1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 34 * $FreeBSD: src/sys/kern/uipc_socket2.c,v 1.55.2.17 2002/08/31 19:04:55 dwmalone Exp $ 35 * $DragonFly: src/sys/kern/uipc_socket2.c,v 1.13 2004/06/06 19:16:06 dillon Exp $ 36 */ 37 38 #include "opt_param.h" 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/domain.h> 42 #include <sys/file.h> /* for maxfiles */ 43 #include <sys/kernel.h> 44 #include <sys/proc.h> 45 #include <sys/malloc.h> 46 #include <sys/mbuf.h> 47 #include <sys/protosw.h> 48 #include <sys/resourcevar.h> 49 #include <sys/stat.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/signalvar.h> 53 #include <sys/sysctl.h> 54 #include <sys/aio.h> /* for aio_swake proto */ 55 #include <sys/event.h> 56 57 #include <sys/thread2.h> 58 #include <sys/msgport2.h> 59 60 int maxsockets; 61 62 /* 63 * Primitive routines for operating on sockets and socket buffers 64 */ 65 66 u_long sb_max = SB_MAX; 67 u_long sb_max_adj = 68 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 69 70 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 71 72 /* 73 * Procedures to manipulate state flags of socket 74 * and do appropriate wakeups. Normal sequence from the 75 * active (originating) side is that soisconnecting() is 76 * called during processing of connect() call, 77 * resulting in an eventual call to soisconnected() if/when the 78 * connection is established. When the connection is torn down 79 * soisdisconnecting() is called during processing of disconnect() call, 80 * and soisdisconnected() is called when the connection to the peer 81 * is totally severed. The semantics of these routines are such that 82 * connectionless protocols can call soisconnected() and soisdisconnected() 83 * only, bypassing the in-progress calls when setting up a ``connection'' 84 * takes no time. 85 * 86 * From the passive side, a socket is created with 87 * two queues of sockets: so_incomp for connections in progress 88 * and so_comp for connections already made and awaiting user acceptance. 89 * As a protocol is preparing incoming connections, it creates a socket 90 * structure queued on so_incomp by calling sonewconn(). When the connection 91 * is established, soisconnected() is called, and transfers the 92 * socket structure to so_comp, making it available to accept(). 93 * 94 * If a socket is closed with sockets on either 95 * so_incomp or so_comp, these sockets are dropped. 96 * 97 * If higher level protocols are implemented in 98 * the kernel, the wakeups done here will sometimes 99 * cause software-interrupt process scheduling. 100 */ 101 102 void 103 soisconnecting(so) 104 struct socket *so; 105 { 106 107 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 108 so->so_state |= SS_ISCONNECTING; 109 } 110 111 void 112 soisconnected(so) 113 struct socket *so; 114 { 115 struct socket *head = so->so_head; 116 117 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 118 so->so_state |= SS_ISCONNECTED; 119 if (head && (so->so_state & SS_INCOMP)) { 120 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 121 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 122 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 123 so->so_rcv.sb_flags |= SB_UPCALL; 124 so->so_options &= ~SO_ACCEPTFILTER; 125 so->so_upcall(so, so->so_upcallarg, 0); 126 return; 127 } 128 TAILQ_REMOVE(&head->so_incomp, so, so_list); 129 head->so_incqlen--; 130 so->so_state &= ~SS_INCOMP; 131 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 132 head->so_qlen++; 133 so->so_state |= SS_COMP; 134 sorwakeup(head); 135 wakeup_one(&head->so_timeo); 136 } else { 137 wakeup(&so->so_timeo); 138 sorwakeup(so); 139 sowwakeup(so); 140 } 141 } 142 143 void 144 soisdisconnecting(so) 145 struct socket *so; 146 { 147 148 so->so_state &= ~SS_ISCONNECTING; 149 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 150 wakeup((caddr_t)&so->so_timeo); 