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 1.6 2003/08/28 01:48:18 hmp 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 int maxsockets; 58 59 /* 60 * Primitive routines for operating on sockets and socket buffers 61 */ 62 63 u_long sb_max = SB_MAX; 64 u_long sb_max_adj = 65 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 66 67 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 68 69 /* 70 * Procedures to manipulate state flags of socket 71 * and do appropriate wakeups. Normal sequence from the 72 * active (originating) side is that soisconnecting() is 73 * called during processing of connect() call, 74 * resulting in an eventual call to soisconnected() if/when the 75 * connection is established. When the connection is torn down 76 * soisdisconnecting() is called during processing of disconnect() call, 77 * and soisdisconnected() is called when the connection to the peer 78 * is totally severed. The semantics of these routines are such that 79 * connectionless protocols can call soisconnected() and soisdisconnected() 80 * only, bypassing the in-progress calls when setting up a ``connection'' 81 * takes no time. 82 * 83 * From the passive side, a socket is created with 84 * two queues of sockets: so_incomp for connections in progress 85 * and so_comp for connections already made and awaiting user acceptance. 86 * As a protocol is preparing incoming connections, it creates a socket 87 * structure queued on so_incomp by calling sonewconn(). When the connection 88 * is established, soisconnected() is called, and transfers the 89 * socket structure to so_comp, making it available to accept(). 90 * 91 * If a socket is closed with sockets on either 92 * so_incomp or so_comp, these sockets are dropped. 93 * 94 * If higher level protocols are implemented in 95 * the kernel, the wakeups done here will sometimes 96 * cause software-interrupt process scheduling. 97 */ 98 99 void 100 soisconnecting(so) 101 struct socket *so; 102 { 103 104 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 105 so->so_state |= SS_ISCONNECTING; 106 } 107 108 void 109 soisconnected(so) 110 struct socket *so; 111 { 112 struct socket *head = so->so_head; 113 114 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 115 so->so_state |= SS_ISCONNECTED; 116 if (head && (so->so_state & SS_INCOMP)) { 117 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 118 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 119 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 120 so->so_rcv.sb_flags |= SB_UPCALL; 121 so->so_options &= ~SO_ACCEPTFILTER; 122 so->so_upcall(so, so->so_upcallarg, 0); 123 return; 124 } 125 TAILQ_REMOVE(&head->so_incomp, so, so_list); 126 head->so_incqlen--; 127 so->so_state &= ~SS_INCOMP; 128 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 129 head->so_qlen++; 130 so->so_state |= SS_COMP; 131 sorwakeup(head); 132 wakeup_one(&head->so_timeo); 133 } else { 134 wakeup(&so->so_timeo); 135 sorwakeup(so); 136 sowwakeup(so); 137 } 138 } 139 140 void 141 soisdisconnecting(so) 142 struct socket *so; 143 { 144 145 so->so_state &= ~SS_ISCONNECTING; 146 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 147 wakeup((caddr_t)&so->so_timeo); 148 sowwakeup(so); 149 sorwakeup(so); 150 } 151 152 void 153 soisdisconnected(so) 154 struct socket *so; 155 { 156 157 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 158 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 159 wakeup((caddr_t)&so->so_timeo); 160 sbdrop(&so->so_snd, so->so_snd.sb_cc); 161 sowwakeup(so); 162 sorwakeup(so); 163 } 164 165 /* 166 * When an attempt at a new connection is noted on a socket 167 * which accepts connections, sonewconn is called. If the 168 * connection is possible (subject to space constraints, etc.) 169 * then we allocate a new structure, propoerly linked into the 170 * data structure of the original socket, and return this. 171 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 172 */ 173 struct socket * 174 sonewconn(struct socket *head, int connstatus) 175 { 176 struct socket *so; 177 178 if (head->so_qlen > 3 * head->so_qlimit / 2) 179 return ((struct socket *)0); 180 so = soalloc(0); 181 if (so == NULL) 182 return ((struct socket *)0); 183 if ((head->so_options & SO_ACCEPTFILTER) != 0) 184 connstatus = 0; 185 so->so_head = head; 186 so->so_type = head->so_type; 187 so->so_options = head->so_options &~ SO_ACCEPTCONN; 188 so->so_linger = head->so_linger; 189 so->so_state = head->so_state | SS_NOFDREF; 190 so->so_proto = head->so_proto; 191 so->so_timeo = head->so_timeo; 192 so->so_cred = crhold(head->so_cred); 193 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) || 194 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 195 sodealloc(so); 196 return ((struct socket *)0); 197 } 198 199 if (connstatus) { 200 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 201 so->so_state |= SS_COMP; 202 head->so_qlen++; 203 } else { 204 if (head->so_incqlen > head->so_qlimit) { 205 struct socket *sp; 206 sp = TAILQ_FIRST(&head->so_incomp); 207 (void) soabort(sp); 208 } 209 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 210 so->so_state |= SS_INCOMP; 211 head->so_incqlen++; 212 } 213 if (connstatus) { 214 sorwakeup(head); 215 wakeup((caddr_t)&head->so_timeo); 216 so->so_state |= connstatus; 217 } 218 return (so); 219 } 220 221 /* 222 * Socantsendmore indicates that no more data will be sent on the 223 * socket; it would normally be applied to a socket when the user 224 * informs the system that no more data is to be sent, by the protocol 225 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 226 * will be received, and will normally be applied to the socket by a 227 * protocol when it detects that the peer will send no more data. 228 * Data queued for reading in the socket may yet be read. 229 */ 230 231 void 232 socantsendmore(so) 233 struct socket *so; 234 { 235 236 so->so_state |= SS_CANTSENDMORE; 237 sowwakeup(so); 238 } 239 240 void 241 socantrcvmore(so) 242 struct socket *so; 243 { 244 245 so->so_state |= SS_CANTRCVMORE; 246 sorwakeup(so); 247 } 248 249 /* 250 * Wait for data to arrive at/drain from a socket buffer. 251 */ 252 int 253 sbwait(sb) 254 struct sockbuf *sb; 255 { 256 257 sb->sb_flags |= SB_WAIT; 258 return (tsleep((caddr_t)&sb->sb_cc, 259 ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), 260 "sbwait", 261 sb->sb_timeo)); 262 } 263 264 /* 265 * Lock a sockbuf already known to be locked; 266 * return any error returned from sleep (EINTR). 267 */ 268 int 269 sb_lock(sb) 270 struct sockbuf *sb; 271 { 272 int error; 273 274 while (sb->sb_flags & SB_LOCK) { 275 sb->sb_flags |= SB_WANT; 276 error = tsleep((caddr_t)&sb->sb_flags, 277 ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), 278 "sblock", 0); 279 if (error) 280 return (error); 281 } 282 sb->sb_flags |= SB_LOCK; 283 return (0); 284 } 285 286 /* 287 * Wakeup processes waiting on a socket buffer. 288 * Do asynchronous notification via SIGIO 289 * if the socket has the SS_ASYNC flag set. 290 */ 291 void 292 sowakeup(so, sb) 293 struct socket *so; 294 struct sockbuf *sb; 295 { 296 selwakeup(&sb->sb_sel); 297 sb->sb_flags &= ~SB_SEL; 298 if (sb->sb_flags & SB_WAIT) { 299 sb->sb_flags &= ~SB_WAIT; 300 wakeup((caddr_t)&sb->sb_cc); 301 } 302 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 303 pgsigio(so->so_sigio, SIGIO, 0); 304 if (sb->sb_flags & SB_UPCALL) 305 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); 306 if (sb->sb_flags & SB_AIO) 307 aio_swake(so, sb); 308 KNOTE(&sb->sb_sel.si_note, 0); 309 } 310 311 /* 312 * Socket buffer (struct sockbuf) utility routines. 