbgroup *inp_pcbgroup; /* (g/i) PCB group list */ 282 CK_LIST_ENTRY(inpcb) inp_pcbgroup_wild; /* (g/i/h) group wildcard entry */ 283 struct ucred *inp_cred; /* (c) cache of socket cred */ 284 u_int32_t inp_flow; /* (i) IPv6 flow information */ 285 u_char inp_vflag; /* (i) IP version flag (v4/v6) */ 286 u_char inp_ip_ttl; /* (i) time to live proto */ 287 u_char inp_ip_p; /* (c) protocol proto */ 288 u_char inp_ip_minttl; /* (i) minimum TTL or drop */ 289 uint32_t inp_flowid; /* (x) flow id / queue id */ 290 struct m_snd_tag *inp_snd_tag; /* (i) send tag for outgoing mbufs */ 291 uint32_t inp_flowtype; /* (x) M_HASHTYPE value */ 292 uint32_t inp_rss_listen_bucket; /* (x) overridden RSS listen bucket */ 293 294 /* Local and foreign ports, local and foreign addr. */ 295 struct in_conninfo inp_inc; /* (i) list for PCB's local port */ 296 297 /* MAC and IPSEC policy information. */ 298 struct label *inp_label; /* (i) MAC label */ 299 struct inpcbpolicy *inp_sp; /* (s) for IPSEC */ 300 301 /* Protocol-dependent part; options. */ 302 struct { 303 u_char inp_ip_tos; /* (i) type of service proto */ 304 struct mbuf *inp_options; /* (i) IP options */ 305 struct ip_moptions *inp_moptions; /* (i) mcast options */ 306 }; 307 struct { 308 /* (i) IP options */ 309 struct mbuf *in6p_options; 310 /* (i) IP6 options for outgoing packets */ 311 struct ip6_pktopts *in6p_outputopts; 312 /* (i) IP multicast options */ 313 struct ip6_moptions *in6p_moptions; 314 /* (i) ICMPv6 code type filter */ 315 struct icmp6_filter *in6p_icmp6filt; 316 /* (i) IPV6_CHECKSUM setsockopt */ 317 int in6p_cksum; 318 short in6p_hops; 319 }; 320 CK_LIST_ENTRY(inpcb) inp_portlist; /* (i/h) */ 321 struct inpcbport *inp_phd; /* (i/h) head of this list */ 322 inp_gen_t inp_gencnt; /* (c) generation count */ 323 struct llentry *inp_lle; /* cached L2 information */ 324 rt_gen_t inp_rt_cookie; /* generation for route entry */ 325 union { /* cached L3 information */ 326 struct route inp_route; 327 struct route_in6 inp_route6; 328 }; 329 CK_LIST_ENTRY(inpcb) inp_list; /* (p/l) list for all PCBs for proto */ 330 /* (e[r]) for list iteration */ 331 /* (p[w]/l) for addition/removal */ 332 struct epoch_context inp_epoch_ctx; 333 }; 334 #endif /* _KERNEL */ 335 336 #define inp_fport inp_inc.inc_fport 337 #define inp_lport inp_inc.inc_lport 338 #define inp_faddr inp_inc.inc_faddr 339 #define inp_laddr inp_inc.inc_laddr 340 341 #define in6p_faddr inp_inc.inc6_faddr 342 #define in6p_laddr inp_inc.inc6_laddr 343 #define in6p_zoneid inp_inc.inc6_zoneid 344 #define in6p_flowinfo inp_flow 345 346 #define inp_vnet inp_pcbinfo->ipi_vnet 347 348 /* 349 * The range of the generation count, as used in this implementation, is 9e19. 350 * We would have to create 300 billion connections per second for this number 351 * to roll over in a year. This seems sufficiently unlikely that we simply 352 * don't concern ourselves with that possibility. 353 */ 354 355 /* 356 * Interface exported to userland by various protocols which use inpcbs. Hack 357 * alert -- only define if struct xsocket is in scope. 