1/* 2 * Copyright (c) 1998-2013 Apple Inc. All rights reserved. 3 * 4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ 5 * 6 * This file contains Original Code and/or Modifications of Original Code 7 * as defined in and that are subject to the Apple Public Source License 8 * Version 2.0 (the 'License'). You may not use this file except in 9 * compliance with the License. The rights granted to you under the License 10 * may not be used to create, or enable the creation or redistribution of, 11 * unlawful or unlicensed copies of an Apple operating system, or to 12 * circumvent, violate, or enable the circumvention or violation of, any 13 * terms of an Apple operating system software license agreement. 14 * 15 * Please obtain a copy of the License at 16 * http://www.opensource.apple.com/apsl/ and read it before using this file. 17 * 18 * The Original Code and all software distributed under the License are 19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 23 * Please see the License for the specific language governing rights and 24 * limitations under the License. 25 * 26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ 27 */ 28/* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */ 29/* 30 * Copyright (c) 1982, 1986, 1988, 1990, 1993 31 * The Regents of the University of California. All rights reserved. 32 * 33 * Redistribution and use in source and binary forms, with or without 34 * modification, are permitted provided that the following conditions 35 * are met: 36 * 1. Redistributions of source code must retain the above copyright 37 * notice, this list of conditions and the following disclaimer. 38 * 2. Redistributions in binary form must reproduce the above copyright 39 * notice, this list of conditions and the following disclaimer in the 40 * documentation and/or other materials provided with the distribution. 41 * 3. All advertising materials mentioning features or use of this software 42 * must display the following acknowledgement: 43 * This product includes software developed by the University of 44 * California, Berkeley and its contributors. 45 * 4. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 62 */ 63/* 64 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce 65 * support for mandatory and extensible security protections. This notice 66 * is included in support of clause 2.2 (b) of the Apple Public License, 67 * Version 2.0. 68 */ 69 70#include <sys/param.h> 71#include <sys/systm.h> 72#include <sys/domain.h> 73#include <sys/kernel.h> 74#include <sys/proc_internal.h> 75#include <sys/kauth.h> 76#include <sys/malloc.h> 77#include <sys/mbuf.h> 78#include <sys/mcache.h> 79#include <sys/protosw.h> 80#include <sys/stat.h> 81#include <sys/socket.h> 82#include <sys/socketvar.h> 83#include <sys/signalvar.h> 84#include <sys/sysctl.h> 85#include <sys/syslog.h> 86#include <sys/ev.h> 87#include <kern/locks.h> 88#include <net/route.h> 89#include <netinet/in.h> 90#include <netinet/in_pcb.h> 91#include <sys/kdebug.h> 92#include <libkern/OSAtomic.h> 93 94#if CONFIG_MACF 95#include <security/mac_framework.h> 96#endif 97 98#include <mach/vm_param.h> 99 100/* TODO: this should be in a header file somewhere */ 101extern void postevent(struct socket *, struct sockbuf *, int); 102 103#define DBG_FNC_SBDROP NETDBG_CODE(DBG_NETSOCK, 4) 104#define DBG_FNC_SBAPPEND NETDBG_CODE(DBG_NETSOCK, 5) 105 106static inline void sbcompress(struct sockbuf *, struct mbuf *, struct mbuf *); 107static struct socket *sonewconn_internal(struct socket *, int); 108static int sbappendaddr_internal(struct sockbuf *, struct sockaddr *, 109 struct mbuf *, struct mbuf *); 110static int sbappendcontrol_internal(struct sockbuf *, struct mbuf *, 111 struct mbuf *); 112static void soevent_ifdenied(struct socket *); 113 114/* 115 * Primitive routines for operating on sockets and socket buffers 116 */ 117static int soqlimitcompat = 1; 118static int soqlencomp = 0; 119 120/* 121 * Based on the number of mbuf clusters configured, high_sb_max and sb_max can 122 * get scaled up or down to suit that memory configuration. high_sb_max is a 123 * higher limit on sb_max that is checked when sb_max gets set through sysctl. 124 */ 125 126u_int32_t sb_max = SB_MAX; /* XXX should be static */ 127u_int32_t high_sb_max = SB_MAX; 128 129static u_int32_t sb_efficiency = 8; /* parameter for sbreserve() */ 130__private_extern__ int32_t total_sbmb_cnt = 0; 131 132/* Control whether to throttle sockets eligible to be throttled */ 133__private_extern__ u_int32_t net_io_policy_throttled = 0; 134static int sysctl_io_policy_throttled SYSCTL_HANDLER_ARGS; 135 136u_int32_t net_io_policy_log = 0; /* log socket policy changes */ 137#if CONFIG_PROC_UUID_POLICY 138u_int32_t net_io_policy_uuid = 1; /* enable UUID socket policy */ 139#endif /* CONFIG_PROC_UUID_POLICY */ 140 141/* 142 * Procedures to manipulate state flags of socket 143 * and do appropriate wakeups. Normal sequence from the 144 * active (originating) side is that soisconnecting() is 145 * called during processing of connect() call, 146 * resulting in an eventual call to soisconnected() if/when the 147 * connection is established. When the connection is torn down 148 * soisdisconnecting() is called during processing of disconnect() call, 149 * and soisdisconnected() is called when the connection to the peer 150 * is totally severed. The semantics of these routines are such that 151 * connectionless protocols can call soisconnected() and soisdisconnected() 152 * only, bypassing the in-progress calls when setting up a ``connection'' 153 * takes no time. 154 * 155 * From the passive side, a socket is created with 156 * two queues of sockets: so_incomp for connections in progress 157 * and so_comp for connections already made and awaiting user acceptance. 158 * As a protocol is preparing incoming connections, it creates a socket 159 * structure queued on so_incomp by calling sonewconn(). When the connection 160 * is established, soisconnected() is called, and transfers the 161 * socket structure to so_comp, making it available to accept(). 162 * 163 * If a socket is closed with sockets on either 164 * so_incomp or so_comp, these sockets are dropped. 165 * 166 * If higher level protocols are implemented in 167 * the kernel, the wakeups done here will sometimes 168 * cause software-interrupt process scheduling. 169 */ 170void 171soisconnecting(struct socket *so) 172{ 173 174 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 175 so->so_state |= SS_ISCONNECTING; 176 177 sflt_notify(so, sock_evt_connecting, NULL); 178} 179 180void 181soisconnected(struct socket *so) 182{ 183 struct socket *head = so->so_head; 184 185 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 186 so->so_state |= SS_ISCONNECTED; 187 188 sflt_notify(so, sock_evt_connected, NULL); 189 190 if (head && (so->so_state & SS_INCOMP)) { 191 so->so_state &= ~SS_INCOMP; 192 so->so_state |= SS_COMP; 193 if (head->so_proto->pr_getlock != NULL) { 194 socket_unlock(so, 0); 195 socket_lock(head, 1); 196 } 197 postevent(head, 0, EV_RCONN); 198 TAILQ_REMOVE(&head->so_incomp, so, so_list); 199 head->so_incqlen--; 200 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 201 sorwakeup(head); 202 wakeup_one((caddr_t)&head->so_timeo); 203 if (head->so_proto->pr_getlock != NULL) { 204 socket_unlock(head, 1); 205 socket_lock(so, 0); 206 } 207 } else { 208 postevent(so, 0, EV_WCONN); 209 wakeup((caddr_t)&so->so_timeo); 210 sorwakeup(so); 211 sowwakeup(so); 212 soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_CONNECTED | 213 SO_FILT_HINT_CONNINFO_UPDATED); 214 } 215} 216 217void 218soisdisconnecting(struct socket *so) 219{ 220 so->so_state &= ~SS_ISCONNECTING; 221 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 222 soevent(so, SO_FILT_HINT_LOCKED); 223 sflt_notify(so, sock_evt_disconnecting, NULL); 224 wakeup((caddr_t)&so->so_timeo); 225 sowwakeup(so); 226 sorwakeup(so); 227} 228 229void 230soisdisconnected(struct socket *so) 231{ 232 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 233 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 234 soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_DISCONNECTED | 235 SO_FILT_HINT_CONNINFO_UPDATED); 236 sflt_notify(so, sock_evt_disconnected, NULL); 237 wakeup((caddr_t)&so->so_timeo); 238 sowwakeup(so); 239 sorwakeup(so); 240} 241 242/* 243 * This function will issue a wakeup like soisdisconnected but it will not 244 * notify the socket filters. This will avoid unlocking the socket 245 * in the midst of closing it. 246 */ 247void 248sodisconnectwakeup(struct socket *so) 249{ 250 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 251 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 252 soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_DISCONNECTED | 253 SO_FILT_HINT_CONNINFO_UPDATED); 254 wakeup((caddr_t)&so->so_timeo); 255 sowwakeup(so); 256 sorwakeup(so); 257} 258 259/* 260 * When an attempt at a new connection is noted on a socket 261 * which accepts connections, sonewconn is called. If the 262 * connection is possible (subject to space constraints, etc.) 263 * then we allocate a new structure, propoerly linked into the 264 * data structure of the original socket, and return this. 265 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 266 */ 267static struct socket * 268sonewconn_internal(struct socket *head, int connstatus) 269{ 270 int so_qlen, error = 0; 271 struct socket *so; 272 lck_mtx_t *mutex_held; 273 274 if (head->so_proto->pr_getlock != NULL) 275 mutex_held = (*head->so_proto->pr_getlock)(head, 0); 276 else 277 mutex_held = head->so_proto->pr_domain->dom_mtx; 278 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 279 280 if (!soqlencomp) { 281 /* 282 * This is the default case; so_qlen represents the 283 * sum of both incomplete and completed queues. 