151 sowwakeup(so); 152 sorwakeup(so); 153 } 154 155 void 156 soisdisconnected(so) 157 struct socket *so; 158 { 159 160 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 161 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 162 wakeup((caddr_t)&so->so_timeo); 163 sbdrop(&so->so_snd, so->so_snd.sb_cc); 164 sowwakeup(so); 165 sorwakeup(so); 166 } 167 168 /* 169 * When an attempt at a new connection is noted on a socket 170 * which accepts connections, sonewconn is called. If the 171 * connection is possible (subject to space constraints, etc.) 172 * then we allocate a new structure, propoerly linked into the 173 * data structure of the original socket, and return this. 174 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 175 */ 176 struct socket * 177 sonewconn(struct socket *head, int connstatus) 178 { 179 struct socket *so; 180 struct pru_attach_info ai; 181 182 if (head->so_qlen > 3 * head->so_qlimit / 2) 183 return ((struct socket *)0); 184 so = soalloc(0); 185 if (so == NULL) 186 return ((struct socket *)0); 187 if ((head->so_options & SO_ACCEPTFILTER) != 0) 188 connstatus = 0; 189 so->so_head = head; 190 so->so_type = head->so_type; 191 so->so_options = head->so_options &~ SO_ACCEPTCONN; 192 so->so_linger = head->so_linger; 193 so->so_state = head->so_state | SS_NOFDREF; 194 so->so_proto = head->so_proto; 195 so->so_timeo = head->so_timeo; 196 so->so_cred = crhold(head->so_cred); 197 ai.sb_rlimit = NULL; 198 ai.p_ucred = NULL; 199 ai.fd_rdir = NULL; /* jail code cruft XXX JH */ 200 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat, NULL) || 201 /* Directly call function since we're already at protocol level. */ 202 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, &ai)) { 203 sodealloc(so); 204 return ((struct socket *)0); 205 } 206 207 if (connstatus) { 208 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 209 so->so_state |= SS_COMP; 210 head->so_qlen++; 211 } else { 212 if (head->so_incqlen > head->so_qlimit) { 213 struct socket *sp; 214 sp = TAILQ_FIRST(&head->so_incomp); 215 (void) soabort(sp); 216 } 217 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 218 so->so_state |= SS_INCOMP; 219 head->so_incqlen++; 220 } 221 if (connstatus) { 222 sorwakeup(head); 223 wakeup((caddr_t)&head->so_timeo); 224 so->so_state |= connstatus; 225 } 226 return (so); 227 } 228 229 /* 230 * Socantsendmore indicates that no more data will be sent on the 231 * socket; it would normally be applied to a socket when the user 232 * informs the system that no more data is to be sent, by the protocol 233 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 234 * will be received, and will normally be applied to the socket by a 235 * protocol when it detects that the peer will send no more data. 236 * Data queued for reading in the socket may yet be read. 237 */ 238 239 void 240 socantsendmore(so) 241 struct socket *so; 242 { 243 244 so->so_state |= SS_CANTSENDMORE; 245 sowwakeup(so); 246 } 247 248 void 249 socantrcvmore(so) 250 struct socket *so; 251 { 252 253 so->so_state |= SS_CANTRCVMORE; 254 sorwakeup(so); 255 } 256 257 /* 258 * Wait for data to arrive at/drain from a socket buffer. 259 */ 260 int 261 sbwait(sb) 262 struct sockbuf *sb; 263 { 264 265 sb->sb_flags |= SB_WAIT; 266 return (tsleep((caddr_t)&sb->sb_cc, 267 ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), 268 "sbwait", 269 sb->sb_timeo)); 270 } 271 272 /* 273 * Lock a sockbuf already known to be locked; 274 * return any error returned from sleep (EINTR). 