358 * Fields prefixed with "xi_" are unique to this structure, and the rest 359 * match fields in the struct inpcb, to ease coding and porting. 360 * 361 * Legend: 362 * (s) - used by userland utilities in src 363 * (p) - used by utilities in ports 364 * (3) - is known to be used by third party software not in ports 365 * (n) - no known usage 366 */ 367 #ifdef _SYS_SOCKETVAR_H_ 368 struct xinpcb { 369 ksize_t xi_len; /* length of this structure */ 370 struct xsocket xi_socket; /* (s,p) */ 371 struct in_conninfo inp_inc; /* (s,p) */ 372 uint64_t inp_gencnt; /* (s,p) */ 373 kvaddr_t inp_ppcb; /* (s) netstat(1) */ 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 ksize_t xig_len; /* length of this structure */ 395 u_int xig_count; /* number of PCBs at this time */ 396 uint32_t _xig_spare32; 397 inp_gen_t xig_gen; /* generation count at this time */ 398 so_gen_t xig_sogen; /* socket generation count this time */ 399 uint64_t _xig_spare64[4]; 400 } __aligned(8); 401 #ifdef _KERNEL 402 void in_pcbtoxinpcb(const struct inpcb *, struct xinpcb *); 403 #endif 404 #endif /* _SYS_SOCKETVAR_H_ */ 405 406 struct inpcbport { 407 struct epoch_context phd_epoch_ctx; 408 CK_LIST_ENTRY(inpcbport) phd_hash; 409 struct inpcbhead phd_pcblist; 410 u_short phd_port; 411 }; 412 413 struct in_pcblist { 414 int il_count; 415 struct epoch_context il_epoch_ctx; 416 struct inpcbinfo *il_pcbinfo; 417 struct inpcb *il_inp_list[0]; 418 }; 419 420 /*- 421 * Global data structure for each high-level protocol (UDP, TCP, ...) in both 422 * IPv4 and IPv6. Holds inpcb lists and information for managing them. 423 * 424 * Each pcbinfo is protected by three locks: ipi_lock, ipi_hash_lock and 425 * ipi_list_lock: 426 * - ipi_lock covering the global pcb list stability during loop iteration, 427 * - ipi_hash_lock covering the hashed lookup tables, 428 * - ipi_list_lock covering mutable global fields (such as the global 429 * pcb list) 430 * 431 * The lock order is: 432 * 433 * ipi_lock (before) 434 * inpcb locks (before) 435 * ipi_list locks (before) 436 * {ipi_hash_lock, pcbgroup locks} 437 * 438 * Locking key: 439 * 440 * (c) Constant or nearly constant after initialisation 441 * (e) - Protected by the net_epoch_prempt epoch 442 * (g) Locked by ipi_lock 443 * (l) Locked by ipi_list_lock 444 * (h) Read using either net_epoch_preempt or inpcb lock; write requires both ipi_hash_lock and inpcb lock 445 * (p) Protected by one or more pcbgroup locks 446 * (x) Synchronisation properties poorly defined 447 */ 448 struct inpcbinfo { 449 /* 450 * Global lock protecting inpcb list modification 451 */ 452 struct mtx ipi_lock; 453 454 /* 455 * Global list of inpcbs on the protocol. 456 */ 457 struct inpcbhead *ipi_listhead; /* [r](e) [w](g/l) */ 458 u_int ipi_count; /* (l) */ 459 460 /* 461 * Generation count -- incremented each time a connection is allocated 462 * or freed. 463 */ 464 u_quad_t ipi_gencnt; /* (l) */ 465 466 /* 467 * Fields associated with port lookup and allocation. 468 */ 469 u_short ipi_lastport; /* (x) */ 470 u_short ipi_lastlow; /* (x) */ 471 u_short ipi_lasthi; /* (x) */ 472 473 /* 474 * UMA zone from which inpcbs are allocated for this protocol. 