284 */ 285 so_qlen = head->so_qlen; 286 } else { 287 /* 288 * When kern.ipc.soqlencomp is set to 1, so_qlen 289 * represents only the completed queue. Since we 290 * cannot let the incomplete queue goes unbounded 291 * (in case of SYN flood), we cap the incomplete 292 * queue length to at most somaxconn, and use that 293 * as so_qlen so that we fail immediately below. 294 */ 295 so_qlen = head->so_qlen - head->so_incqlen; 296 if (head->so_incqlen > somaxconn) 297 so_qlen = somaxconn; 298 } 299 300 if (so_qlen >= 301 (soqlimitcompat ? head->so_qlimit : (3 * head->so_qlimit / 2))) 302 return ((struct socket *)0); 303 so = soalloc(1, SOCK_DOM(head), head->so_type); 304 if (so == NULL) 305 return ((struct socket *)0); 306 /* check if head was closed during the soalloc */ 307 if (head->so_proto == NULL) { 308 sodealloc(so); 309 return ((struct socket *)0); 310 } 311 312 so->so_type = head->so_type; 313 so->so_options = head->so_options &~ SO_ACCEPTCONN; 314 so->so_linger = head->so_linger; 315 so->so_state = head->so_state | SS_NOFDREF; 316 so->so_proto = head->so_proto; 317 so->so_timeo = head->so_timeo; 318 so->so_pgid = head->so_pgid; 319 kauth_cred_ref(head->so_cred); 320 so->so_cred = head->so_cred; 321 so->last_pid = head->last_pid; 322 so->last_upid = head->last_upid; 323 memcpy(so->last_uuid, head->last_uuid, sizeof (so->last_uuid)); 324 if (head->so_flags & SOF_DELEGATED) { 325 so->e_pid = head->e_pid; 326 so->e_upid = head->e_upid; 327 memcpy(so->e_uuid, head->e_uuid, sizeof (so->e_uuid)); 328 } 329 /* inherit socket options stored in so_flags */ 330 so->so_flags = head->so_flags & 331 (SOF_NOSIGPIPE | SOF_NOADDRAVAIL | SOF_REUSESHAREUID | 332 SOF_NOTIFYCONFLICT | SOF_BINDRANDOMPORT | SOF_NPX_SETOPTSHUT | 333 SOF_NODEFUNCT | SOF_PRIVILEGED_TRAFFIC_CLASS| SOF_NOTSENT_LOWAT | 334 SOF_USELRO | SOF_DELEGATED); 335 so->so_usecount = 1; 336 so->next_lock_lr = 0; 337 so->next_unlock_lr = 0; 338 339 so->so_rcv.sb_flags |= SB_RECV; /* XXX */ 340 so->so_rcv.sb_so = so->so_snd.sb_so = so; 341 TAILQ_INIT(&so->so_evlist); 342 343#if CONFIG_MACF_SOCKET 344 mac_socket_label_associate_accept(head, so); 345#endif 346 347 /* inherit traffic management properties of listener */ 348 so->so_traffic_mgt_flags = 349 head->so_traffic_mgt_flags & (TRAFFIC_MGT_SO_BACKGROUND); 350 so->so_background_thread = head->so_background_thread; 351 so->so_traffic_class = head->so_traffic_class; 352 353 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 354 sodealloc(so); 355 return ((struct socket *)0); 356 } 357 so->so_rcv.sb_flags |= (head->so_rcv.sb_flags & SB_USRSIZE); 358 so->so_snd.sb_flags |= (head->so_snd.sb_flags & SB_USRSIZE); 359 360 /* 361 * Must be done with head unlocked to avoid deadlock 362 * for protocol with per socket mutexes. 363 */ 364 if (head->so_proto->pr_unlock) 365 socket_unlock(head, 0); 366 if (((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL) != 0) || 367 error) { 368 sodealloc(so); 369 if (head->so_proto->pr_unlock) 370 socket_lock(head, 0); 371 return ((struct socket *)0); 372 } 373 if (head->so_proto->pr_unlock) { 374 socket_lock(head, 0); 375 /* 376 * Radar 7385998 Recheck that the head is still accepting 377 * to avoid race condition when head is getting closed. 378 */ 379 if ((head->so_options & SO_ACCEPTCONN) == 0) { 380 so->so_state &= ~SS_NOFDREF; 381 soclose(so); 382 return ((struct socket *)0); 383 } 384 } 385 386 atomic_add_32(&so->so_proto->pr_domain->dom_refs, 1); 387 388 /* Insert in head appropriate lists */ 389 so->so_head = head; 390 391 /* 392 * Since this socket is going to be inserted into the incomp 393 * queue, it can be picked up by another thread in 394 * tcp_dropdropablreq to get dropped before it is setup.. 395 * To prevent this race, set in-progress flag which can be 396 * cleared later 397 */ 398 so->so_flags |= SOF_INCOMP_INPROGRESS; 399 400 if (connstatus) { 401 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 402 so->so_state |= SS_COMP; 403 } else { 404 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 405 so->so_state |= SS_INCOMP; 406 head->so_incqlen++; 407 } 408 head->so_qlen++; 409 410 /* Attach socket filters for this protocol */ 411 sflt_initsock(so); 412 413 if (connstatus) { 414 so->so_state |= connstatus; 415 sorwakeup(head); 416 wakeup((caddr_t)&head->so_timeo); 417 } 418 return (so); 419} 420 421 422struct socket * 423sonewconn(struct socket *head, int connstatus, const struct sockaddr *from) 424{ 425 int error = sflt_connectin(head, from); 426 if (error) { 427 return (NULL); 428 } 429 430 return (sonewconn_internal(head, connstatus)); 431} 432 433/* 434 * Socantsendmore indicates that no more data will be sent on the 435 * socket; it would normally be applied to a socket when the user 436 * informs the system that no more data is to be sent, by the protocol 437 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 438 * will be received, and will normally be applied to the socket by a 439 * protocol when it detects that the peer will send no more data. 440 * Data queued for reading in the socket may yet be read. 441 */ 442 443void 444socantsendmore(struct socket *so) 445{ 446 so->so_state |= SS_CANTSENDMORE; 447 soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_CANTSENDMORE); 448 sflt_notify(so, sock_evt_cantsendmore, NULL); 449 sowwakeup(so); 450} 451 452void 453socantrcvmore(struct socket *so) 454{ 455 so->so_state |= SS_CANTRCVMORE; 456 soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_CANTRCVMORE); 457 sflt_notify(so, sock_evt_cantrecvmore, NULL); 458 sorwakeup(so); 459} 460 461/* 462 * Wait for data to arrive at/drain from a socket buffer. 463 */ 464int 465sbwait(struct sockbuf *sb) 466{ 467 boolean_t nointr = (sb->sb_flags & SB_NOINTR); 468 void *lr_saved = __builtin_return_address(0); 469 struct socket *so = sb->sb_so; 470 lck_mtx_t *mutex_held; 471 struct timespec ts; 472 int error = 0; 473 474 if (so == NULL) { 475 panic("%s: null so, sb=%p sb_flags=0x%x lr=%p\n", 476 __func__, sb, sb->sb_flags, lr_saved); 477 /* NOTREACHED */ 478 } else if (so->so_usecount < 1) { 479 panic("%s: sb=%p sb_flags=0x%x sb_so=%p usecount=%d lr=%p " 480 "lrh= %s\n", __func__, sb, sb->sb_flags, so, 481 so->so_usecount, lr_saved, solockhistory_nr(so)); 482 /* NOTREACHED */ 483 } 484 485 if (so->so_proto->pr_getlock != NULL) 486 mutex_held = (*so->so_proto->pr_getlock)(so, 0); 487 else 488 mutex_held = so->so_proto->pr_domain->dom_mtx; 489 490 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 491 492 ts.tv_sec = sb->sb_timeo.tv_sec; 493 ts.tv_nsec = sb->sb_timeo.tv_usec * 1000; 494 495 sb->sb_waiters++; 496 VERIFY(sb->sb_waiters != 0); 497 498 error = msleep((caddr_t)&sb->sb_cc, mutex_held, 499 nointr ? PSOCK : PSOCK | PCATCH, 500 nointr ? "sbwait_nointr" : "sbwait", &ts); 501 502 VERIFY(sb->sb_waiters != 0); 503 sb->sb_waiters--; 504 505 if (so->so_usecount < 1) { 506 panic("%s: 2 sb=%p sb_flags=0x%x sb_so=%p usecount=%d lr=%p " 507 "lrh= %s\n", __func__, sb, sb->sb_flags, so, 508 so->so_usecount, lr_saved, solockhistory_nr(so)); 509 /* NOTREACHED */ 510 } 511 512 if ((so->so_state & SS_DRAINING) || (so->so_flags & SOF_DEFUNCT)) { 513 error = EBADF; 514 if (so->so_flags & SOF_DEFUNCT) { 515 SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " 516 "(%d)\n", __func__, proc_selfpid(), 517 (uint64_t)VM_KERNEL_ADDRPERM(so), 518 SOCK_DOM(so), SOCK_TYPE(so), error)); 519 } 520 } 521 522 return (error); 523} 524 525void 526sbwakeup(struct sockbuf *sb) 527{ 528 if (sb->sb_waiters > 0) 529 wakeup((caddr_t)&sb->sb_cc); 530} 531 532/* 533 * Wakeup processes waiting on a socket buffer. 534 * Do asynchronous notification via SIGIO 535 * if the socket has the SS_ASYNC flag set. 536 */ 537void 538sowakeup(struct socket *so, struct sockbuf *sb) 539{ 540 if (so->so_flags & SOF_DEFUNCT) { 541 SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] si 0x%x, " 542 "fl 0x%x [%s]\n", __func__, proc_selfpid(), 543 (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), 544 SOCK_TYPE(so), (uint32_t)sb->sb_sel.si_flags, sb->sb_flags, 545 (sb->sb_flags & SB_RECV) ? "rcv" : "snd")); 546 } 547 548 sb->sb_flags &= ~SB_SEL; 549 selwakeup(&sb->sb_sel); 550 sbwakeup(sb); 551 if (so->so_state & SS_ASYNC) { 552 if (so->so_pgid < 0) 553 gsignal(-so->so_pgid, SIGIO); 554 else if (so->so_pgid > 0) 555 proc_signal(so->so_pgid, SIGIO); 556 } 557 if (sb->sb_flags & SB_KNOTE) { 558 KNOTE(&sb->sb_sel.si_note, SO_FILT_HINT_LOCKED); 559 } 560 if (sb->sb_flags & SB_UPCALL) { 561 void (*sb_upcall)(struct socket *, void *, int); 562 caddr_t sb_upcallarg; 563 564 sb_upcall = sb->sb_upcall; 565 sb_upcallarg = sb->sb_upcallarg; 566 /* Let close know that we're about to do an upcall */ 567 so->so_upcallusecount++; 568 569 socket_unlock(so, 0); 570 (*sb_upcall)(so, sb_upcallarg, M_DONTWAIT); 571 socket_lock(so, 0); 572 573 so->so_upcallusecount--; 574 /* Tell close that it's safe to proceed */ 575 if ((so->so_flags & SOF_CLOSEWAIT) && 576 so->so_upcallusecount == 0) 577 wakeup((caddr_t)&so->so_upcallusecount); 578 } 579} 580 581/* 582 * Socket buffer (struct sockbuf) utility routines. 583 * 584 * Each socket contains two socket buffers: one for sending data and 585 * one for receiving data. Each buffer contains a queue of mbufs, 586 * information about the number of mbufs and amount of data in the 587 * queue, and other fields allowing select() statements and notification 588 * on data availability to be implemented. 589 * 590 * Data stored in a socket buffer is maintained as a list of records. 