275 */ 276 int 277 sb_lock(sb) 278 struct sockbuf *sb; 279 { 280 int error; 281 282 while (sb->sb_flags & SB_LOCK) { 283 sb->sb_flags |= SB_WANT; 284 error = tsleep((caddr_t)&sb->sb_flags, 285 ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), 286 "sblock", 0); 287 if (error) 288 return (error); 289 } 290 sb->sb_flags |= SB_LOCK; 291 return (0); 292 } 293 294 /* 295 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 296 * via SIGIO if the socket has the SS_ASYNC flag set. 297 */ 298 void 299 sowakeup(so, sb) 300 struct socket *so; 301 struct sockbuf *sb; 302 { 303 struct selinfo *selinfo = &sb->sb_sel; 304 305 selwakeup(selinfo); 306 sb->sb_flags &= ~SB_SEL; 307 if (sb->sb_flags & SB_WAIT) { 308 sb->sb_flags &= ~SB_WAIT; 309 wakeup((caddr_t)&sb->sb_cc); 310 } 311 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 312 pgsigio(so->so_sigio, SIGIO, 0); 313 if (sb->sb_flags & SB_UPCALL) 314 (*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT); 315 if (sb->sb_flags & SB_AIO) 316 aio_swake(so, sb); 317 KNOTE(&selinfo->si_note, 0); 318 if (sb->sb_flags & SB_MEVENT) { 319 struct netmsg_so_notify *msg, *nmsg; 320 321 TAILQ_FOREACH_MUTABLE(msg, &selinfo->si_mlist, nm_list, nmsg) { 322 if (msg->nm_predicate((struct netmsg *)msg)) { 323 TAILQ_REMOVE(&selinfo->si_mlist, msg, nm_list); 324 lwkt_replymsg(&msg->nm_lmsg, 325 msg->nm_lmsg.ms_error); 326 } 327 } 328 if (TAILQ_EMPTY(&sb->sb_sel.si_mlist)) 329 sb->sb_flags &= ~SB_MEVENT; 330 } 331 } 332 333 /* 334 * Socket buffer (struct sockbuf) utility routines. 335 * 336 * Each socket contains two socket buffers: one for sending data and 337 * one for receiving data. Each buffer contains a queue of mbufs, 338 * information about the number of mbufs and amount of data in the 339 * queue, and other fields allowing select() statements and notification 340 * on data availability to be implemented. 341 * 342 * Data stored in a socket buffer is maintained as a list of records. 343 * Each record is a list of mbufs chained together with the m_next 344 * field. Records are chained together with the m_nextpkt field. The upper 345 * level routine soreceive() expects the following conventions to be 346 * observed when placing information in the receive buffer: 347 * 348 * 1. If the protocol requires each message be preceded by the sender's 349 * name, then a record containing that name must be present before 350 * any associated data (mbuf's must be of type MT_SONAME). 351 * 2. If the protocol supports the exchange of ``access rights'' (really 352 * just additional data associated with the message), and there are 353 * ``rights'' to be received, then a record containing this data 354 * should be present (mbuf's must be of type MT_RIGHTS). 355 * 3. If a name or rights record exists, then it must be followed by 356 * a data record, perhaps of zero length. 357 * 358 * Before using a new socket structure it is first necessary to reserve 359 * buffer space to the socket, by calling sbreserve(). This should commit 360 * some of the available buffer space in the system buffer pool for the 361 * socket (currently, it does nothing but enforce limits). The space 362 * should be released by calling sbrelease() when the socket is destroyed. 363 */ 364 365 int 366 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl) 367 { 368 if (sbreserve(&so->so_snd, sndcc, so, rl) == 0) 369 goto bad; 370 if (sbreserve(&so->so_rcv, rcvcc, so, rl) == 0) 371 goto bad2; 372 if (so->so_rcv.sb_lowat == 0) 373 so->so_rcv.sb_lowat = 1; 374 if (so->so_snd.sb_lowat == 0) 375 so->so_snd.sb_lowat = MCLBYTES; 376 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 377 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 378 return (0); 379 bad2: 380 sbrelease(&so->so_snd, so); 381 bad: 382 return (ENOBUFS); 383 } 384 385 static int 386 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 387 { 388 int error = 0; 389 u_long old_sb_max = sb_max; 390 391 error = SYSCTL_OUT(req, arg1, sizeof(int)); 392 if (error || !req->newptr) 393 return (error); 394 error = SYSCTL_IN(req, arg1, sizeof(int)); 395 if (error) 396 return (error); 397 if (sb_max < MSIZE + MCLBYTES) { 398 sb_max = old_sb_max; 399 return (EINVAL); 400 } 401 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 402 return (0); 403 } 404 405 /* 406 * Allot mbufs to a sockbuf. 