475 */ 476 struct uma_zone *ipi_zone; /* (c) */ 477 478 /* 479 * Connection groups associated with this protocol. These fields are 480 * constant, but pcbgroup structures themselves are protected by 481 * per-pcbgroup locks. 482 */ 483 struct inpcbgroup *ipi_pcbgroups; /* (c) */ 484 u_int ipi_npcbgroups; /* (c) */ 485 u_int ipi_hashfields; /* (c) */ 486 487 /* 488 * Global lock protecting modification non-pcbgroup hash lookup tables. 489 */ 490 struct mtx ipi_hash_lock; 491 492 /* 493 * Global hash of inpcbs, hashed by local and foreign addresses and 494 * port numbers. 495 */ 496 struct inpcbhead *ipi_hashbase; /* (h) */ 497 u_long ipi_hashmask; /* (h) */ 498 499 /* 500 * Global hash of inpcbs, hashed by only local port number. 501 */ 502 struct inpcbporthead *ipi_porthashbase; /* (h) */ 503 u_long ipi_porthashmask; /* (h) */ 504 505 /* 506 * List of wildcard inpcbs for use with pcbgroups. In the past, was 507 * per-pcbgroup but is now global. All pcbgroup locks must be held 508 * to modify the list, so any is sufficient to read it. 509 */ 510 struct inpcbhead *ipi_wildbase; /* (p) */ 511 u_long ipi_wildmask; /* (p) */ 512 513 /* 514 * Load balance groups used for the SO_REUSEPORT_LB option, 515 * hashed by local port. 516 */ 517 struct inpcblbgrouphead *ipi_lbgrouphashbase; /* (h) */ 518 u_long ipi_lbgrouphashmask; /* (h) */ 519 520 /* 521 * Pointer to network stack instance 522 */ 523 struct vnet *ipi_vnet; /* (c) */ 524 525 /* 526 * general use 2 527 */ 528 void *ipi_pspare[2]; 529 530 /* 531 * Global lock protecting global inpcb list, inpcb count, etc. 532 */ 533 struct rwlock ipi_list_lock; 534 }; 535 536 #ifdef _KERNEL 537 /* 538 * Connection groups hold sets of connections that have similar CPU/thread 539 * affinity. Each connection belongs to exactly one connection group. 540 */ 541 struct inpcbgroup { 542 /* 543 * Per-connection group hash of inpcbs, hashed by local and foreign 544 * addresses and port numbers. 545 */ 546 struct inpcbhead *ipg_hashbase; /* (c) */ 547 u_long ipg_hashmask; /* (c) */ 548 549 /* 550 * Notional affinity of this pcbgroup. 551 */ 552 u_int ipg_cpu; /* (p) */ 553 554 /* 555 * Per-connection group lock, not to be confused with ipi_lock. 556 * Protects the hash table hung off the group, but also the global 557 * wildcard list in inpcbinfo. 558 */ 559 struct mtx ipg_lock; 560 } __aligned(CACHE_LINE_SIZE); 561 562 /* 563 * Load balance groups used for the SO_REUSEPORT_LB socket option. Each group 564 * (or unique address:port combination) can be re-used at most 565 * INPCBLBGROUP_SIZMAX (256) times. The inpcbs are stored in il_inp which 566 * is dynamically resized as processes bind/unbind to that specific group. 