591 * Each record is a list of mbufs chained together with the m_next 592 * field. Records are chained together with the m_nextpkt field. The upper 593 * level routine soreceive() expects the following conventions to be 594 * observed when placing information in the receive buffer: 595 * 596 * 1. If the protocol requires each message be preceded by the sender's 597 * name, then a record containing that name must be present before 598 * any associated data (mbuf's must be of type MT_SONAME). 599 * 2. If the protocol supports the exchange of ``access rights'' (really 600 * just additional data associated with the message), and there are 601 * ``rights'' to be received, then a record containing this data 602 * should be present (mbuf's must be of type MT_RIGHTS). 603 * 3. If a name or rights record exists, then it must be followed by 604 * a data record, perhaps of zero length. 605 * 606 * Before using a new socket structure it is first necessary to reserve 607 * buffer space to the socket, by calling sbreserve(). This should commit 608 * some of the available buffer space in the system buffer pool for the 609 * socket (currently, it does nothing but enforce limits). The space 610 * should be released by calling sbrelease() when the socket is destroyed. 611 */ 612 613/* 614 * Returns: 0 Success 615 * ENOBUFS 616 */ 617int 618soreserve(struct socket *so, u_int32_t sndcc, u_int32_t rcvcc) 619{ 620 621 if (sbreserve(&so->so_snd, sndcc) == 0) 622 goto bad; 623 else 624 so->so_snd.sb_idealsize = sndcc; 625 626 if (sbreserve(&so->so_rcv, rcvcc) == 0) 627 goto bad2; 628 else 629 so->so_rcv.sb_idealsize = rcvcc; 630 631 if (so->so_rcv.sb_lowat == 0) 632 so->so_rcv.sb_lowat = 1; 633 if (so->so_snd.sb_lowat == 0) 634 so->so_snd.sb_lowat = MCLBYTES; 635 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 636 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 637 return (0); 638bad2: 639 so->so_snd.sb_flags &= ~SB_SEL; 640 selthreadclear(&so->so_snd.sb_sel); 641 sbrelease(&so->so_snd); 642bad: 643 return (ENOBUFS); 644} 645 646/* 647 * Allot mbufs to a sockbuf. 648 * Attempt to scale mbmax so that mbcnt doesn't become limiting 649 * if buffering efficiency is near the normal case. 650 */ 651int 652sbreserve(struct sockbuf *sb, u_int32_t cc) 653{ 654 if ((u_quad_t)cc > (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES)) 655 return (0); 656 sb->sb_hiwat = cc; 657 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 658 if (sb->sb_lowat > sb->sb_hiwat) 659 sb->sb_lowat = sb->sb_hiwat; 660 return (1); 661} 662 663/* 664 * Free mbufs held by a socket, and reserved mbuf space. 665 */ 666/* WARNING needs to do selthreadclear() before calling this */ 667void 668sbrelease(struct sockbuf *sb) 669{ 670 sbflush(sb); 671 sb->sb_hiwat = 0; 672 sb->sb_mbmax = 0; 673} 674 675/* 676 * Routines to add and remove 677 * data from an mbuf queue. 678 * 679 * The routines sbappend() or sbappendrecord() are normally called to 680 * append new mbufs to a socket buffer, after checking that adequate 681 * space is available, comparing the function sbspace() with the amount 682 * of data to be added. sbappendrecord() differs from sbappend() in 683 * that data supplied is treated as the beginning of a new record. 684 * To place a sender's address, optional access rights, and data in a 685 * socket receive buffer, sbappendaddr() should be used. To place 686 * access rights and data in a socket receive buffer, sbappendrights() 687 * should be used. In either case, the new data begins a new record. 688 * Note that unlike sbappend() and sbappendrecord(), these routines check 689 * for the caller that there will be enough space to store the data. 690 * Each fails if there is not enough space, or if it cannot find mbufs 691 * to store additional information in. 692 * 693 * Reliable protocols may use the socket send buffer to hold data 694 * awaiting acknowledgement. Data is normally copied from a socket 695 * send buffer in a protocol with m_copy for output to a peer, 696 * and then removing the data from the socket buffer with sbdrop() 697 * or sbdroprecord() when the data is acknowledged by the peer. 698 */ 699 700/* 701 * Append mbuf chain m to the last record in the 702 * socket buffer sb. The additional space associated 703 * the mbuf chain is recorded in sb. Empty mbufs are 704 * discarded and mbufs are compacted where possible. 705 */ 706int 707sbappend(struct sockbuf *sb, struct mbuf *m) 708{ 709 struct socket *so = sb->sb_so; 710 711 if (m == NULL || (sb->sb_flags & SB_DROP)) { 712 if (m != NULL) 713 m_freem(m); 714 return (0); 715 } 716 717 SBLASTRECORDCHK(sb, "sbappend 1"); 718 719 if (sb->sb_lastrecord != NULL && (sb->sb_mbtail->m_flags & M_EOR)) 720 return (sbappendrecord(sb, m)); 721 722 if (sb->sb_flags & SB_RECV) { 723 int error = sflt_data_in(so, NULL, &m, NULL, 0); 724 SBLASTRECORDCHK(sb, "sbappend 2"); 725 if (error != 0) { 726 if (error != EJUSTRETURN) 727 m_freem(m); 728 return (0); 729 } 730 } 731 732 /* If this is the first record, it's also the last record */ 733 if (sb->sb_lastrecord == NULL) 734 sb->sb_lastrecord = m; 735 736 sbcompress(sb, m, sb->sb_mbtail); 737 SBLASTRECORDCHK(sb, "sbappend 3"); 738 return (1); 739} 740 741/* 742 * Similar to sbappend, except that this is optimized for stream sockets. 743 */ 744int 745sbappendstream(struct sockbuf *sb, struct mbuf *m) 746{ 747 struct socket *so = sb->sb_so; 748 749 if (m == NULL || (sb->sb_flags & SB_DROP)) { 750 if (m != NULL) 751 m_freem(m); 752 return (0); 753 } 754 755 if (m->m_nextpkt != NULL || (sb->sb_mb != sb->sb_lastrecord)) { 756 panic("sbappendstream: nexpkt %p || mb %p != lastrecord %p\n", 757 m->m_nextpkt, sb->sb_mb, sb->sb_lastrecord); 758 /* NOTREACHED */ 759 } 760 761 SBLASTMBUFCHK(sb, __func__); 762 763 if (sb->sb_flags & SB_RECV) { 764 int error = sflt_data_in(so, NULL, &m, NULL, 0); 765 SBLASTRECORDCHK(sb, "sbappendstream 1"); 766 if (error != 0) { 767 if (error != EJUSTRETURN) 768 m_freem(m); 769 return (0); 770 } 771 } 772 773 sbcompress(sb, m, sb->sb_mbtail); 774 sb->sb_lastrecord = sb->sb_mb; 775 SBLASTRECORDCHK(sb, "sbappendstream 2"); 776 return (1); 777} 778 779#ifdef SOCKBUF_DEBUG 780void 781sbcheck(struct sockbuf *sb) 782{ 783 struct mbuf *m; 784 struct mbuf *n = 0; 785 u_int32_t len = 0, mbcnt = 0; 786 lck_mtx_t *mutex_held; 787 788 if (sb->sb_so->so_proto->pr_getlock != NULL) 789 mutex_held = (*sb->sb_so->so_proto->pr_getlock)(sb->sb_so, 0); 790 else 791 mutex_held = sb->sb_so->so_proto->pr_domain->dom_mtx; 792 793 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 794 795 if (sbchecking == 0) 796 return; 797 798 for (m = sb->sb_mb; m; m = n) { 799 n = m->m_nextpkt; 800 for (; m; m = m->m_next) { 801 len += m->m_len; 802 mbcnt += MSIZE; 803 /* XXX pretty sure this is bogus */ 804 if (m->m_flags & M_EXT) 805 mbcnt += m->m_ext.ext_size; 806 } 807 } 808 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 809 panic("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc, 810 mbcnt, sb->sb_mbcnt); 811 } 812} 813#endif 814 815void 816sblastrecordchk(struct sockbuf *sb, const char *where) 817{ 818 struct mbuf *m = sb->sb_mb; 819 820 while (m && m->m_nextpkt) 821 m = m->m_nextpkt; 822 823 if (m != sb->sb_lastrecord) { 824 printf("sblastrecordchk: mb %p lastrecord %p last %p\n", 825 sb->sb_mb, sb->sb_lastrecord, m); 826 printf("packet chain:\n"); 827 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 828 printf("\t%p\n", m); 829 panic("sblastrecordchk from %s", where); 830 } 831} 832 833void 834sblastmbufchk(struct sockbuf *sb, const char *where) 835{ 836 struct mbuf *m = sb->sb_mb; 837 struct mbuf *n; 838 839 while (m && m->m_nextpkt) 840 m = m->m_nextpkt; 841 842 while (m && m->m_next) 843 m = m->m_next; 844 845 if (m != sb->sb_mbtail) { 846 printf("sblastmbufchk: mb %p mbtail %p last %p\n", 847 sb->sb_mb, sb->sb_mbtail, m); 848 printf("packet tree:\n"); 849 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 850 printf("\t"); 851 for (n = m; n != NULL; n = n->m_next) 852 printf("%p ", n); 853 printf("\n"); 854 } 855 panic("sblastmbufchk from %s", where); 856 } 857} 858 859/* 860 * Similar to sbappend, except the mbuf chain begins a new record. 861 */ 862int 863sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 864{ 865 struct mbuf *m; 866 int space = 0; 867 868 if (m0 == NULL || (sb->sb_flags & SB_DROP)) { 869 if (m0 != NULL) 870 m_freem(m0); 871 return (0); 872 } 873 874 for (m = m0; m != NULL; m = m->m_next) 875 space += m->m_len; 876 877 if (space > sbspace(sb) && !(sb->sb_flags & SB_UNIX)) { 878 m_freem(m0); 879 return (0); 880 } 881 882 if (sb->sb_flags & SB_RECV) { 883 int error = sflt_data_in(sb->sb_so, NULL, &m0, NULL, 884 sock_data_filt_flag_record); 885 if (error != 0) { 886 SBLASTRECORDCHK(sb, "sbappendrecord 1"); 887 if (error != EJUSTRETURN) 888 m_freem(m0); 889 return (0); 890 } 891 } 892 893 /* 894 * Note this permits zero length records. 895 */ 896 sballoc(sb, m0); 897 SBLASTRECORDCHK(sb, "sbappendrecord 2"); 898 if (sb->sb_lastrecord != NULL) { 899 sb->sb_lastrecord->m_nextpkt = m0; 900 } else { 901 sb->sb_mb = m0; 902 } 903 sb->sb_lastrecord = m0; 904 sb->sb_mbtail = m0; 905 906 m = m0->m_next; 907 m0->m_next = 0; 908 if (m && (m0->m_flags & M_EOR)) { 909 m0->m_flags &= ~M_EOR; 910 m->m_flags |= M_EOR; 911 } 912 sbcompress(sb, m, m0); 913 SBLASTRECORDCHK(sb, "sbappendrecord 3"); 914 return (1); 915} 916 917/* 918 * As above except that OOB data 919 * is inserted at the beginning of the sockbuf, 920 * but after any other OOB data. 921 */ 922int 923sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 924{ 925 struct mbuf *m; 926 struct mbuf **mp; 927 928 if (m0 == 0) 929 return (0); 930 931 SBLASTRECORDCHK(sb, "sbinsertoob 1"); 932 933 if ((sb->sb_flags & SB_RECV) != 0) { 934 int error = sflt_data_in(sb->sb_so, NULL, &m0, NULL, 935 sock_data_filt_flag_oob); 936 937 SBLASTRECORDCHK(sb, "sbinsertoob 2"); 938 if (error) { 939 if (error != EJUSTRETURN) { 940 m_freem(m0); 941 } 942 return (0); 943 } 944 } 945 946 for (mp = &sb->sb_mb; *mp; mp = &((*mp)->m_nextpkt)) { 947 m = *mp; 948again: 949 switch (m->m_type) { 950 951 case MT_OOBDATA: 952 continue; /* WANT next train */ 953 954 case MT_CONTROL: 955 m = m->m_next; 956 if (m) 957 goto again; /* inspect THIS train further */ 958 } 959 break; 960 } 961 /* 962 * Put the first mbuf on the queue. 