313 * 314 * Each socket contains two socket buffers: one for sending data and 315 * one for receiving data. Each buffer contains a queue of mbufs, 316 * information about the number of mbufs and amount of data in the 317 * queue, and other fields allowing select() statements and notification 318 * on data availability to be implemented. 319 * 320 * Data stored in a socket buffer is maintained as a list of records. 321 * Each record is a list of mbufs chained together with the m_next 322 * field. Records are chained together with the m_nextpkt field. The upper 323 * level routine soreceive() expects the following conventions to be 324 * observed when placing information in the receive buffer: 325 * 326 * 1. If the protocol requires each message be preceded by the sender's 327 * name, then a record containing that name must be present before 328 * any associated data (mbuf's must be of type MT_SONAME). 329 * 2. If the protocol supports the exchange of ``access rights'' (really 330 * just additional data associated with the message), and there are 331 * ``rights'' to be received, then a record containing this data 332 * should be present (mbuf's must be of type MT_RIGHTS). 333 * 3. If a name or rights record exists, then it must be followed by 334 * a data record, perhaps of zero length. 335 * 336 * Before using a new socket structure it is first necessary to reserve 337 * buffer space to the socket, by calling sbreserve(). This should commit 338 * some of the available buffer space in the system buffer pool for the 339 * socket (currently, it does nothing but enforce limits). The space 340 * should be released by calling sbrelease() when the socket is destroyed. 341 */ 342 343 int 344 soreserve(so, sndcc, rcvcc) 345 struct socket *so; 346 u_long sndcc, rcvcc; 347 { 348 struct proc *p = curproc; 349 350 if (sbreserve(&so->so_snd, sndcc, so, p) == 0) 351 goto bad; 352 if (sbreserve(&so->so_rcv, rcvcc, so, p) == 0) 353 goto bad2; 354 if (so->so_rcv.sb_lowat == 0) 355 so->so_rcv.sb_lowat = 1; 356 if (so->so_snd.sb_lowat == 0) 357 so->so_snd.sb_lowat = MCLBYTES; 358 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 359 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 360 return (0); 361 bad2: 362 sbrelease(&so->so_snd, so); 363 bad: 364 return (ENOBUFS); 365 } 366 367 static int 368 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 369 { 370 int error = 0; 371 u_long old_sb_max = sb_max; 372 373 error = SYSCTL_OUT(req, arg1, sizeof(int)); 374 if (error || !req->newptr) 375 return (error); 376 error = SYSCTL_IN(req, arg1, sizeof(int)); 377 if (error) 378 return (error); 379 if (sb_max < MSIZE + MCLBYTES) { 380 sb_max = old_sb_max; 381 return (EINVAL); 382 } 383 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 384 return (0); 385 } 386 387 /* 388 * Allot mbufs to a sockbuf. 389 * Attempt to scale mbmax so that mbcnt doesn't become limiting 390 * if buffering efficiency is near the normal case. 391 */ 392 int 393 sbreserve(sb, cc, so, p) 394 struct sockbuf *sb; 395 u_long cc; 396 struct socket *so; 397 struct proc *p; 398 { 399 400 /* 401 * p will only be NULL when we're in an interrupt 402 * (e.g. in tcp_input()) 403 */ 404 if (cc > sb_max_adj) 405 return (0); 406 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 407 p ? p->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) { 408 return (0); 409 } 410 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 411 if (sb->sb_lowat > sb->sb_hiwat) 412 sb->sb_lowat = sb->sb_hiwat; 413 return (1); 414 } 415 416 /* 417 * Free mbufs held by a socket, and reserved mbuf space. 