567 */ 568 struct inpcblbgroup { 569 LIST_ENTRY(inpcblbgroup) il_list; 570 uint16_t il_lport; /* (c) */ 571 u_char il_vflag; /* (c) */ 572 u_char il_pad; 573 uint32_t il_pad2; 574 union in_dependaddr il_dependladdr; /* (c) */ 575 #define il_laddr il_dependladdr.id46_addr.ia46_addr4 576 #define il6_laddr il_dependladdr.id6_addr 577 uint32_t il_inpsiz; /* max count in il_inp[] (h) */ 578 uint32_t il_inpcnt; /* cur count in il_inp[] (h) */ 579 struct inpcb *il_inp[]; /* (h) */ 580 }; 581 LIST_HEAD(inpcblbgrouphead, inpcblbgroup); 582 583 #define INP_LOCK_INIT(inp, d, t) \ 584 rw_init_flags(&(inp)->inp_lock, (t), RW_RECURSE | RW_DUPOK) 585 #define INP_LOCK_DESTROY(inp) rw_destroy(&(inp)->inp_lock) 586 #define INP_RLOCK(inp) rw_rlock(&(inp)->inp_lock) 587 #define INP_WLOCK(inp) rw_wlock(&(inp)->inp_lock) 588 #define INP_TRY_RLOCK(inp) rw_try_rlock(&(inp)->inp_lock) 589 #define INP_TRY_WLOCK(inp) rw_try_wlock(&(inp)->inp_lock) 590 #define INP_RUNLOCK(inp) rw_runlock(&(inp)->inp_lock) 591 #define INP_WUNLOCK(inp) rw_wunlock(&(inp)->inp_lock) 592 #define INP_TRY_UPGRADE(inp) rw_try_upgrade(&(inp)->inp_lock) 593 #define INP_DOWNGRADE(inp) rw_downgrade(&(inp)->inp_lock) 594 #define INP_WLOCKED(inp) rw_wowned(&(inp)->inp_lock) 595 #define INP_LOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_LOCKED) 596 #define INP_RLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_RLOCKED) 597 #define INP_WLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_WLOCKED) 598 #define INP_UNLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_UNLOCKED) 599 600 /* 601 * These locking functions are for inpcb consumers outside of sys/netinet, 602 * more specifically, they were added for the benefit of TOE drivers. The 603 * macros are reserved for use by the stack. 604 */ 605 void inp_wlock(struct inpcb *); 606 void inp_wunlock(struct inpcb *); 607 void inp_rlock(struct inpcb *); 608 void inp_runlock(struct inpcb *); 609 610 #ifdef INVARIANT_SUPPORT 611 void inp_lock_assert(struct inpcb *); 612 void inp_unlock_assert(struct inpcb *); 613 #else 614 #define inp_lock_assert(inp) do {} while (0) 615 #define inp_unlock_assert(inp) do {} while (0) 616 #endif 617 618 void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg); 619 int inp_ip_tos_get(const struct inpcb *inp); 620 void inp_ip_tos_set(struct inpcb *inp, int val); 621 struct socket * 622 inp_inpcbtosocket(struct inpcb *inp); 623 struct tcpcb * 624 inp_inpcbtotcpcb(struct inpcb *inp); 625 void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 626 uint32_t *faddr, uint16_t *fp); 627 int inp_so_options(const struct inpcb *inp); 628 629 #endif /* _KERNEL */ 630 631 #define INP_INFO_LOCK_INIT(ipi, d) \ 632 mtx_init(&(ipi)->ipi_lock, (d), NULL, MTX_DEF| MTX_RECURSE) 633 #define INP_INFO_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_lock) 634 #define INP_INFO_RLOCK_ET(ipi, et) NET_EPOCH_ENTER_ET((et)) 635 #define INP_INFO_WLOCK(ipi) mtx_lock(&(ipi)->ipi_lock) 636 #define INP_INFO_TRY_WLOCK(ipi) mtx_trylock(&(ipi)->ipi_lock) 637 #define INP_INFO_WLOCKED(ipi) mtx_owned(&(ipi)->ipi_lock) 638 #define INP_INFO_RUNLOCK_ET(ipi, et) NET_EPOCH_EXIT_ET((et)) 639 #define INP_INFO_RUNLOCK_TP(ipi, tp) NET_EPOCH_EXIT_ET(*(tp)->t_inpcb->inp_et) 640 #define INP_INFO_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_lock) 641 #define INP_INFO_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_lock)) 642 #define INP_INFO_RLOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt)) 643 #define