407 * Attempt to scale mbmax so that mbcnt doesn't become limiting 408 * if buffering efficiency is near the normal case. 409 */ 410 int 411 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so, struct rlimit *rl) 412 { 413 414 /* 415 * rl will only be NULL when we're in an interrupt (eg, in tcp_input) 416 * or when called from netgraph (ie, ngd_attach) 417 */ 418 if (cc > sb_max_adj) 419 return (0); 420 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 421 rl ? rl->rlim_cur : RLIM_INFINITY)) { 422 return (0); 423 } 424 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 425 if (sb->sb_lowat > sb->sb_hiwat) 426 sb->sb_lowat = sb->sb_hiwat; 427 return (1); 428 } 429 430 /* 431 * Free mbufs held by a socket, and reserved mbuf space. 432 */ 433 void 434 sbrelease(sb, so) 435 struct sockbuf *sb; 436 struct socket *so; 437 { 438 439 sbflush(sb); 440 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 441 RLIM_INFINITY); 442 sb->sb_mbmax = 0; 443 } 444 445 /* 446 * Routines to add and remove 447 * data from an mbuf queue. 448 * 449 * The routines sbappend() or sbappendrecord() are normally called to 450 * append new mbufs to a socket buffer, after checking that adequate 451 * space is available, comparing the function sbspace() with the amount 452 * of data to be added. sbappendrecord() differs from sbappend() in 453 * that data supplied is treated as the beginning of a new record. 454 * To place a sender's address, optional access rights, and data in a 455 * socket receive buffer, sbappendaddr() should be used. To place 456 * access rights and data in a socket receive buffer, sbappendrights() 457 * should be used. In either case, the new data begins a new record. 458 * Note that unlike sbappend() and sbappendrecord(), these routines check 459 * for the caller that there will be enough space to store the data. 460 * Each fails if there is not enough space, or if it cannot find mbufs 461 * to store additional information in. 462 * 463 * Reliable protocols may use the socket send buffer to hold data 464 * awaiting acknowledgement. Data is normally copied from a socket 465 * send buffer in a protocol with m_copy for output to a peer, 466 * and then removing the data from the socket buffer with sbdrop() 467 * or sbdroprecord() when the data is acknowledged by the peer. 468 */ 469 470 /* 471 * Append mbuf chain m to the last record in the 472 * socket buffer sb. The additional space associated 473 * the mbuf chain is recorded in sb. Empty mbufs are 474 * discarded and mbufs are compacted where possible. 475 */ 476 void 477 sbappend(sb, m) 478 struct sockbuf *sb; 479 struct mbuf *m; 480 { 481 struct mbuf *n; 482 483 if (m == 0) 484 return; 485 n = sb->sb_mb; 486 if (n) { 487 while (n->m_nextpkt) 488 n = n->m_nextpkt; 489 do { 490 if (n->m_flags & M_EOR) { 491 sbappendrecord(sb, m); /* XXXXXX!!!! */ 492 return; 493 } 494 } while (n->m_next && (n = n->m_next)); 495 } 496 sbcompress(sb, m, n); 497 } 498 499 #ifdef SOCKBUF_DEBUG 500 void 501 sbcheck(sb) 502 struct sockbuf *sb; 503 { 504 struct mbuf *m; 505 struct mbuf *n = 0; 506 u_long len = 0, mbcnt = 0; 507 508 for (m = sb->sb_mb; m; m = n) { 509 n = m->m_nextpkt; 510 for (; m; m = m->m_next) { 511 len += m->m_len; 512 mbcnt += MSIZE; 513 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 514 mbcnt += m->m_ext.ext_size; 515 } 516 } 517 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 518 printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc, 519 mbcnt, sb->sb_mbcnt); 520 panic("sbcheck"); 521 } 522 } 523 #endif 524 525 /* 526 * As above, except the mbuf chain 527 * begins a new record. 528 */ 529 void 530 sbappendrecord(sb, m0) 531 struct sockbuf *sb; 532 struct mbuf *m0; 533 { 534 struct mbuf *m; 535 536 if (m0 == 0) 537 return; 538 m = sb->sb_mb; 539 if (m) 540 while (m->m_nextpkt) 541 m = m->m_nextpkt; 542 /* 543 * Put the first mbuf on the queue. 