418 */ 419 void 420 sbrelease(sb, so) 421 struct sockbuf *sb; 422 struct socket *so; 423 { 424 425 sbflush(sb); 426 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 427 RLIM_INFINITY); 428 sb->sb_mbmax = 0; 429 } 430 431 /* 432 * Routines to add and remove 433 * data from an mbuf queue. 434 * 435 * The routines sbappend() or sbappendrecord() are normally called to 436 * append new mbufs to a socket buffer, after checking that adequate 437 * space is available, comparing the function sbspace() with the amount 438 * of data to be added. sbappendrecord() differs from sbappend() in 439 * that data supplied is treated as the beginning of a new record. 440 * To place a sender's address, optional access rights, and data in a 441 * socket receive buffer, sbappendaddr() should be used. To place 442 * access rights and data in a socket receive buffer, sbappendrights() 443 * should be used. In either case, the new data begins a new record. 444 * Note that unlike sbappend() and sbappendrecord(), these routines check 445 * for the caller that there will be enough space to store the data. 446 * Each fails if there is not enough space, or if it cannot find mbufs 447 * to store additional information in. 448 * 449 * Reliable protocols may use the socket send buffer to hold data 450 * awaiting acknowledgement. Data is normally copied from a socket 451 * send buffer in a protocol with m_copy for output to a peer, 452 * and then removing the data from the socket buffer with sbdrop() 453 * or sbdroprecord() when the data is acknowledged by the peer. 454 */ 455 456 /* 457 * Append mbuf chain m to the last record in the 458 * socket buffer sb. The additional space associated 459 * the mbuf chain is recorded in sb. Empty mbufs are 460 * discarded and mbufs are compacted where possible. 461 */ 462 void 463 sbappend(sb, m) 464 struct sockbuf *sb; 465 struct mbuf *m; 466 { 467 struct mbuf *n; 468 469 if (m == 0) 470 return; 471 n = sb->sb_mb; 472 if (n) { 473 while (n->m_nextpkt) 474 n = n->m_nextpkt; 475 do { 476 if (n->m_flags & M_EOR) { 477 sbappendrecord(sb, m); /* XXXXXX!!!! */ 478 return; 479 } 480 } while (n->m_next && (n = n->m_next)); 481 } 482 sbcompress(sb, m, n); 483 } 484 485 #ifdef SOCKBUF_DEBUG 486 void 487 sbcheck(sb) 488 struct sockbuf *sb; 489 { 490 struct mbuf *m; 491 struct mbuf *n = 0; 492 u_long len = 0, mbcnt = 0; 493 494 for (m = sb->sb_mb; m; m = n) { 495 n = m->m_nextpkt; 496 for (; m; m = m->m_next) { 497 len += m->m_len; 498 mbcnt += MSIZE; 499 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 500 mbcnt += m->m_ext.ext_size; 501 } 502 } 503 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 504 printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc, 505 mbcnt, sb->sb_mbcnt); 506 panic("sbcheck"); 507 } 508 } 509 #endif 510 511 /* 512 * As above, except the mbuf chain 513 * begins a new record. 514 */ 515 void 516 sbappendrecord(sb, m0) 517 struct sockbuf *sb; 518 struct mbuf *m0; 519 { 520 struct mbuf *m; 521 522 if (m0 == 0) 523 return; 524 m = sb->sb_mb; 525 if (m) 526 while (m->m_nextpkt) 527 m = m->m_nextpkt; 528 /* 529 * Put the first mbuf on the queue. 530 * Note this permits zero length records. 531 */ 532 sballoc(sb, m0); 533 if (m) 534 m->m_nextpkt = m0; 535 else 536 sb->sb_mb = m0; 537 m = m0->m_next; 538 m0->m_next = 0; 539 if (m && (m0->m_flags & M_EOR)) { 540 m0->m_flags &= ~M_EOR; 541 m->m_flags |= M_EOR; 542 } 543 sbcompress(sb, m, m0); 544 } 545 546 /* 547 * As above except that OOB data 548 * is inserted at the beginning of the sockbuf, 549 * but after any other OOB data. 550 */ 551 void 552 sbinsertoob(sb, m0) 553 struct sockbuf *sb; 554 struct mbuf *m0; 555 { 556 struct mbuf *m; 557 struct mbuf **mp; 558 559 if (m0 == 0) 560 return; 561 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) { 562 m = *mp; 563 again: 564 switch (m->m_type) { 565 566 case MT_OOBDATA: 567 continue; /* WANT next train */ 568 569 case MT_CONTROL: 570 m = m->m_next; 571 if (m) 572 goto again; /* inspect THIS train further */ 573 } 574 break; 575 } 576 /* 577 * Put the first mbuf on the queue. 