INP_INFO_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_lock, MA_OWNED) 644 #define INP_INFO_UNLOCK_ASSERT(ipi) MPASS(!in_epoch(net_epoch_preempt) && !mtx_owned(&(ipi)->ipi_lock)) 645 646 #define INP_LIST_LOCK_INIT(ipi, d) \ 647 rw_init_flags(&(ipi)->ipi_list_lock, (d), 0) 648 #define INP_LIST_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_list_lock) 649 #define INP_LIST_RLOCK(ipi) rw_rlock(&(ipi)->ipi_list_lock) 650 #define INP_LIST_WLOCK(ipi) rw_wlock(&(ipi)->ipi_list_lock) 651 #define INP_LIST_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_list_lock) 652 #define INP_LIST_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_list_lock) 653 #define INP_LIST_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_list_lock) 654 #define INP_LIST_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_list_lock) 655 #define INP_LIST_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_list_lock) 656 #define INP_LIST_LOCK_ASSERT(ipi) \ 657 rw_assert(&(ipi)->ipi_list_lock, RA_LOCKED) 658 #define INP_LIST_RLOCK_ASSERT(ipi) \ 659 rw_assert(&(ipi)->ipi_list_lock, RA_RLOCKED) 660 #define INP_LIST_WLOCK_ASSERT(ipi) \ 661 rw_assert(&(ipi)->ipi_list_lock, RA_WLOCKED) 662 #define INP_LIST_UNLOCK_ASSERT(ipi) \ 663 rw_assert(&(ipi)->ipi_list_lock, RA_UNLOCKED) 664 665 #define INP_HASH_LOCK_INIT(ipi, d) mtx_init(&(ipi)->ipi_hash_lock, (d), NULL, MTX_DEF) 666 #define INP_HASH_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_hash_lock) 667 #define INP_HASH_RLOCK(ipi) struct epoch_tracker inp_hash_et; epoch_enter_preempt(net_epoch_preempt, &inp_hash_et) 668 #define INP_HASH_RLOCK_ET(ipi, et) epoch_enter_preempt(net_epoch_preempt, &(et)) 669 #define INP_HASH_WLOCK(ipi) mtx_lock(&(ipi)->ipi_hash_lock) 670 #define INP_HASH_RUNLOCK(ipi) NET_EPOCH_EXIT_ET(inp_hash_et) 671 #define INP_HASH_RUNLOCK_ET(ipi, et) NET_EPOCH_EXIT_ET((et)) 672 #define INP_HASH_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_hash_lock) 673 #define INP_HASH_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_hash_lock)) 674 #define INP_HASH_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_hash_lock, MA_OWNED); 675 676 #define INP_GROUP_LOCK_INIT(ipg, d) mtx_init(&(ipg)->ipg_lock, (d), NULL, \ 677 MTX_DEF | MTX_DUPOK) 678 #define INP_GROUP_LOCK_DESTROY(ipg) mtx_destroy(&(ipg)->ipg_lock) 679 680 #define INP_GROUP_LOCK(ipg) mtx_lock(&(ipg)->ipg_lock) 681 #define INP_GROUP_LOCK_ASSERT(ipg) mtx_assert(&(ipg)->ipg_lock, MA_OWNED) 682 #define INP_GROUP_UNLOCK(ipg) mtx_unlock(&(ipg)->ipg_lock) 683 684 #define INP_PCBHASH(faddr, lport, fport, mask) \ 685 (((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask)) 686 #define INP_PCBPORTHASH(lport, mask) \ 687 (ntohs((lport)) & (mask)) 688 #define INP_PCBLBGROUP_PORTHASH(lport, mask) \ 689 (ntohs((lport)) & (mask)) 690 #define INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) \ 691 ((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) 692 #define INP6_PCBHASHKEY(faddr) ((faddr)->s6_addr32[3]) 693 694 /* 695 * Flags for inp_vflags -- historically version flags only 696 */ 697 #define INP_IPV4 0x1 698 #define INP_IPV6 0x2 699 #define INP_IPV6PROTO 0x4 /* opened under IPv6 protocol */ 700 701 /* 702 * Flags for inp_flags. 