544 * Note this permits zero length records. 545 */ 546 sballoc(sb, m0); 547 if (m) 548 m->m_nextpkt = m0; 549 else 550 sb->sb_mb = m0; 551 m = m0->m_next; 552 m0->m_next = 0; 553 if (m && (m0->m_flags & M_EOR)) { 554 m0->m_flags &= ~M_EOR; 555 m->m_flags |= M_EOR; 556 } 557 sbcompress(sb, m, m0); 558 } 559 560 /* 561 * As above except that OOB data 562 * is inserted at the beginning of the sockbuf, 563 * but after any other OOB data. 564 */ 565 void 566 sbinsertoob(sb, m0) 567 struct sockbuf *sb; 568 struct mbuf *m0; 569 { 570 struct mbuf *m; 571 struct mbuf **mp; 572 573 if (m0 == 0) 574 return; 575 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) { 576 m = *mp; 577 again: 578 switch (m->m_type) { 579 580 case MT_OOBDATA: 581 continue; /* WANT next train */ 582 583 case MT_CONTROL: 584 m = m->m_next; 585 if (m) 586 goto again; /* inspect THIS train further */ 587 } 588 break; 589 } 590 /* 591 * Put the first mbuf on the queue. 592 * Note this permits zero length records. 593 */ 594 sballoc(sb, m0); 595 m0->m_nextpkt = *mp; 596 *mp = m0; 597 m = m0->m_next; 598 m0->m_next = 0; 599 if (m && (m0->m_flags & M_EOR)) { 600 m0->m_flags &= ~M_EOR; 601 m->m_flags |= M_EOR; 602 } 603 sbcompress(sb, m, m0); 604 } 605 606 /* 607 * Append address and data, and optionally, control (ancillary) data 608 * to the receive queue of a socket. If present, 609 * m0 must include a packet header with total length. 610 * Returns 0 if no space in sockbuf or insufficient mbufs. 611 */ 612 int 613 sbappendaddr(sb, asa, m0, control) 614 struct sockbuf *sb; 615 struct sockaddr *asa; 616 struct mbuf *m0, *control; 617 { 618 struct mbuf *m, *n; 619 int space = asa->sa_len; 620 621 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 622 panic("sbappendaddr"); 623 624 if (m0) 625 space += m0->m_pkthdr.len; 626 for (n = control; n; n = n->m_next) { 627 space += n->m_len; 628 if (n->m_next == 0) /* keep pointer to last control buf */ 629 break; 630 } 631 if (space > sbspace(sb)) 632 return (0); 633 if (asa->sa_len > MLEN) 634 return (0); 635 MGET(m, MB_DONTWAIT, MT_SONAME); 636 if (m == 0) 637 return (0); 638 m->m_len = asa->sa_len; 639 bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len); 640 if (n) 641 n->m_next = m0; /* concatenate data to control */ 642 else 643 control = m0; 644 m->m_next = control; 645 for (n = m; n; n = n->m_next) 646 sballoc(sb, n); 647 n = sb->sb_mb; 648 if (n) { 649 while (n->m_nextpkt) 650 n = n->m_nextpkt; 651 n->m_nextpkt = m; 652 } else 653 sb->sb_mb = m; 654 return (1); 655 } 656 657 int 658 sbappendcontrol(sb, m0, control) 659 struct sockbuf *sb; 660 struct mbuf *control, *m0; 661 { 662 struct mbuf *m, *n; 663 int space = 0; 664 665 if (control == 0) 666 panic("sbappendcontrol"); 667 for (m = control; ; m = m->m_next) { 668 space += m->m_len; 669 if (m->m_next == 0) 670 break; 671 } 672 n = m; /* save pointer to last control buffer */ 673 for (m = m0; m; m = m->m_next) 674 space += m->m_len; 675 if (space > sbspace(sb)) 676 return (0); 677 n->m_next = m0; /* concatenate data to control */ 678 for (m = control; m; m = m->m_next) 679 sballoc(sb, m); 680 n = sb->sb_mb; 681 if (n) { 682 while (n->m_nextpkt) 683 n = n->m_nextpkt; 684 n->m_nextpkt = control; 685 } else 686 sb->sb_mb = control; 687 return (1); 688 } 689 690 /* 691 * Compress mbuf chain m into the socket 692 * buffer sb following mbuf n. If n 693 * is null, the buffer is presumed empty. 