963 * Note this permits zero length records. 964 */ 965 sballoc(sb, m0); 966 m0->m_nextpkt = *mp; 967 if (*mp == NULL) { 968 /* m0 is actually the new tail */ 969 sb->sb_lastrecord = m0; 970 } 971 *mp = m0; 972 m = m0->m_next; 973 m0->m_next = 0; 974 if (m && (m0->m_flags & M_EOR)) { 975 m0->m_flags &= ~M_EOR; 976 m->m_flags |= M_EOR; 977 } 978 sbcompress(sb, m, m0); 979 SBLASTRECORDCHK(sb, "sbinsertoob 3"); 980 return (1); 981} 982 983/* 984 * Append address and data, and optionally, control (ancillary) data 985 * to the receive queue of a socket. If present, 986 * m0 must include a packet header with total length. 987 * Returns 0 if no space in sockbuf or insufficient mbufs. 988 * 989 * Returns: 0 No space/out of mbufs 990 * 1 Success 991 */ 992static int 993sbappendaddr_internal(struct sockbuf *sb, struct sockaddr *asa, 994 struct mbuf *m0, struct mbuf *control) 995{ 996 struct mbuf *m, *n, *nlast; 997 int space = asa->sa_len; 998 999 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1000 panic("sbappendaddr"); 1001 1002 if (m0) 1003 space += m0->m_pkthdr.len; 1004 for (n = control; n; n = n->m_next) { 1005 space += n->m_len; 1006 if (n->m_next == 0) /* keep pointer to last control buf */ 1007 break; 1008 } 1009 if (space > sbspace(sb)) 1010 return (0); 1011 if (asa->sa_len > MLEN) 1012 return (0); 1013 MGET(m, M_DONTWAIT, MT_SONAME); 1014 if (m == 0) 1015 return (0); 1016 m->m_len = asa->sa_len; 1017 bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len); 1018 if (n) 1019 n->m_next = m0; /* concatenate data to control */ 1020 else 1021 control = m0; 1022 m->m_next = control; 1023 1024 SBLASTRECORDCHK(sb, "sbappendadddr 1"); 1025 1026 for (n = m; n->m_next != NULL; n = n->m_next) 1027 sballoc(sb, n); 1028 sballoc(sb, n); 1029 nlast = n; 1030 1031 if (sb->sb_lastrecord != NULL) { 1032 sb->sb_lastrecord->m_nextpkt = m; 1033 } else { 1034 sb->sb_mb = m; 1035 } 1036 sb->sb_lastrecord = m; 1037 sb->sb_mbtail = nlast; 1038 1039 SBLASTMBUFCHK(sb, __func__); 1040 SBLASTRECORDCHK(sb, "sbappendadddr 2"); 1041 1042 postevent(0, sb, EV_RWBYTES); 1043 return (1); 1044} 1045 1046/* 1047 * Returns: 0 Error: No space/out of mbufs/etc. 1048 * 1 Success 1049 * 1050 * Imputed: (*error_out) errno for error 1051 * ENOBUFS 1052 * sflt_data_in:??? [whatever a filter author chooses] 1053 */ 1054int 1055sbappendaddr(struct sockbuf *sb, struct sockaddr *asa, struct mbuf *m0, 1056 struct mbuf *control, int *error_out) 1057{ 1058 int result = 0; 1059 boolean_t sb_unix = (sb->sb_flags & SB_UNIX); 1060 1061 if (error_out) 1062 *error_out = 0; 1063 1064 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1065 panic("sbappendaddrorfree"); 1066 1067 if (sb->sb_flags & SB_DROP) { 1068 if (m0 != NULL) 1069 m_freem(m0); 1070 if (control != NULL && !sb_unix) 1071 m_freem(control); 1072 if (error_out != NULL) 1073 *error_out = EINVAL; 1074 return (0); 1075 } 1076 1077 /* Call socket data in filters */ 1078 if ((sb->sb_flags & SB_RECV) != 0) { 1079 int error; 1080 error = sflt_data_in(sb->sb_so, asa, &m0, &control, 0); 1081 SBLASTRECORDCHK(sb, __func__); 1082 if (error) { 1083 if (error != EJUSTRETURN) { 1084 if (m0) 1085 m_freem(m0); 1086 if (control != NULL && !sb_unix) 1087 m_freem(control); 1088 if (error_out) 1089 *error_out = error; 1090 } 1091 return (0); 1092 } 1093 } 1094 1095 result = sbappendaddr_internal(sb, asa, m0, control); 1096 if (result == 0) { 1097 if (m0) 1098 m_freem(m0); 1099 if (control != NULL && !sb_unix) 1100 m_freem(control); 1101 if (error_out) 1102 *error_out = ENOBUFS; 1103 } 1104 1105 return (result); 1106} 1107 1108static int 1109sbappendcontrol_internal(struct sockbuf *sb, struct mbuf *m0, 1110 struct mbuf *control) 1111{ 1112 struct mbuf *m, *mlast, *n; 1113 int space = 0; 1114 1115 if (control == 0) 1116 panic("sbappendcontrol"); 1117 1118 for (m = control; ; m = m->m_next) { 1119 space += m->m_len; 1120 if (m->m_next == 0) 1121 break; 1122 } 1123 n = m; /* save pointer to last control buffer */ 1124 for (m = m0; m; m = m->m_next) 1125 space += m->m_len; 1126 if (space > sbspace(sb) && !(sb->sb_flags & SB_UNIX)) 1127 return (0); 1128 n->m_next = m0; /* concatenate data to control */ 1129 SBLASTRECORDCHK(sb, "sbappendcontrol 1"); 1130 1131 for (m = control; m->m_next != NULL; m = m->m_next) 1132 sballoc(sb, m); 1133 sballoc(sb, m); 1134 mlast = m; 1135 1136 if (sb->sb_lastrecord != NULL) { 1137 sb->sb_lastrecord->m_nextpkt = control; 1138 } else { 1139 sb->sb_mb = control; 1140 } 1141 sb->sb_lastrecord = control; 1142 sb->sb_mbtail = mlast; 1143 1144 SBLASTMBUFCHK(sb, __func__); 1145 SBLASTRECORDCHK(sb, "sbappendcontrol 2"); 1146 1147 postevent(0, sb, EV_RWBYTES); 1148 return (1); 1149} 1150 1151int 1152sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control, 1153 int *error_out) 1154{ 1155 int result = 0; 1156 boolean_t sb_unix = (sb->sb_flags & SB_UNIX); 1157 1158 if (error_out) 1159 *error_out = 0; 1160 1161 if (sb->sb_flags & SB_DROP) { 1162 if (m0 != NULL) 1163 m_freem(m0); 1164 if (control != NULL && !sb_unix) 1165 m_freem(control); 1166 if (error_out != NULL) 1167 *error_out = EINVAL; 1168 return (0); 1169 } 1170 1171 if (sb->sb_flags & SB_RECV) { 1172 int error; 1173 1174 error = sflt_data_in(sb->sb_so, NULL, &m0, &control, 0); 1175 SBLASTRECORDCHK(sb, __func__); 1176 if (error) { 1177 if (error != EJUSTRETURN) { 1178 if (m0) 1179 m_freem(m0); 1180 if (control != NULL && !sb_unix) 1181 m_freem(control); 1182 if (error_out) 1183 *error_out = error; 1184 } 1185 return (0); 1186 } 1187 } 1188 1189 result = sbappendcontrol_internal(sb, m0, control); 1190 if (result == 0) { 1191 if (m0) 1192 m_freem(m0); 1193 if (control != NULL && !sb_unix) 1194 m_freem(control); 1195 if (error_out) 1196 *error_out = ENOBUFS; 1197 } 1198 1199 return (result); 1200} 1201 1202/* 1203 * Append a contiguous TCP data blob with TCP sequence number as control data 1204 * as a new msg to the receive socket buffer. 1205 */ 1206int 1207sbappendmsgstream_rcv(struct sockbuf *sb, struct mbuf *m, uint32_t seqnum, 1208 int unordered) 1209{ 1210 struct mbuf *m_eor = NULL; 1211 u_int32_t data_len = 0; 1212 int ret = 0; 1213 struct socket *so = sb->sb_so; 1214 1215 VERIFY((m->m_flags & M_PKTHDR) && m_pktlen(m) > 0); 1216 VERIFY(so->so_msg_state != NULL); 1217 VERIFY(sb->sb_flags & SB_RECV); 1218 1219 /* Keep the TCP sequence number in the mbuf pkthdr */ 1220 m->m_pkthdr.msg_seq = seqnum; 1221 1222 /* find last mbuf and set M_EOR */ 1223 for (m_eor = m; ; m_eor = m_eor->m_next) { 1224 /* 1225 * If the msg is unordered, we need to account for 1226 * these bytes in receive socket buffer size. Otherwise, 1227 * the receive window advertised will shrink because 1228 * of the additional unordered bytes added to the 1229 * receive buffer. 1230 */ 1231 if (unordered) { 1232 m_eor->m_flags |= M_UNORDERED_DATA; 1233 data_len += m_eor->m_len; 1234 so->so_msg_state->msg_uno_bytes += m_eor->m_len; 1235 } else { 1236 m_eor->m_flags &= ~M_UNORDERED_DATA; 1237 } 1238 1239 if (m_eor->m_next == NULL) 1240 break; 1241 } 1242 1243 /* set EOR flag at end of byte blob */ 1244 m_eor->m_flags |= M_EOR; 1245 1246 /* expand the receive socket buffer to allow unordered data */ 1247 if (unordered && !sbreserve(sb, sb->sb_hiwat + data_len)) { 1248 /* 1249 * Could not allocate memory for unordered data, it 1250 * means this packet will have to be delivered in order 1251 */ 1252 printf("%s: could not reserve space for unordered data\n", 1253 __func__); 1254 } 1255 1256 ret = sbappendrecord(sb, m); 1257 return (ret); 1258} 1259 1260/* 1261 * TCP streams have message based out of order delivery support, or have 1262 * Multipath TCP support, or are regular TCP sockets 1263 */ 1264int 1265sbappendstream_rcvdemux(struct socket *so, struct mbuf *m, uint32_t seqnum, 1266 int unordered) 1267{ 1268 int ret = 0; 1269 1270 if ((m != NULL) && (m_pktlen(m) <= 0)) { 1271 m_freem(m); 1272 return (ret); 1273 } 1274 1275 if (so->so_flags & SOF_ENABLE_MSGS) { 1276 ret = sbappendmsgstream_rcv(&so->so_rcv, m, seqnum, unordered); 1277 } 1278#if MPTCP 1279 else if (so->so_flags & SOF_MPTCP_TRUE) { 1280 ret = sbappendmptcpstream_rcv(&so->so_rcv, m); 1281 } 1282#endif /* MPTCP */ 1283 else { 1284 ret = sbappendstream(&so->so_rcv, m); 1285 } 1286 return (ret); 1287} 1288 1289#if MPTCP 1290int 1291sbappendmptcpstream_rcv(struct sockbuf *sb, struct mbuf *m) 1292{ 1293 struct socket *so = sb->sb_so; 1294 1295 VERIFY(m == NULL || (m->m_flags & M_PKTHDR)); 1296 /* SB_NOCOMPRESS must be set prevent loss of M_PKTHDR data */ 1297 VERIFY((sb->sb_flags & (SB_RECV|SB_NOCOMPRESS)) == 1298 (SB_RECV|SB_NOCOMPRESS)); 1299 1300 if (m == NULL || m_pktlen(m) == 0 || (sb->sb_flags & SB_DROP) || 1301 (so->so_state & SS_CANTRCVMORE)) { 1302 if (m != NULL) 1303 m_freem(m); 1304 return (0); 1305 } 1306 /* the socket is not closed, so SOF_MP_SUBFLOW must be set */ 1307 VERIFY(so->so_flags & SOF_MP_SUBFLOW); 1308 1309 if (m->m_nextpkt != NULL || (sb->sb_mb != sb->sb_lastrecord)) { 1310 panic("%s: nexpkt %p || mb %p != lastrecord %p\n", __func__, 1311 m->m_nextpkt, sb->sb_mb, sb->sb_lastrecord); 1312 /* NOTREACHED */ 1313 } 1314 1315 SBLASTMBUFCHK(sb, __func__); 1316 1317 mptcp_adj_rmap(so, m); 1318 1319 /* No filter support (SB_RECV) on mptcp subflow sockets */ 1320 1321 sbcompress(sb, m, sb->sb_mbtail); 1322 sb->sb_lastrecord = sb->sb_mb; 1323 SBLASTRECORDCHK(sb, __func__); 1324 return (1); 1325} 1326#endif /* MPTCP */ 1327 1328/* 1329 * Append message to send socket buffer based on priority. 