703 */ 704 #define INP_RECVOPTS 0x00000001 /* receive incoming IP options */ 705 #define INP_RECVRETOPTS 0x00000002 /* receive IP options for reply */ 706 #define INP_RECVDSTADDR 0x00000004 /* receive IP dst address */ 707 #define INP_HDRINCL 0x00000008 /* user supplies entire IP header */ 708 #define INP_HIGHPORT 0x00000010 /* user wants "high" port binding */ 709 #define INP_LOWPORT 0x00000020 /* user wants "low" port binding */ 710 #define INP_ANONPORT 0x00000040 /* port chosen for user */ 711 #define INP_RECVIF 0x00000080 /* receive incoming interface */ 712 #define INP_MTUDISC 0x00000100 /* user can do MTU discovery */ 713 /* 0x000200 unused: was INP_FAITH */ 714 #define INP_RECVTTL 0x00000400 /* receive incoming IP TTL */ 715 #define INP_DONTFRAG 0x00000800 /* don't fragment packet */ 716 #define INP_BINDANY 0x00001000 /* allow bind to any address */ 717 #define INP_INHASHLIST 0x00002000 /* in_pcbinshash() has been called */ 718 #define INP_RECVTOS 0x00004000 /* receive incoming IP TOS */ 719 #define IN6P_IPV6_V6ONLY 0x00008000 /* restrict AF_INET6 socket for v6 */ 720 #define IN6P_PKTINFO 0x00010000 /* receive IP6 dst and I/F */ 721 #define IN6P_HOPLIMIT 0x00020000 /* receive hoplimit */ 722 #define IN6P_HOPOPTS 0x00040000 /* receive hop-by-hop options */ 723 #define IN6P_DSTOPTS 0x00080000 /* receive dst options after rthdr */ 724 #define IN6P_RTHDR 0x00100000 /* receive routing header */ 725 #define IN6P_RTHDRDSTOPTS 0x00200000 /* receive dstoptions before rthdr */ 726 #define IN6P_TCLASS 0x00400000 /* receive traffic class value */ 727 #define IN6P_AUTOFLOWLABEL 0x00800000 /* attach flowlabel automatically */ 728 #define INP_TIMEWAIT 0x01000000 /* in TIMEWAIT, ppcb is tcptw */ 729 #define INP_ONESBCAST 0x02000000 /* send all-ones broadcast */ 730 #define INP_DROPPED 0x04000000 /* protocol drop flag */ 731 #define INP_SOCKREF 0x08000000 /* strong socket reference */ 732 #define INP_RESERVED_0 0x10000000 /* reserved field */ 733 #define INP_RESERVED_1 0x20000000 /* reserved field */ 734 #define IN6P_RFC2292 0x40000000 /* used RFC2292 API on the socket */ 735 #define IN6P_MTU 0x80000000 /* receive path MTU */ 736 737 #define INP_CONTROLOPTS (INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\ 738 INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\ 739 IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\ 740 IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\ 741 IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\ 742 IN6P_MTU) 743 744 /* 745 * Flags for inp_flags2. 746 */ 747 #define INP_LLE_VALID 0x00000001 /* cached lle is valid */ 748 #define INP_RT_VALID 0x00000002 /* cached rtentry is valid */ 749 #define INP_PCBGROUPWILD 0x00000004 /* in pcbgroup wildcard list */ 750 #define INP_REUSEPORT 0x00000008 /* SO_REUSEPORT option is set */ 751 #define INP_FREED 0x00000010 /* inp itself is not valid */ 752 #define INP_REUSEADDR 0x00000020 /* SO_REUSEADDR option is set */ 753 #define INP_BINDMULTI 0x00000040 /* IP_BINDMULTI option is set */ 754 #define INP_RSS_BUCKET_SET 