578 * Note this permits zero length records. 579 */ 580 sballoc(sb, m0); 581 m0->m_nextpkt = *mp; 582 *mp = m0; 583 m = m0->m_next; 584 m0->m_next = 0; 585 if (m && (m0->m_flags & M_EOR)) { 586 m0->m_flags &= ~M_EOR; 587 m->m_flags |= M_EOR; 588 } 589 sbcompress(sb, m, m0); 590 } 591 592 /* 593 * Append address and data, and optionally, control (ancillary) data 594 * to the receive queue of a socket. If present, 595 * m0 must include a packet header with total length. 596 * Returns 0 if no space in sockbuf or insufficient mbufs. 597 */ 598 int 599 sbappendaddr(sb, asa, m0, control) 600 struct sockbuf *sb; 601 struct sockaddr *asa; 602 struct mbuf *m0, *control; 603 { 604 struct mbuf *m, *n; 605 int space = asa->sa_len; 606 607 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 608 panic("sbappendaddr"); 609 610 if (m0) 611 space += m0->m_pkthdr.len; 612 for (n = control; n; n = n->m_next) { 613 space += n->m_len; 614 if (n->m_next == 0) /* keep pointer to last control buf */ 615 break; 616 } 617 if (space > sbspace(sb)) 618 return (0); 619 if (asa->sa_len > MLEN) 620 return (0); 621 MGET(m, M_DONTWAIT, MT_SONAME); 622 if (m == 0) 623 return (0); 624 m->m_len = asa->sa_len; 625 bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len); 626 if (n) 627 n->m_next = m0; /* concatenate data to control */ 628 else 629 control = m0; 630 m->m_next = control; 631 for (n = m; n; n = n->m_next) 632 sballoc(sb, n); 633 n = sb->sb_mb; 634 if (n) { 635 while (n->m_nextpkt) 636 n = n->m_nextpkt; 637 n->m_nextpkt = m; 638 } else 639 sb->sb_mb = m; 640 return (1); 641 } 642 643 int 644 sbappendcontrol(sb, m0, control) 645 struct sockbuf *sb; 646 struct mbuf *control, *m0; 647 { 648 struct mbuf *m, *n; 649 int space = 0; 650 651 if (control == 0) 652 panic("sbappendcontrol"); 653 for (m = control; ; m = m->m_next) { 654 space += m->m_len; 655 if (m->m_next == 0) 656 break; 657 } 658 n = m; /* save pointer to last control buffer */ 659 for (m = m0; m; m = m->m_next) 660 space += m->m_len; 661 if (space > sbspace(sb)) 662 return (0); 663 n->m_next = m0; /* concatenate data to control */ 664 for (m = control; m; m = m->m_next) 665 sballoc(sb, m); 666 n = sb->sb_mb; 667 if (n) { 668 while (n->m_nextpkt) 669 n = n->m_nextpkt; 670 n->m_nextpkt = control; 671 } else 672 sb->sb_mb = control; 673 return (1); 674 } 675 676 /* 677 * Compress mbuf chain m into the socket 678 * buffer sb following mbuf n. If n 679 * is null, the buffer is presumed empty. 680 */ 681 void 682 sbcompress(sb, m, n) 683 struct sockbuf *sb; 684 struct mbuf *m, *n; 685 { 686 int eor = 0; 687 struct mbuf *o; 688 689 while (m) { 690 eor |= m->m_flags & M_EOR; 691 if (m->m_len == 0 && 692 (eor == 0 || 693 (((o = m->m_next) || (o = n)) && 694 o->m_type == m->m_type))) { 695 m = m_free(m); 696 continue; 697 } 698 if (n && (n->m_flags & M_EOR) == 0 && 699 M_WRITABLE(n) && 700 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 701 m->m_len <= M_TRAILINGSPACE(n) && 702 n->m_type == m->m_type) { 703 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, 704 (unsigned)m->m_len); 705 n->m_len += m->m_len; 706 sb->sb_cc += m->m_len; 707 m = m_free(m); 708 continue; 709 } 710 if (n) 711 n->m_next = m; 712 else 713 sb->sb_mb = m; 714 sballoc(sb, m); 715 n = m; 716 m->m_flags &= ~M_EOR; 717 m = m->m_next; 718 n->m_next = 0; 719 } 720 if (eor) { 721 if (n) 722 n->m_flags |= eor; 723 else 724 printf("semi-panic: sbcompress\n"); 725 } 726 } 727 728 /* 729 * Free all mbufs in a sockbuf. 730 * Check that all resources are reclaimed. 