1330 */ 1331int 1332sbappendmsg_snd(struct sockbuf *sb, struct mbuf *m) 1333{ 1334 struct socket *so = sb->sb_so; 1335 struct msg_priq *priq; 1336 int set_eor = 0; 1337 1338 VERIFY(so->so_msg_state != NULL); 1339 1340 if (m->m_nextpkt != NULL || (sb->sb_mb != sb->sb_lastrecord)) 1341 panic("sbappendstream: nexpkt %p || mb %p != lastrecord %p\n", 1342 m->m_nextpkt, sb->sb_mb, sb->sb_lastrecord); 1343 1344 SBLASTMBUFCHK(sb, __func__); 1345 1346 if (m == NULL || (sb->sb_flags & SB_DROP) || so->so_msg_state == NULL) { 1347 if (m != NULL) 1348 m_freem(m); 1349 return (0); 1350 } 1351 1352 priq = &so->so_msg_state->msg_priq[m->m_pkthdr.msg_pri]; 1353 1354 /* note if we need to propogate M_EOR to the last mbuf */ 1355 if (m->m_flags & M_EOR) { 1356 set_eor = 1; 1357 1358 /* Reset M_EOR from the first mbuf */ 1359 m->m_flags &= ~(M_EOR); 1360 } 1361 1362 if (priq->msgq_head == NULL) { 1363 VERIFY(priq->msgq_tail == NULL && priq->msgq_lastmsg == NULL); 1364 priq->msgq_head = priq->msgq_lastmsg = m; 1365 } else { 1366 VERIFY(priq->msgq_tail->m_next == NULL); 1367 1368 /* Check if the last message has M_EOR flag set */ 1369 if (priq->msgq_tail->m_flags & M_EOR) { 1370 /* Insert as a new message */ 1371 priq->msgq_lastmsg->m_nextpkt = m; 1372 1373 /* move the lastmsg pointer */ 1374 priq->msgq_lastmsg = m; 1375 } else { 1376 /* Append to the existing message */ 1377 priq->msgq_tail->m_next = m; 1378 } 1379 } 1380 1381 /* Update accounting and the queue tail pointer */ 1382 1383 while (m->m_next != NULL) { 1384 sballoc(sb, m); 1385 priq->msgq_bytes += m->m_len; 1386 m = m->m_next; 1387 } 1388 sballoc(sb, m); 1389 priq->msgq_bytes += m->m_len; 1390 1391 if (set_eor) { 1392 m->m_flags |= M_EOR; 1393 1394 /* 1395 * Since the user space can not write a new msg 1396 * without completing the previous one, we can 1397 * reset this flag to start sending again. 1398 */ 1399 priq->msgq_flags &= ~(MSGQ_MSG_NOTDONE); 1400 } 1401 1402 priq->msgq_tail = m; 1403 1404 SBLASTRECORDCHK(sb, "sbappendstream 2"); 1405 postevent(0, sb, EV_RWBYTES); 1406 return (1); 1407} 1408 1409/* 1410 * Pull data from priority queues to the serial snd queue 1411 * right before sending. 1412 */ 1413void 1414sbpull_unordered_data(struct socket *so, int32_t off, int32_t len) 1415{ 1416 int32_t topull, i; 1417 struct msg_priq *priq = NULL; 1418 1419 VERIFY(so->so_msg_state != NULL); 1420 1421 topull = (off + len) - so->so_msg_state->msg_serial_bytes; 1422 1423 i = MSG_PRI_MAX; 1424 while (i >= MSG_PRI_MIN && topull > 0) { 1425 struct mbuf *m = NULL, *mqhead = NULL, *mend = NULL; 1426 priq = &so->so_msg_state->msg_priq[i]; 1427 if ((priq->msgq_flags & MSGQ_MSG_NOTDONE) && 1428 priq->msgq_head == NULL) { 1429 /* 1430 * We were in the middle of sending 1431 * a message and we have not seen the 1432 * end of it. 1433 */ 1434 VERIFY(priq->msgq_lastmsg == NULL && 1435 priq->msgq_tail == NULL); 1436 return; 1437 } 1438 if (priq->msgq_head != NULL) { 1439 int32_t bytes = 0, topull_tmp = topull; 1440 /* 1441 * We found a msg while scanning the priority 1442 * queue from high to low priority. 1443 */ 1444 m = priq->msgq_head; 1445 mqhead = m; 1446 mend = m; 1447 1448 /* 1449 * Move bytes from the priority queue to the 1450 * serial queue. Compute the number of bytes 1451 * being added. 1452 */ 1453 while (mqhead->m_next != NULL && topull_tmp > 0) { 1454 bytes += mqhead->m_len; 1455 topull_tmp -= mqhead->m_len; 1456 mend = mqhead; 1457 mqhead = mqhead->m_next; 1458 } 1459 1460 if (mqhead->m_next == NULL) { 1461 /* 1462 * If we have only one more mbuf left, 1463 * move the last mbuf of this message to 1464 * serial queue and set the head of the 1465 * queue to be the next message. 1466 */ 1467 bytes += mqhead->m_len; 1468 mend = mqhead; 1469 mqhead = m->m_nextpkt; 1470 if (!(mend->m_flags & M_EOR)) { 1471 /* 1472 * We have not seen the end of 1473 * this message, so we can not 1474 * pull anymore. 1475 */ 1476 priq->msgq_flags |= MSGQ_MSG_NOTDONE; 1477 } else { 1478 /* Reset M_EOR */ 1479 mend->m_flags &= ~(M_EOR); 1480 } 1481 } else { 1482 /* propogate the next msg pointer */ 1483 mqhead->m_nextpkt = m->m_nextpkt; 1484 } 1485 priq->msgq_head = mqhead; 1486 1487 /* 1488 * if the lastmsg pointer points to 1489 * the mbuf that is being dequeued, update 1490 * it to point to the new head. 1491 */ 1492 if (priq->msgq_lastmsg == m) 1493 priq->msgq_lastmsg = priq->msgq_head; 1494 1495 m->m_nextpkt = NULL; 1496 mend->m_next = NULL; 1497 1498 if (priq->msgq_head == NULL) { 1499 /* Moved all messages, update tail */ 1500 priq->msgq_tail = NULL; 1501 VERIFY(priq->msgq_lastmsg == NULL); 1502 } 1503 1504 /* Move it to serial sb_mb queue */ 1505 if (so->so_snd.sb_mb == NULL) { 1506 so->so_snd.sb_mb = m; 1507 } else { 1508 so->so_snd.sb_mbtail->m_next = m; 1509 } 1510 1511 priq->msgq_bytes -= bytes; 1512 VERIFY(priq->msgq_bytes >= 0); 1513 sbwakeup(&so->so_snd); 1514 1515 so->so_msg_state->msg_serial_bytes += bytes; 1516 so->so_snd.sb_mbtail = mend; 1517 so->so_snd.sb_lastrecord = so->so_snd.sb_mb; 1518 1519 topull = 1520 (off + len) - so->so_msg_state->msg_serial_bytes; 1521 1522 if (priq->msgq_flags & MSGQ_MSG_NOTDONE) 1523 break; 1524 } else { 1525 --i; 1526 } 1527 } 1528 sblastrecordchk(&so->so_snd, "sbpull_unordered_data"); 1529 sblastmbufchk(&so->so_snd, "sbpull_unordered_data"); 1530} 1531 1532/* 1533 * Compress mbuf chain m into the socket 1534 * buffer sb following mbuf n. If n 1535 * is null, the buffer is presumed empty. 1536 */ 1537static inline void 1538sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1539{ 1540 int eor = 0, compress = (!(sb->sb_flags & SB_NOCOMPRESS)); 1541 struct mbuf *o; 1542 1543 if (m == NULL) { 1544 /* There is nothing to compress; just update the tail */ 1545 for (; n->m_next != NULL; n = n->m_next) 1546 ; 1547 sb->sb_mbtail = n; 1548 goto done; 1549 } 1550 1551 while (m != NULL) { 1552 eor |= m->m_flags & M_EOR; 1553 if (compress && m->m_len == 0 && (eor == 0 || 1554 (((o = m->m_next) || (o = n)) && o->m_type == m->m_type))) { 1555 if (sb->sb_lastrecord == m) 1556 sb->sb_lastrecord = m->m_next; 1557 m = m_free(m); 1558 continue; 1559 } 1560 if (compress && n != NULL && (n->m_flags & M_EOR) == 0 && 1561#ifndef __APPLE__ 1562 M_WRITABLE(n) && 1563#endif 1564 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1565 m->m_len <= M_TRAILINGSPACE(n) && 1566 n->m_type == m->m_type) { 1567 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, 1568 (unsigned)m->m_len); 1569 n->m_len += m->m_len; 1570 sb->sb_cc += m->m_len; 1571 if (m->m_type != MT_DATA && m->m_type != MT_HEADER && 1572 m->m_type != MT_OOBDATA) { 1573 /* XXX: Probably don't need */ 1574 sb->sb_ctl += m->m_len; 1575 } 1576 m = m_free(m); 1577 continue; 1578 } 1579 if (n != NULL) 1580 n->m_next = m; 1581 else 1582 sb->sb_mb = m; 1583 sb->sb_mbtail = m; 1584 sballoc(sb, m); 1585 n = m; 1586 m->m_flags &= ~M_EOR; 1587 m = m->m_next; 1588 n->m_next = NULL; 1589 } 1590 if (eor != 0) { 1591 if (n != NULL) 1592 n->m_flags |= eor; 1593 else 1594 printf("semi-panic: sbcompress\n"); 1595 } 1596done: 1597 SBLASTMBUFCHK(sb, __func__); 1598 postevent(0, sb, EV_RWBYTES); 1599} 1600 1601void 1602sb_empty_assert(struct sockbuf *sb, const char *where) 1603{ 1604 if (!(sb->sb_cc == 0 && sb->sb_mb == NULL && sb->sb_mbcnt == 0 && 1605 sb->sb_mbtail == NULL && sb->sb_lastrecord == NULL)) { 1606 panic("%s: sb %p so %p cc %d mbcnt %d mb %p mbtail %p " 1607 "lastrecord %p\n", where, sb, sb->sb_so, sb->sb_cc, 1608 sb->sb_mbcnt, sb->sb_mb, sb->sb_mbtail, 1609 sb->sb_lastrecord); 1610 /* NOTREACHED */ 1611 } 1612} 1613 1614static void 1615sbflush_priq(struct msg_priq *priq) 1616{ 1617 struct mbuf *m; 1618 m = priq->msgq_head; 1619 if (m != NULL) 1620 m_freem_list(m); 1621 priq->msgq_head = priq->msgq_tail = priq->msgq_lastmsg = NULL; 1622 priq->msgq_bytes = priq->msgq_flags = 0; 1623} 1624 1625/* 1626 * Free all mbufs in a sockbuf. 1627 * Check that all resources are reclaimed. 1628 */ 1629void 1630sbflush(struct sockbuf *sb) 1631{ 1632 void *lr_saved = __builtin_return_address(0); 1633 struct socket *so = sb->sb_so; 1634#ifdef notyet 1635 lck_mtx_t *mutex_held; 1636#endif 1637 u_int32_t i; 1638 1639 /* so_usecount may be 0 if we get here from sofreelastref() */ 1640 if (so == NULL) { 1641 panic("%s: null so, sb=%p sb_flags=0x%x lr=%p\n", 1642 __func__, sb, sb->sb_flags, lr_saved); 1643 /* NOTREACHED */ 1644 } else if (so->so_usecount < 0) { 1645 panic("%s: sb=%p sb_flags=0x%x sb_so=%p usecount=%d lr=%p " 1646 "lrh= %s\n", __func__, sb, sb->sb_flags, so, 1647 so->so_usecount, lr_saved, solockhistory_nr(so)); 1648 /* NOTREACHED */ 1649 } 1650#ifdef notyet 1651 /* 1652 * XXX: This code is currently commented out, because we may get here 1653 * as part of sofreelastref(), and at that time, pr_getlock() may no 1654 * longer be able to return us the lock; this will be fixed in future. 1655 */ 1656 if (so->so_proto->pr_getlock != NULL) 1657 mutex_held = (*so->so_proto->pr_getlock)(so, 0); 1658 else 1659 mutex_held = so->so_proto->pr_domain->dom_mtx; 1660 1661 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 1662#endif 1663 1664 /* 1665 * Obtain lock on the socket buffer (SB_LOCK). This is required 1666 * to prevent the socket buffer from being unexpectedly altered 1667 * while it is used by another thread in socket send/receive. 1668 * 1669 * sblock() must not fail here, hence the assertion. 1670 */ 1671 (void) sblock(sb, SBL_WAIT | SBL_NOINTR | SBL_IGNDEFUNCT); 1672 VERIFY(sb->sb_flags & SB_LOCK); 1673 1674 while (sb->sb_mbcnt > 0) { 1675 /* 1676 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty: 1677 * we would loop forever. Panic instead. 