0x00000080 /* IP_RSS_LISTEN_BUCKET is set */ 755 #define INP_RECVFLOWID 0x00000100 /* populate recv datagram with flow info */ 756 #define INP_RECVRSSBUCKETID 0x00000200 /* populate recv datagram with bucket id */ 757 #define INP_RATE_LIMIT_CHANGED 0x00000400 /* rate limit needs attention */ 758 #define INP_ORIGDSTADDR 0x00000800 /* receive IP dst address/port */ 759 #define INP_CANNOT_DO_ECN 0x00001000 /* The stack does not do ECN */ 760 #define INP_REUSEPORT_LB 0x00002000 /* SO_REUSEPORT_LB option is set */ 761 762 /* 763 * Flags passed to in_pcblookup*() functions. 764 */ 765 #define INPLOOKUP_WILDCARD 0x00000001 /* Allow wildcard sockets. */ 766 #define INPLOOKUP_RLOCKPCB 0x00000002 /* Return inpcb read-locked. */ 767 #define INPLOOKUP_WLOCKPCB 0x00000004 /* Return inpcb write-locked. */ 768 769 #define INPLOOKUP_MASK (INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB | \ 770 INPLOOKUP_WLOCKPCB) 771 772 #define sotoinpcb(so) ((struct inpcb *)(so)->so_pcb) 773 #define sotoin6pcb(so) sotoinpcb(so) /* for KAME src sync over BSD*'s */ 774 775 #define INP_SOCKAF(so) so->so_proto->pr_domain->dom_family 776 777 #define INP_CHECK_SOCKAF(so, af) (INP_SOCKAF(so) == af) 778 779 /* 780 * Constants for pcbinfo.ipi_hashfields. 781 */ 782 #define IPI_HASHFIELDS_NONE 0 783 #define IPI_HASHFIELDS_2TUPLE 1 784 #define IPI_HASHFIELDS_4TUPLE 2 785 786 #ifdef _KERNEL 787 VNET_DECLARE(int, ipport_reservedhigh); 788 VNET_DECLARE(int, ipport_reservedlow); 789 VNET_DECLARE(int, ipport_lowfirstauto); 790 VNET_DECLARE(int, ipport_lowlastauto); 791 VNET_DECLARE(int, ipport_firstauto); 792 VNET_DECLARE(int, ipport_lastauto); 793 VNET_DECLARE(int, ipport_hifirstauto); 794 VNET_DECLARE(int, ipport_hilastauto); 795 VNET_DECLARE(int, ipport_randomized); 796 VNET_DECLARE(int, ipport_randomcps); 797 VNET_DECLARE(int, ipport_randomtime); 798 VNET_DECLARE(int, ipport_stoprandom); 799 VNET_DECLARE(int, ipport_tcpallocs); 800 801 #define V_ipport_reservedhigh VNET(ipport_reservedhigh) 802 #define V_ipport_reservedlow VNET(ipport_reservedlow) 803 #define V_ipport_lowfirstauto VNET(ipport_lowfirstauto) 804 #define V_ipport_lowlastauto VNET(ipport_lowlastauto) 805 #define V_ipport_firstauto VNET(ipport_firstauto) 806 #define V_ipport_lastauto VNET(ipport_lastauto) 807 #define V_ipport_hifirstauto VNET(ipport_hifirstauto) 808 #define V_ipport_hilastauto VNET(ipport_hilastauto) 809 #define V_ipport_randomized VNET(ipport_randomized) 810 #define V_ipport_randomcps VNET(ipport_randomcps) 811 #define V_ipport_randomtime VNET(ipport_randomtime) 812 #define V_ipport_stoprandom VNET(ipport_stoprandom) 813 #define V_ipport_tcpallocs VNET(ipport_tcpallocs) 814 815 void in_pcbinfo_destroy(struct inpcbinfo *); 816 void in_pcbinfo_init(struct inpcbinfo *, const char *, struct inpcbhead *, 817 int, int, char *, uma_init, u_int); 818 819 int in_pcbbind_check_bindmulti(const struct inpcb *ni, 820 const struct inpcb *oi); 821 822 struct inpcbgroup * 823 in_pcbgroup_byhash(struct inpcbinfo *, u_int, uint32_t); 824 struct inpcbgroup * 825 in_pcbgroup_byinpcb(struct inpcb *); 826 struct inpcbgroup * 827 