694 */ 695 void 696 sbcompress(sb, m, n) 697 struct sockbuf *sb; 698 struct mbuf *m, *n; 699 { 700 int eor = 0; 701 struct mbuf *o; 702 703 while (m) { 704 eor |= m->m_flags & M_EOR; 705 if (m->m_len == 0 && 706 (eor == 0 || 707 (((o = m->m_next) || (o = n)) && 708 o->m_type == m->m_type))) { 709 m = m_free(m); 710 continue; 711 } 712 if (n && (n->m_flags & M_EOR) == 0 && 713 M_WRITABLE(n) && 714 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 715 m->m_len <= M_TRAILINGSPACE(n) && 716 n->m_type == m->m_type) { 717 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, 718 (unsigned)m->m_len); 719 n->m_len += m->m_len; 720 sb->sb_cc += m->m_len; 721 m = m_free(m); 722 continue; 723 } 724 if (n) 725 n->m_next = m; 726 else 727 sb->sb_mb = m; 728 sballoc(sb, m); 729 n = m; 730 m->m_flags &= ~M_EOR; 731 m = m->m_next; 732 n->m_next = 0; 733 } 734 if (eor) { 735 if (n) 736 n->m_flags |= eor; 737 else 738 printf("semi-panic: sbcompress"); 739 } 740 } 741 742 /* 743 * Free all mbufs in a sockbuf. 744 * Check that all resources are reclaimed. 745 */ 746 void 747 sbflush(sb) 748 struct sockbuf *sb; 749 { 750 751 if (sb->sb_flags & SB_LOCK) 752 panic("sbflush: locked"); 753 while (sb->sb_mbcnt) { 754 /* 755 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty: 756 * we would loop forever. Panic instead. 757 */ 758 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 759 break; 760 sbdrop(sb, (int)sb->sb_cc); 761 } 762 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt) 763 panic("sbflush: cc %ld || mb %p || mbcnt %ld", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt); 764 } 765 766 /* 767 * Drop data from (the front of) a sockbuf. 768 */ 769 void 770 sbdrop(sb, len) 771 struct sockbuf *sb; 772 int len; 773 { 774 struct mbuf *m; 775 struct mbuf *next; 776 777 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 778 while (len > 0) { 779 if (m == 0) { 780 if (next == 0) 781 panic("sbdrop"); 782 m = next; 783 next = m->m_nextpkt; 784 continue; 785 } 786 if (m->m_len > len) { 787 m->m_len -= len; 788 m->m_data += len; 789 sb->sb_cc -= len; 790 break; 791 } 792 len -= m->m_len; 793 sbfree(sb, m); 794 m = m_free(m); 795 } 796 while (m && m->m_len == 0) { 797 sbfree(sb, m); 798 m = m_free(m); 799 } 800 if (m) { 801 sb->sb_mb = m; 802 m->m_nextpkt = next; 803 } else 804 sb->sb_mb = next; 805 } 806 807 /* 808 * Drop a record off the front of a sockbuf 809 * and move the next record to the front. 810 */ 811 void 812 sbdroprecord(sb) 813 struct sockbuf *sb; 814 { 815 struct mbuf *m; 816 817 m = sb->sb_mb; 818 if (m) { 819 sb->sb_mb = m->m_nextpkt; 820 do { 821 sbfree(sb, m); 822 m = m_free(m); 823 } while (m); 824 } 825 } 826 827 /* 828 * Create a "control" mbuf containing the specified data 829 * with the specified type for presentation on a socket buffer. 830 */ 831 struct mbuf * 832 sbcreatecontrol(p, size, type, level) 833 caddr_t p; 834 int size; 835 int type, level; 836 { 837 struct cmsghdr *cp; 838 struct mbuf *m; 839 840 if (CMSG_SPACE((u_int)size) > MCLBYTES) 841 return ((struct mbuf *) NULL); 842 if ((m = m_get(MB_DONTWAIT, MT_CONTROL)) == NULL) 843 return ((struct mbuf *) NULL); 844 if (CMSG_SPACE((u_int)size) > MLEN) { 845 MCLGET(m, MB_DONTWAIT); 846 if ((m->m_flags & M_EXT) == 0) { 847 m_free(m); 848 return ((struct mbuf *) NULL); 849 } 850 } 851 cp = mtod(m, struct cmsghdr *); 852 m->m_len = 0; 853 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), 854 ("sbcreatecontrol: short mbuf")); 855 if (p != NULL) 856 (void)memcpy(CMSG_DATA(cp), p, size); 857 m->m_len = CMSG_SPACE(size); 858 cp->cmsg_len = CMSG_LEN(size); 859 cp->cmsg_level = level; 860 cp->cmsg_type = type; 861 return (m); 862 } 863 864 /* 865 * Some routines that return EOPNOTSUPP for entry points that are not 866 * supported by a protocol. Fill in as needed. 