1678 */ 1679 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 1680 break; 1681 sbdrop(sb, (int)sb->sb_cc); 1682 } 1683 1684 if (!(sb->sb_flags & SB_RECV) && (so->so_flags & SOF_ENABLE_MSGS)) { 1685 VERIFY(so->so_msg_state != NULL); 1686 for (i = MSG_PRI_MIN; i <= MSG_PRI_MAX; ++i) { 1687 sbflush_priq(&so->so_msg_state->msg_priq[i]); 1688 } 1689 so->so_msg_state->msg_serial_bytes = 0; 1690 so->so_msg_state->msg_uno_bytes = 0; 1691 } 1692 1693 sb_empty_assert(sb, __func__); 1694 postevent(0, sb, EV_RWBYTES); 1695 1696 sbunlock(sb, TRUE); /* keep socket locked */ 1697} 1698 1699/* 1700 * Drop data from (the front of) a sockbuf. 1701 * use m_freem_list to free the mbuf structures 1702 * under a single lock... this is done by pruning 1703 * the top of the tree from the body by keeping track 1704 * of where we get to in the tree and then zeroing the 1705 * two pertinent pointers m_nextpkt and m_next 1706 * the socket buffer is then updated to point at the new 1707 * top of the tree and the pruned area is released via 1708 * m_freem_list. 1709 */ 1710void 1711sbdrop(struct sockbuf *sb, int len) 1712{ 1713 struct mbuf *m, *free_list, *ml; 1714 struct mbuf *next, *last; 1715 1716 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1717#if MPTCP 1718 if ((m != NULL) && (len > 0) && 1719 (!(sb->sb_flags & SB_RECV)) && 1720 ((sb->sb_so->so_flags & SOF_MP_SUBFLOW) || 1721 ((SOCK_CHECK_DOM(sb->sb_so, PF_MULTIPATH)) && 1722 (SOCK_CHECK_PROTO(sb->sb_so, IPPROTO_TCP))))) { 1723 mptcp_preproc_sbdrop(m, (unsigned int)len); 1724 } 1725#endif /* MPTCP */ 1726 KERNEL_DEBUG((DBG_FNC_SBDROP | DBG_FUNC_START), sb, len, 0, 0, 0); 1727 1728 free_list = last = m; 1729 ml = (struct mbuf *)0; 1730 1731 while (len > 0) { 1732 if (m == 0) { 1733 if (next == 0) { 1734 /* 1735 * temporarily replacing this panic with printf 1736 * because it occurs occasionally when closing 1737 * a socket when there is no harm in ignoring 1738 * it. This problem will be investigated 1739 * further. 1740 */ 1741 /* panic("sbdrop"); */ 1742 printf("sbdrop - count not zero\n"); 1743 len = 0; 1744 /* 1745 * zero the counts. if we have no mbufs, 1746 * we have no data (PR-2986815) 1747 */ 1748 sb->sb_cc = 0; 1749 sb->sb_mbcnt = 0; 1750 if (!(sb->sb_flags & SB_RECV) && 1751 (sb->sb_so->so_flags & SOF_ENABLE_MSGS)) { 1752 sb->sb_so->so_msg_state-> 1753 msg_serial_bytes = 0; 1754 } 1755 break; 1756 } 1757 m = last = next; 1758 next = m->m_nextpkt; 1759 continue; 1760 } 1761 if (m->m_len > len) { 1762 m->m_len -= len; 1763 m->m_data += len; 1764 sb->sb_cc -= len; 1765 if (m->m_type != MT_DATA && m->m_type != MT_HEADER && 1766 m->m_type != MT_OOBDATA) 1767 sb->sb_ctl -= len; 1768 break; 1769 } 1770 len -= m->m_len; 1771 sbfree(sb, m); 1772 1773 ml = m; 1774 m = m->m_next; 1775 } 1776 while (m && m->m_len == 0) { 1777 sbfree(sb, m); 1778 1779 ml = m; 1780 m = m->m_next; 1781 } 1782 if (ml) { 1783 ml->m_next = (struct mbuf *)0; 1784 last->m_nextpkt = (struct mbuf *)0; 1785 m_freem_list(free_list); 1786 } 1787 if (m) { 1788 sb->sb_mb = m; 1789 m->m_nextpkt = next; 1790 } else { 1791 sb->sb_mb = next; 1792 } 1793 1794 /* 1795 * First part is an inline SB_EMPTY_FIXUP(). Second part 1796 * makes sure sb_lastrecord is up-to-date if we dropped 1797 * part of the last record. 1798 */ 1799 m = sb->sb_mb; 1800 if (m == NULL) { 1801 sb->sb_mbtail = NULL; 1802 sb->sb_lastrecord = NULL; 1803 } else if (m->m_nextpkt == NULL) { 1804 sb->sb_lastrecord = m; 1805 } 1806 1807 postevent(0, sb, EV_RWBYTES); 1808 1809 KERNEL_DEBUG((DBG_FNC_SBDROP | DBG_FUNC_END), sb, 0, 0, 0, 0); 1810} 1811 1812/* 1813 * Drop a record off the front of a sockbuf 1814 * and move the next record to the front. 1815 */ 1816void 1817sbdroprecord(struct sockbuf *sb) 1818{ 1819 struct mbuf *m, *mn; 1820 1821 m = sb->sb_mb; 1822 if (m) { 1823 sb->sb_mb = m->m_nextpkt; 1824 do { 1825 sbfree(sb, m); 1826 MFREE(m, mn); 1827 m = mn; 1828 } while (m); 1829 } 1830 SB_EMPTY_FIXUP(sb); 1831 postevent(0, sb, EV_RWBYTES); 1832} 1833 1834/* 1835 * Create a "control" mbuf containing the specified data 1836 * with the specified type for presentation on a socket buffer. 1837 */ 1838struct mbuf * 1839sbcreatecontrol(caddr_t p, int size, int type, int level) 1840{ 1841 struct cmsghdr *cp; 1842 struct mbuf *m; 1843 1844 if (CMSG_SPACE((u_int)size) > MLEN) 1845 return ((struct mbuf *)NULL); 1846 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL) 1847 return ((struct mbuf *)NULL); 1848 cp = mtod(m, struct cmsghdr *); 1849 VERIFY(IS_P2ALIGNED(cp, sizeof (u_int32_t))); 1850 /* XXX check size? */ 1851 (void) memcpy(CMSG_DATA(cp), p, size); 1852 m->m_len = CMSG_SPACE(size); 1853 cp->cmsg_len = CMSG_LEN(size); 1854 cp->cmsg_level = level; 1855 cp->cmsg_type = type; 1856 return (m); 1857} 1858 1859struct mbuf ** 1860sbcreatecontrol_mbuf(caddr_t p, int size, int type, int level, struct mbuf **mp) 1861{ 1862 struct mbuf *m; 1863 struct cmsghdr *cp; 1864 1865 if (*mp == NULL) { 1866 *mp = sbcreatecontrol(p, size, type, level); 1867 return (mp); 1868 } 1869 1870 if (CMSG_SPACE((u_int)size) + (*mp)->m_len > MLEN) { 1871 mp = &(*mp)->m_next; 1872 *mp = sbcreatecontrol(p, size, type, level); 1873 return (mp); 1874 } 1875 1876 m = *mp; 1877 1878 cp = (struct cmsghdr *)(void *)(mtod(m, char *) + m->m_len); 1879 /* CMSG_SPACE ensures 32-bit alignment */ 1880 VERIFY(IS_P2ALIGNED(cp, sizeof (u_int32_t))); 1881 m->m_len += CMSG_SPACE(size); 1882 1883 /* XXX check size? */ 1884 (void) memcpy(CMSG_DATA(cp), p, size); 1885 cp->cmsg_len = CMSG_LEN(size); 1886 cp->cmsg_level = level; 1887 cp->cmsg_type = type; 1888 1889 return (mp); 1890} 1891 1892 1893/* 1894 * Some routines that return EOPNOTSUPP for entry points that are not 1895 * supported by a protocol. Fill in as needed. 1896 */ 1897int 1898pru_abort_notsupp(struct socket *so) 1899{ 1900#pragma unused(so) 1901 return (EOPNOTSUPP); 1902} 1903 1904int 1905pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 1906{ 1907#pragma unused(so, nam) 1908 return (EOPNOTSUPP); 1909} 1910 1911int 1912pru_attach_notsupp(struct socket *so, int proto, struct proc *p) 1913{ 1914#pragma unused(so, proto, p) 1915 return (EOPNOTSUPP); 1916} 1917 1918int 1919pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct proc *p) 1920{ 1921#pragma unused(so, nam, p) 1922 return (EOPNOTSUPP); 1923} 1924 1925int 1926pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct proc *p) 1927{ 1928#pragma unused(so, nam, p) 1929 return (EOPNOTSUPP); 1930} 1931 1932int 1933pru_connect2_notsupp(struct socket *so1, struct socket *so2) 1934{ 1935#pragma unused(so1, so2) 1936 return (EOPNOTSUPP); 1937} 1938 1939int 1940pru_connectx_notsupp(struct socket *so, struct sockaddr_list **src_sl, 1941 struct sockaddr_list **dst_sl, struct proc *p, uint32_t ifscope, 1942 associd_t aid, connid_t *pcid, uint32_t flags, void *arg, 1943 uint32_t arglen) 1944{ 1945#pragma unused(so, src_sl, dst_sl, p, ifscope, aid, pcid, flags, arg, arglen) 1946 return (EOPNOTSUPP); 1947} 1948 1949int 1950pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 1951 struct ifnet *ifp, struct proc *p) 1952{ 1953#pragma unused(so, cmd, data, ifp, p) 1954 return (EOPNOTSUPP); 1955} 1956 1957int 1958pru_detach_notsupp(struct socket *so) 1959{ 1960#pragma unused(so) 1961 return (EOPNOTSUPP); 1962} 1963 1964int 1965pru_disconnect_notsupp(struct socket *so) 1966{ 1967#pragma unused(so) 1968 return (EOPNOTSUPP); 1969} 1970 1971int 1972pru_disconnectx_notsupp(struct socket *so, associd_t aid, connid_t cid) 1973{ 1974#pragma unused(so, aid, cid) 1975 return (EOPNOTSUPP); 1976} 1977 1978int 1979pru_listen_notsupp(struct socket *so, struct proc *p) 1980{ 1981#pragma unused(so, p) 1982 return (EOPNOTSUPP); 1983} 1984 1985int 1986pru_peeloff_notsupp(struct socket *so, associd_t aid, struct socket **psop) 1987{ 1988#pragma unused(so, aid, psop) 1989 return (EOPNOTSUPP); 1990} 1991 1992int 1993pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 1994{ 1995#pragma unused(so, nam) 1996 return (EOPNOTSUPP); 1997} 1998 1999int 2000pru_rcvd_notsupp(struct socket *so, int flags) 2001{ 2002#pragma unused(so, flags) 2003 return (EOPNOTSUPP); 2004} 2005 2006int 2007pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 2008{ 2009#pragma unused(so, m, flags) 2010 return (EOPNOTSUPP); 2011} 2012 2013int 2014pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 2015 struct sockaddr *addr, struct mbuf *control, struct proc *p) 2016{ 2017#pragma unused(so, flags, m, addr, control, p) 2018 return (EOPNOTSUPP); 2019} 2020 2021/* 2022 * This isn't really a ``null'' operation, but it's the default one 2023 * and doesn't do anything destructive. 2024 */ 2025int 2026pru_sense_null(struct socket *so, void *ub, int isstat64) 2027{ 2028 if (isstat64 != 0) { 2029 struct stat64 *sb64; 2030 2031 sb64 = (struct stat64 *)ub; 2032 sb64->st_blksize = so->so_snd.sb_hiwat; 2033 } else { 2034 struct stat *sb; 2035 2036 sb = (struct stat *)ub; 2037 sb->st_blksize = so->so_snd.sb_hiwat; 2038 } 2039 2040 return (0); 2041} 2042 2043 2044int 2045pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 2046 struct mbuf *top, struct mbuf *control, int flags) 2047{ 2048#pragma unused(so, addr, uio, top, control, flags) 2049 return (EOPNOTSUPP); 2050} 2051 2052int 2053pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 2054 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2055{ 2056#pragma unused(so, paddr, uio, mp0, controlp, flagsp) 2057 return (EOPNOTSUPP); 2058} 2059 2060int 2061pru_shutdown_notsupp(struct socket *so) 2062{ 2063#pragma unused(so) 2064 return (EOPNOTSUPP); 2065} 2066 2067int 2068pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 2069{ 2070#pragma unused(so, nam) 2071 return (EOPNOTSUPP); 2072} 2073 2074int 2075pru_sopoll_notsupp(struct socket *so, int events, kauth_cred_t cred, void *wql) 2076{ 2077#pragma unused(so, events, cred, wql) 2078 return (EOPNOTSUPP); 2079} 2080 2081int 2082pru_socheckopt_null(struct socket *so, struct sockopt *sopt) 2083{ 2084#pragma unused(so, sopt) 2085 /* 2086 * Allow all options for set/get by default. 