in_pcbgroup_bytuple(struct inpcbinfo *, struct in_addr, u_short, 828 struct in_addr, u_short); 829 void in_pcbgroup_destroy(struct inpcbinfo *); 830 int in_pcbgroup_enabled(struct inpcbinfo *); 831 void in_pcbgroup_init(struct inpcbinfo *, u_int, int); 832 void in_pcbgroup_remove(struct inpcb *); 833 void in_pcbgroup_update(struct inpcb *); 834 void in_pcbgroup_update_mbuf(struct inpcb *, struct mbuf *); 835 836 void in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *); 837 int in_pcballoc(struct socket *, struct inpcbinfo *); 838 int in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *); 839 int in_pcb_lport(struct inpcb *, struct in_addr *, u_short *, 840 struct ucred *, int); 841 int in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *, 842 u_short *, struct ucred *); 843 int in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *); 844 int in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *, 845 struct mbuf *); 846 int in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *, 847 u_short *, in_addr_t *, u_short *, struct inpcb **, 848 struct ucred *); 849 void in_pcbdetach(struct inpcb *); 850 void in_pcbdisconnect(struct inpcb *); 851 void in_pcbdrop(struct inpcb *); 852 void in_pcbfree(struct inpcb *); 853 int in_pcbinshash(struct inpcb *); 854 int in_pcbinshash_nopcbgroup(struct inpcb *); 855 int in_pcbladdr(struct inpcb *, struct in_addr *, struct in_addr *, 856 struct ucred *); 857 struct inpcb * 858 in_pcblookup_local(struct inpcbinfo *, 859 struct in_addr, u_short, int, struct ucred *); 860 struct inpcb * 861 in_pcblookup(struct inpcbinfo *, struct in_addr, u_int, 862 struct in_addr, u_int, int, struct ifnet *); 863 struct inpcb * 864 in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int, 865 struct in_addr, u_int, int, struct ifnet *, struct mbuf *); 866 void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr, 867 int, struct inpcb *(*)(struct inpcb *, int)); 868 void in_pcbref(struct inpcb *); 869 void in_pcbrehash(struct inpcb *); 870 void in_pcbrehash_mbuf(struct inpcb *, struct mbuf *); 871 int in_pcbrele(struct inpcb *); 872 int in_pcbrele_rlocked(struct inpcb *); 873 int in_pcbrele_wlocked(struct inpcb *); 874 void in_pcblist_rele_rlocked(epoch_context_t ctx); 875 void in_losing(struct inpcb *); 876 void in_pcbsetsolabel(struct socket *so); 877 int in_getpeeraddr(struct socket *so, struct sockaddr **nam); 878 int in_getsockaddr(struct socket *so, struct sockaddr **nam); 879 struct sockaddr * 880 in_sockaddr(in_port_t port, struct in_addr *addr); 881 void in_pcbsosetlabel(struct socket *so); 882 #ifdef RATELIMIT 883 int in_pcbattach_txrtlmt(struct inpcb *, struct ifnet *, uint32_t, uint32_t, uint32_t); 884 void in_pcbdetach_txrtlmt(struct inpcb *); 885 int in_pcbmodify_txrtlmt(struct inpcb *, uint32_t); 886 int in_pcbquery_txrtlmt(struct inpcb *, uint32_t *); 887 int in_pcbquery_txrlevel(struct inpcb *, uint32_t *); 888 void in_pcboutput_txrtlmt(struct inpcb *, struct ifnet *, struct mbuf *); 889 void in_pcboutput_eagain(struct inpcb *); 890 #endif 891 #endif /* _KERNEL */ 892 893 #endif /* !_NETINET_IN_PCB_H_ */ 894