867 */ 868 int 869 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 870 { 871 return EOPNOTSUPP; 872 } 873 874 int 875 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 876 { 877 return EOPNOTSUPP; 878 } 879 880 int 881 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 882 { 883 return EOPNOTSUPP; 884 } 885 886 int 887 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 888 struct ifnet *ifp, struct thread *td) 889 { 890 return EOPNOTSUPP; 891 } 892 893 int 894 pru_listen_notsupp(struct socket *so, struct thread *td) 895 { 896 return EOPNOTSUPP; 897 } 898 899 int 900 pru_rcvd_notsupp(struct socket *so, int flags) 901 { 902 return EOPNOTSUPP; 903 } 904 905 int 906 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 907 { 908 return EOPNOTSUPP; 909 } 910 911 /* 912 * This isn't really a ``null'' operation, but it's the default one 913 * and doesn't do anything destructive. 914 */ 915 int 916 pru_sense_null(struct socket *so, struct stat *sb) 917 { 918 sb->st_blksize = so->so_snd.sb_hiwat; 919 return 0; 920 } 921 922 /* 923 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. Callers 924 * of this routine assume that it always succeeds, so we have to use a 925 * blockable allocation even though we might be called from a critical thread. 926 */ 927 struct sockaddr * 928 dup_sockaddr(struct sockaddr *sa) 929 { 930 struct sockaddr *sa2; 931 932 sa2 = malloc(sa->sa_len, M_SONAME, M_INTWAIT); 933 bcopy(sa, sa2, sa->sa_len); 934 return (sa2); 935 } 936 937 /* 938 * Create an external-format (``xsocket'') structure using the information 939 * in the kernel-format socket structure pointed to by so. This is done 940 * to reduce the spew of irrelevant information over this interface, 941 * to isolate user code from changes in the kernel structure, and 942 * potentially to provide information-hiding if we decide that 943 * some of this information should be hidden from users. 944 */ 945 void 946 sotoxsocket(struct socket *so, struct xsocket *xso) 947 { 948 xso->xso_len = sizeof *xso; 949 xso->xso_so = so; 950 xso->so_type = so->so_type; 951 xso->so_options = so->so_options; 952 xso->so_linger = so->so_linger; 953 xso->so_state = so->so_state; 954 xso->so_pcb = so->so_pcb; 955 xso->xso_protocol = so->so_proto->pr_protocol; 956 xso->xso_family = so->so_proto->pr_domain->dom_family; 957 xso->so_qlen = so->so_qlen; 958 xso->so_incqlen = so->so_incqlen; 959 xso->so_qlimit = so->so_qlimit; 960 xso->so_timeo = so->so_timeo; 961 xso->so_error = so->so_error; 962 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 963 xso->so_oobmark = so->so_oobmark; 964 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 965 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 966 xso->so_uid = so->so_cred->cr_uid; 967 } 968 969 /* 970 * This does the same for sockbufs. Note that the xsockbuf structure, 971 * since it is always embedded in a socket, does not include a self 972 * pointer nor a length. We make this entry point public in case 973 * some other mechanism needs it. 974 */ 975 void 976 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 977 { 978 xsb->sb_cc = sb->sb_cc; 979 xsb->sb_hiwat = sb->sb_hiwat; 980 xsb->sb_mbcnt = sb->sb_mbcnt; 981 xsb->sb_mbmax = sb->sb_mbmax; 982 xsb->sb_lowat = sb->sb_lowat; 983 xsb->sb_flags = sb->sb_flags; 984 xsb->sb_timeo = sb->sb_timeo; 985 } 986 987 /* 988 * Here is the definition of some of the basic objects in the kern.ipc 989 * branch of the MIB. 990 */ 991 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 992 993 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 994 static int dummy; 995 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 996 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, 997 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); 998 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, 999 &maxsockets, 0, "Maximum number of sockets avaliable"); 1000 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1001 &sb_efficiency, 0, ""); 1002 1003 /* 1004 * Initialise maxsockets 1005 */ 1006 static void init_maxsockets(void *ignored) 1007 { 1008 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 1009 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 1010 } 1011 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 1012