2087 */ 2088 return (0); 2089} 2090 2091void 2092pru_sanitize(struct pr_usrreqs *pru) 2093{ 2094#define DEFAULT(foo, bar) if ((foo) == NULL) (foo) = (bar) 2095 DEFAULT(pru->pru_abort, pru_abort_notsupp); 2096 DEFAULT(pru->pru_accept, pru_accept_notsupp); 2097 DEFAULT(pru->pru_attach, pru_attach_notsupp); 2098 DEFAULT(pru->pru_bind, pru_bind_notsupp); 2099 DEFAULT(pru->pru_connect, pru_connect_notsupp); 2100 DEFAULT(pru->pru_connect2, pru_connect2_notsupp); 2101 DEFAULT(pru->pru_connectx, pru_connectx_notsupp); 2102 DEFAULT(pru->pru_control, pru_control_notsupp); 2103 DEFAULT(pru->pru_detach, pru_detach_notsupp); 2104 DEFAULT(pru->pru_disconnect, pru_disconnect_notsupp); 2105 DEFAULT(pru->pru_disconnectx, pru_disconnectx_notsupp); 2106 DEFAULT(pru->pru_listen, pru_listen_notsupp); 2107 DEFAULT(pru->pru_peeloff, pru_peeloff_notsupp); 2108 DEFAULT(pru->pru_peeraddr, pru_peeraddr_notsupp); 2109 DEFAULT(pru->pru_rcvd, pru_rcvd_notsupp); 2110 DEFAULT(pru->pru_rcvoob, pru_rcvoob_notsupp); 2111 DEFAULT(pru->pru_send, pru_send_notsupp); 2112 DEFAULT(pru->pru_sense, pru_sense_null); 2113 DEFAULT(pru->pru_shutdown, pru_shutdown_notsupp); 2114 DEFAULT(pru->pru_sockaddr, pru_sockaddr_notsupp); 2115 DEFAULT(pru->pru_sopoll, pru_sopoll_notsupp); 2116 DEFAULT(pru->pru_soreceive, pru_soreceive_notsupp); 2117 DEFAULT(pru->pru_sosend, pru_sosend_notsupp); 2118 DEFAULT(pru->pru_socheckopt, pru_socheckopt_null); 2119#undef DEFAULT 2120} 2121 2122/* 2123 * The following are macros on BSD and functions on Darwin 2124 */ 2125 2126/* 2127 * Do we need to notify the other side when I/O is possible? 2128 */ 2129 2130int 2131sb_notify(struct sockbuf *sb) 2132{ 2133 return (sb->sb_waiters > 0 || 2134 (sb->sb_flags & (SB_SEL|SB_ASYNC|SB_UPCALL|SB_KNOTE))); 2135} 2136 2137/* 2138 * How much space is there in a socket buffer (so->so_snd or so->so_rcv)? 2139 * This is problematical if the fields are unsigned, as the space might 2140 * still be negative (cc > hiwat or mbcnt > mbmax). Should detect 2141 * overflow and return 0. 2142 */ 2143int 2144sbspace(struct sockbuf *sb) 2145{ 2146 int space = imin((int)(sb->sb_hiwat - sb->sb_cc), 2147 (int)(sb->sb_mbmax - sb->sb_mbcnt)); 2148 if (space < 0) 2149 space = 0; 2150 2151 return (space); 2152} 2153 2154/* 2155 * If this socket has priority queues, check if there is enough 2156 * space in the priority queue for this msg. 2157 */ 2158int 2159msgq_sbspace(struct socket *so, struct mbuf *control) 2160{ 2161 int space = 0, error; 2162 u_int32_t msgpri; 2163 VERIFY(so->so_type == SOCK_STREAM && SOCK_PROTO(so) == IPPROTO_TCP && 2164 control != NULL); 2165 error = tcp_get_msg_priority(control, &msgpri); 2166 if (error) 2167 return (0); 2168 space = (so->so_snd.sb_idealsize / MSG_PRI_COUNT) - 2169 so->so_msg_state->msg_priq[msgpri].msgq_bytes; 2170 if (space < 0) 2171 space = 0; 2172 return (space); 2173} 2174 2175/* do we have to send all at once on a socket? */ 2176int 2177sosendallatonce(struct socket *so) 2178{ 2179 return (so->so_proto->pr_flags & PR_ATOMIC); 2180} 2181 2182/* can we read something from so? */ 2183int 2184soreadable(struct socket *so) 2185{ 2186 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat || 2187 (so->so_state & SS_CANTRCVMORE) || 2188 so->so_comp.tqh_first || so->so_error); 2189} 2190 2191/* can we write something to so? */ 2192 2193int 2194sowriteable(struct socket *so) 2195{ 2196 return ((!so_wait_for_if_feedback(so) && 2197 sbspace(&(so)->so_snd) >= (so)->so_snd.sb_lowat && 2198 ((so->so_state & SS_ISCONNECTED) || 2199 (so->so_proto->pr_flags & PR_CONNREQUIRED) == 0)) || 2200 (so->so_state & SS_CANTSENDMORE) || 2201 so->so_error); 2202} 2203 2204/* adjust counters in sb reflecting allocation of m */ 2205 2206void 2207sballoc(struct sockbuf *sb, struct mbuf *m) 2208{ 2209 u_int32_t cnt = 1; 2210 sb->sb_cc += m->m_len; 2211 if (m->m_type != MT_DATA && m->m_type != MT_HEADER && 2212 m->m_type != MT_OOBDATA) 2213 sb->sb_ctl += m->m_len; 2214 sb->sb_mbcnt += MSIZE; 2215 2216 if (m->m_flags & M_EXT) { 2217 sb->sb_mbcnt += m->m_ext.ext_size; 2218 cnt += (m->m_ext.ext_size >> MSIZESHIFT); 2219 } 2220 OSAddAtomic(cnt, &total_sbmb_cnt); 2221 VERIFY(total_sbmb_cnt > 0); 2222} 2223 2224/* adjust counters in sb reflecting freeing of m */ 2225void 2226sbfree(struct sockbuf *sb, struct mbuf *m) 2227{ 2228 int cnt = -1; 2229 2230 sb->sb_cc -= m->m_len; 2231 if (m->m_type != MT_DATA && m->m_type != MT_HEADER && 2232 m->m_type != MT_OOBDATA) 2233 sb->sb_ctl -= m->m_len; 2234 sb->sb_mbcnt -= MSIZE; 2235 if (m->m_flags & M_EXT) { 2236 sb->sb_mbcnt -= m->m_ext.ext_size; 2237 cnt -= (m->m_ext.ext_size >> MSIZESHIFT); 2238 } 2239 OSAddAtomic(cnt, &total_sbmb_cnt); 2240 VERIFY(total_sbmb_cnt >= 0); 2241} 2242 2243/* 2244 * Set lock on sockbuf sb; sleep if lock is already held. 2245 * Unless SB_NOINTR is set on sockbuf, sleep is interruptible. 2246 * Returns error without lock if sleep is interrupted. 2247 */ 2248int 2249sblock(struct sockbuf *sb, uint32_t flags) 2250{ 2251 boolean_t nointr = ((sb->sb_flags & SB_NOINTR) || (flags & SBL_NOINTR)); 2252 void *lr_saved = __builtin_return_address(0); 2253 struct socket *so = sb->sb_so; 2254 void * wchan; 2255 int error = 0; 2256 2257 VERIFY((flags & SBL_VALID) == flags); 2258 2259 /* so_usecount may be 0 if we get here from sofreelastref() */ 2260 if (so == NULL) { 2261 panic("%s: null so, sb=%p sb_flags=0x%x lr=%p\n", 2262 __func__, sb, sb->sb_flags, lr_saved); 2263 /* NOTREACHED */ 2264 } else if (so->so_usecount < 0) { 2265 panic("%s: sb=%p sb_flags=0x%x sb_so=%p usecount=%d lr=%p " 2266 "lrh= %s\n", __func__, sb, sb->sb_flags, so, 2267 so->so_usecount, lr_saved, solockhistory_nr(so)); 2268 /* NOTREACHED */ 2269 } 2270 2271 if ((sb->sb_flags & SB_LOCK) && !(flags & SBL_WAIT)) 2272 return (EWOULDBLOCK); 2273 2274 /* 2275 * We may get here from sorflush(), in which case "sb" may not 2276 * point to the real socket buffer. Use the actual socket buffer 2277 * address from the socket instead. 2278 */ 2279 wchan = (sb->sb_flags & SB_RECV) ? 2280 &so->so_rcv.sb_flags : &so->so_snd.sb_flags; 2281 2282 while (sb->sb_flags & SB_LOCK) { 2283 lck_mtx_t *mutex_held; 2284 2285 /* 2286 * XXX: This code should be moved up above outside of this loop; 2287 * however, we may get here as part of sofreelastref(), and 2288 * at that time pr_getlock() may no longer be able to return 2289 * us the lock. This will be fixed in future. 2290 */ 2291 if (so->so_proto->pr_getlock != NULL) 2292 mutex_held = (*so->so_proto->pr_getlock)(so, 0); 2293 else 2294 mutex_held = so->so_proto->pr_domain->dom_mtx; 2295 2296 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 2297 2298 sb->sb_wantlock++; 2299 VERIFY(sb->sb_wantlock != 0); 2300 2301 error = msleep(wchan, mutex_held, 2302 nointr ? PSOCK : PSOCK | PCATCH, 2303 nointr ? "sb_lock_nointr" : "sb_lock", NULL); 2304 2305 VERIFY(sb->sb_wantlock != 0); 2306 sb->sb_wantlock--; 2307 2308 if (error == 0 && (so->so_flags & SOF_DEFUNCT) && 2309 !(flags & SBL_IGNDEFUNCT)) { 2310 error = EBADF; 2311 SODEFUNCTLOG(("%s[%d]: defunct so 0x%llx [%d,%d] " 2312 "(%d)\n", __func__, proc_selfpid(), 2313 (uint64_t)VM_KERNEL_ADDRPERM(so), 2314 SOCK_DOM(so), SOCK_TYPE(so), error)); 2315 } 2316 2317 if (error != 0) 2318 return (error); 2319 } 2320 sb->sb_flags |= SB_LOCK; 2321 return (0); 2322} 2323 2324/* 2325 * Release lock on sockbuf sb 2326 */ 2327void 2328sbunlock(struct sockbuf *sb, boolean_t keeplocked) 2329{ 2330 void *lr_saved = __builtin_return_address(0); 2331 struct socket *so = sb->sb_so; 2332 2333 /* so_usecount may be 0 if we get here from sofreelastref() */ 2334 if (so == NULL) { 2335 panic("%s: null so, sb=%p sb_flags=0x%x lr=%p\n", 2336 __func__, sb, sb->sb_flags, lr_saved); 2337 /* NOTREACHED */ 2338 } else if (so->so_usecount < 0) { 2339 panic("%s: sb=%p sb_flags=0x%x sb_so=%p usecount=%d lr=%p " 2340 "lrh= %s\n", __func__, sb, sb->sb_flags, so, 2341 so->so_usecount, lr_saved, solockhistory_nr(so)); 2342 /* NOTREACHED */ 2343 } 2344 2345 VERIFY(sb->sb_flags & SB_LOCK); 2346 sb->sb_flags &= ~SB_LOCK; 2347 2348 if (sb->sb_wantlock > 0) { 2349 /* 2350 * We may get here from sorflush(), in which case "sb" may not 2351 * point to the real socket buffer. Use the actual socket 2352 * buffer address from the socket instead. 2353 */ 2354 wakeup((sb->sb_flags & SB_RECV) ? &so->so_rcv.sb_flags : 2355 &so->so_snd.sb_flags); 2356 } 2357 2358 if (!keeplocked) { /* unlock on exit */ 2359 lck_mtx_t *mutex_held; 2360 2361 if (so->so_proto->pr_getlock != NULL) 2362 mutex_held = (*so->so_proto->pr_getlock)(so, 0); 2363 else 2364 mutex_held = so->so_proto->pr_domain->dom_mtx; 2365 2366 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 2367 2368 VERIFY(so->so_usecount != 0); 2369 so->so_usecount--; 2370 so->unlock_lr[so->next_unlock_lr] = lr_saved; 2371 so->next_unlock_lr = (so->next_unlock_lr + 1) % SO_LCKDBG_MAX; 2372 lck_mtx_unlock(mutex_held); 2373 } 2374} 2375 2376void 2377sorwakeup(struct socket *so) 2378{ 2379 if (sb_notify(&so->so_rcv)) 2380 sowakeup(so, &so->so_rcv); 2381} 2382 2383void 2384sowwakeup(struct socket *so) 2385{ 2386 if (sb_notify(&so->so_snd)) 2387 sowakeup(so, &so->so_snd); 2388} 2389 2390void 2391soevent(struct socket *so, long hint) 2392{ 2393 if (so->so_flags & SOF_KNOTE) 2394 KNOTE(&so->so_klist, hint); 2395 2396 soevupcall(so, hint); 2397 2398 /* Don't post an event if this a subflow socket */ 2399 if ((hint & SO_FILT_HINT_IFDENIED) && !(so->so_flags & SOF_MP_SUBFLOW)) 2400 soevent_ifdenied(so); 2401} 2402 2403void 2404soevupcall(struct socket *so, u_int32_t hint) 2405{ 2406 void (*so_event)(struct socket *, void *, uint32_t); 2407 2408 if ((so_event = so->so_event) != NULL) { 2409 caddr_t so_eventarg = so->so_eventarg; 2410 2411 hint &= so->so_eventmask; 2412 if (hint != 0) { 2413 socket_unlock(so, 0); 2414 so->so_event(so, so_eventarg, hint); 2415 socket_lock(so, 0); 2416 } 2417 } 2418} 2419 2420static void 2421soevent_ifdenied(struct socket *so) 2422{ 2423 struct kev_netpolicy_ifdenied ev_ifdenied; 2424 2425 bzero(&ev_ifdenied, sizeof (ev_ifdenied)); 2426 /* 2427 * The event consumer is interested about the effective {upid,pid,uuid} 2428 * info which can be different than the those related to the process 2429 * that recently performed a system call on the socket, i.e. when the 2430 * socket is delegated. 2431 */ 2432 if (so->so_flags & SOF_DELEGATED) { 2433 ev_ifdenied.ev_data.eupid = so->e_upid; 2434 ev_ifdenied.ev_data.epid = so->e_pid; 2435 uuid_copy(ev_ifdenied.ev_data.euuid, so->e_uuid); 2436 } else { 2437 ev_ifdenied.ev_data.eupid = so->last_upid; 2438 ev_ifdenied.ev_data.epid = so->last_pid; 2439 uuid_copy(ev_ifdenied.ev_data.euuid, so->last_uuid); 2440 } 2441 2442 if (++so->so_ifdenied_notifies > 1) { 2443 /* 2444 * Allow for at most one kernel event to be generated per 2445 * socket; so_ifdenied_notifies is reset upon changes in 2446 * the UUID policy. See comments in inp_update_policy. 2447 */ 2448 if (net_io_policy_log) { 2449 uuid_string_t buf; 2450 2451 uuid_unparse(ev_ifdenied.ev_data.euuid, buf); 2452 log(LOG_DEBUG, "%s[%d]: so 0x%llx [%d,%d] epid %d " 2453 "euuid %s%s has %d redundant events supressed\n", 2454 __func__, so->last_pid, 2455 (uint64_t)VM_KERNEL_ADDRPERM(so), SOCK_DOM(so), 2456 SOCK_TYPE(so), ev_ifdenied.ev_data.epid, buf, 2457 ((so->so_flags & SOF_DELEGATED) ? 2458 " [delegated]" : ""), so->so_ifdenied_notifies); 2459 } 2460 } else { 2461 if (net_io_policy_log) { 2462 uuid_string_t buf; 2463 2464 uuid_unparse(ev_ifdenied.ev_data.euuid, buf); 2465 log(LOG_DEBUG, "%s[%d]: so 0x%llx [%d,%d] epid %d " 2466 "euuid %s%s event posted\n", __func__, 2467 so->last_pid, (uint64_t)VM_KERNEL_ADDRPERM(so), 2468 SOCK_DOM(so), SOCK_TYPE(so), 2469 ev_ifdenied.ev_data.epid, buf, 2470 ((so->so_flags & SOF_DELEGATED) ? 2471 " [delegated]" : "")); 2472 } 2473 netpolicy_post_msg(KEV_NETPOLICY_IFDENIED, &ev_ifdenied.ev_data, 2474 sizeof (ev_ifdenied)); 2475 } 2476} 2477 2478/* 2479 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 2480 */ 2481struct sockaddr * 2482dup_sockaddr(struct sockaddr *sa, int canwait) 2483{ 2484 struct sockaddr *sa2; 2485 2486 MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME, 2487 canwait ? M_WAITOK : M_NOWAIT); 2488 if (sa2) 2489 bcopy(sa, sa2, sa->sa_len); 2490 return (sa2); 2491} 2492 2493/* 2494 * Create an external-format (``xsocket'') structure using the information 2495 * in the kernel-format socket structure pointed to by so. This is done 2496 * to reduce the spew of irrelevant information over this interface, 2497 * to isolate user code from changes in the kernel structure, and 2498 * potentially to provide information-hiding if we decide that 2499 * some of this information should be hidden from users. 2500 */ 2501void 2502sotoxsocket(struct socket *so, struct xsocket *xso) 2503{ 2504 xso->xso_len = sizeof (*xso); 2505 xso->xso_so = (_XSOCKET_PTR(struct socket *))VM_KERNEL_ADDRPERM(so); 2506 xso->so_type = so->so_type; 2507 xso->so_options = (short)(so->so_options & 0xffff); 2508 xso->so_linger = so->so_linger; 2509 xso->so_state = so->so_state; 2510 xso->so_pcb = (_XSOCKET_PTR(caddr_t))VM_KERNEL_ADDRPERM(so->so_pcb); 2511 if (so->so_proto) { 2512 xso->xso_protocol = SOCK_PROTO(so); 2513 xso->xso_family = SOCK_DOM(so); 2514 } else { 2515 xso->xso_protocol = xso->xso_family = 0; 2516 } 2517 xso->so_qlen = so->so_qlen; 2518 xso->so_incqlen = so->so_incqlen; 2519 xso->so_qlimit = so->so_qlimit; 2520 xso->so_timeo = so->so_timeo; 2521 xso->so_error = so->so_error; 2522 xso->so_pgid = so->so_pgid; 2523 xso->so_oobmark = so->so_oobmark; 2524 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 2525 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 2526 xso->so_uid = kauth_cred_getuid(so->so_cred); 2527} 2528 2529 2530 2531void 2532sotoxsocket64(struct socket *so, struct xsocket64 *xso) 2533{ 2534 xso->xso_len = sizeof (*xso); 2535 xso->xso_so = (u_int64_t)VM_KERNEL_ADDRPERM(so); 2536 xso->so_type = so->so_type; 2537 xso->so_options = (short)(so->so_options & 0xffff); 2538 xso->so_linger = so->so_linger; 2539 xso->so_state = so->so_state; 2540 xso->so_pcb = (u_int64_t)VM_KERNEL_ADDRPERM(so->so_pcb); 2541 if (so->so_proto) { 2542 xso->xso_protocol = SOCK_PROTO(so); 2543 xso->xso_family = SOCK_DOM(so); 2544 } else { 2545 xso->xso_protocol = xso->xso_family = 0; 2546 } 2547 xso->so_qlen = so->so_qlen; 2548 xso->so_incqlen = so->so_incqlen; 2549 xso->so_qlimit = so->so_qlimit; 2550 xso->so_timeo = so->so_timeo; 2551 xso->so_error = so->so_error; 2552 xso->so_pgid = so->so_pgid; 2553 xso->so_oobmark = so->so_oobmark; 2554 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 2555 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 2556 xso->so_uid = kauth_cred_getuid(so->so_cred); 2557} 2558 2559 2560/* 2561 * This does the same for sockbufs. Note that the xsockbuf structure, 2562 * since it is always embedded in a socket, does not include a self 2563 * pointer nor a length. We make this entry point public in case 2564 * some other mechanism needs it. 2565 */ 2566void 2567sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 2568{ 2569 xsb->sb_cc = sb->sb_cc; 2570 xsb->sb_hiwat = sb->sb_hiwat; 2571 xsb->sb_mbcnt = sb->sb_mbcnt; 2572 xsb->sb_mbmax = sb->sb_mbmax; 2573 xsb->sb_lowat = sb->sb_lowat; 2574 xsb->sb_flags = sb->sb_flags; 2575 xsb->sb_timeo = (short) 2576 (sb->sb_timeo.tv_sec * hz) + sb->sb_timeo.tv_usec / tick; 2577 if (xsb->sb_timeo == 0 && sb->sb_timeo.tv_usec != 0) 2578 xsb->sb_timeo = 1; 2579} 2580 2581/* 2582 * Based on the policy set by an all knowing decison maker, throttle sockets 2583 * that either have been marked as belonging to "background" process. 2584 */ 2585int 2586soisthrottled(struct socket *so) 2587{ 2588 /* 2589 * On non-embedded, we rely on implicit throttling by the 2590 * application, as we're missing the system wide "decision maker" 2591 */ 2592 return ( 2593 (so->so_traffic_mgt_flags & TRAFFIC_MGT_SO_BACKGROUND)); 2594} 2595 2596int 2597soisprivilegedtraffic(struct socket *so) 2598{ 2599 return ((so->so_flags & SOF_PRIVILEGED_TRAFFIC_CLASS) ? 1 : 0); 2600} 2601 2602int 2603soissrcbackground(struct socket *so) 2604{ 2605 return ((so->so_traffic_mgt_flags & TRAFFIC_MGT_SO_BACKGROUND) || 2606 IS_SO_TC_BACKGROUND(so->so_traffic_class)); 2607} 2608 2609/* 2610 * Here is the definition of some of the basic objects in the kern.ipc 2611 * branch of the MIB. 2612 */ 2613SYSCTL_NODE(_kern, KERN_IPC, ipc, 2614 CTLFLAG_RW|CTLFLAG_LOCKED|CTLFLAG_ANYBODY, 0, "IPC"); 2615 2616/* Check that the maximum socket buffer size is within a range */ 2617 2618static int 2619sysctl_sb_max SYSCTL_HANDLER_ARGS 2620{ 2621#pragma unused(oidp, arg1, arg2) 2622 u_int32_t new_value; 2623 int changed = 0; 2624 int error = sysctl_io_number(req, sb_max, sizeof (u_int32_t), 2625 &new_value, &changed); 2626 if (!error && changed) { 2627 if (new_value > LOW_SB_MAX && new_value <= high_sb_max) { 2628 sb_max = new_value; 2629 } else { 2630 error = ERANGE; 2631 } 2632 } 2633 return (error); 2634} 2635 2636static int 2637sysctl_io_policy_throttled SYSCTL_HANDLER_ARGS 2638{ 2639#pragma unused(arg1, arg2) 2640 int i, err; 2641 2642 i = net_io_policy_throttled; 2643 2644 err = sysctl_handle_int(oidp, &i, 0, req); 2645 if (err != 0 || req->newptr == USER_ADDR_NULL) 2646 return (err); 2647 2648 if (i != net_io_policy_throttled) 2649 SOTHROTTLELOG(("throttle: network IO policy throttling is " 2650 "now %s\n", i ? "ON" : "OFF")); 2651 2652 net_io_policy_throttled = i; 2653 2654 return (err); 2655} 2656 2657SYSCTL_PROC(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, 2658 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, 2659 &sb_max, 0, &sysctl_sb_max, "IU", "Maximum socket buffer size"); 2660 2661SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, 2662 CTLFLAG_RD | CTLFLAG_LOCKED, &maxsockets, 0, 2663 "Maximum number of sockets avaliable"); 2664 2665SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, 2666 CTLFLAG_RW | CTLFLAG_LOCKED, &sb_efficiency, 0, ""); 2667 2668SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, 2669 CTLFLAG_RD | CTLFLAG_LOCKED, &nmbclusters, 0, ""); 2670 2671SYSCTL_INT(_kern_ipc, OID_AUTO, njcl, 2672 CTLFLAG_RD | CTLFLAG_LOCKED, &njcl, 0, ""); 2673 2674SYSCTL_INT(_kern_ipc, OID_AUTO, njclbytes, 2675 CTLFLAG_RD | CTLFLAG_LOCKED, &njclbytes, 0, ""); 2676 2677SYSCTL_INT(_kern_ipc, KIPC_SOQLIMITCOMPAT, soqlimitcompat, 2678 CTLFLAG_RW | CTLFLAG_LOCKED, &soqlimitcompat, 1, 2679 "Enable socket queue limit compatibility"); 2680 2681SYSCTL_INT(_kern_ipc, OID_AUTO, soqlencomp, CTLFLAG_RW | CTLFLAG_LOCKED, 2682 &soqlencomp, 0, "Listen backlog represents only complete queue"); 2683 2684SYSCTL_NODE(_kern_ipc, OID_AUTO, io_policy, CTLFLAG_RW, 0, "network IO policy"); 2685 2686SYSCTL_PROC(_kern_ipc_io_policy, OID_AUTO, throttled, 2687 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &net_io_policy_throttled, 0, 2688 sysctl_io_policy_throttled, "I", ""); 2689 2690SYSCTL_INT(_kern_ipc_io_policy, OID_AUTO, log, CTLFLAG_RW | CTLFLAG_LOCKED, 2691 &net_io_policy_log, 0, ""); 2692 2693#if CONFIG_PROC_UUID_POLICY 2694SYSCTL_INT(_kern_ipc_io_policy, OID_AUTO, uuid, CTLFLAG_RW | CTLFLAG_LOCKED, 2695 &net_io_policy_uuid, 0, ""); 2696#endif /* CONFIG_PROC_UUID_POLICY */ 2697