731 */ 732 void 733 sbflush(sb) 734 struct sockbuf *sb; 735 { 736 737 if (sb->sb_flags & SB_LOCK) 738 panic("sbflush: locked"); 739 while (sb->sb_mbcnt) { 740 /* 741 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty: 742 * we would loop forever. Panic instead. 743 */ 744 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 745 break; 746 sbdrop(sb, (int)sb->sb_cc); 747 } 748 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt) 749 panic("sbflush: cc %ld || mb %p || mbcnt %ld", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt); 750 } 751 752 /* 753 * Drop data from (the front of) a sockbuf. 754 */ 755 void 756 sbdrop(sb, len) 757 struct sockbuf *sb; 758 int len; 759 { 760 struct mbuf *m; 761 struct mbuf *next; 762 763 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 764 while (len > 0) { 765 if (m == 0) { 766 if (next == 0) 767 panic("sbdrop"); 768 m = next; 769 next = m->m_nextpkt; 770 continue; 771 } 772 if (m->m_len > len) { 773 m->m_len -= len; 774 m->m_data += len; 775 sb->sb_cc -= len; 776 break; 777 } 778 len -= m->m_len; 779 sbfree(sb, m); 780 m = m_free(m); 781 } 782 while (m && m->m_len == 0) { 783 sbfree(sb, m); 784 m = m_free(m); 785 } 786 if (m) { 787 sb->sb_mb = m; 788 m->m_nextpkt = next; 789 } else 790 sb->sb_mb = next; 791 } 792 793 /* 794 * Drop a record off the front of a sockbuf 795 * and move the next record to the front. 796 */ 797 void 798 sbdroprecord(sb) 799 struct sockbuf *sb; 800 { 801 struct mbuf *m; 802 803 m = sb->sb_mb; 804 if (m) { 805 sb->sb_mb = m->m_nextpkt; 806 do { 807 sbfree(sb, m); 808 m = m_free(m); 809 } while (m); 810 } 811 } 812 813 /* 814 * Create a "control" mbuf containing the specified data 815 * with the specified type for presentation on a socket buffer. 816 */ 817 struct mbuf * 818 sbcreatecontrol(p, size, type, level) 819 caddr_t p; 820 int size; 821 int type, level; 822 { 823 struct cmsghdr *cp; 824 struct mbuf *m; 825 826 if (CMSG_SPACE((u_int)size) > MCLBYTES) 827 return ((struct mbuf *) NULL); 828 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL) 829 return ((struct mbuf *) NULL); 830 if (CMSG_SPACE((u_int)size) > MLEN) { 831 MCLGET(m, M_DONTWAIT); 832 if ((m->m_flags & M_EXT) == 0) { 833 m_free(m); 834 return ((struct mbuf *) NULL); 835 } 836 } 837 cp = mtod(m, struct cmsghdr *); 838 m->m_len = 0; 839 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), 840 ("sbcreatecontrol: short mbuf")); 841 if (p != NULL) 842 (void)memcpy(CMSG_DATA(cp), p, size); 843 m->m_len = CMSG_SPACE(size); 844 cp->cmsg_len = CMSG_LEN(size); 845 cp->cmsg_level = level; 846 cp->cmsg_type = type; 847 return (m); 848 } 849 850 /* 851 * Some routines that return EOPNOTSUPP for entry points that are not 852 * supported by a protocol. Fill in as needed. 853 */ 854 int 855 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 856 { 857 return EOPNOTSUPP; 858 } 859 860 int 861 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 862 { 863 return EOPNOTSUPP; 864 } 865 866 int 867 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 868 { 869 return EOPNOTSUPP; 870 } 871 872 int 873 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 874 struct ifnet *ifp, struct thread *td) 875 { 876 return EOPNOTSUPP; 877 } 878 879 int 880 pru_listen_notsupp(struct socket *so, struct thread *td) 881 { 882 return EOPNOTSUPP; 883 } 884 885 int 886 pru_rcvd_notsupp(struct socket *so, int flags) 887 { 888 return EOPNOTSUPP; 889 } 890 891 int 892 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 893 { 894 return EOPNOTSUPP; 895 } 896 897 /* 898 * This isn't really a ``null'' operation, but it's the default one 899 * and doesn't do anything destructive. 900 */ 901 int 902 pru_sense_null(struct socket *so, struct stat *sb) 903 { 904 sb->st_blksize = so->so_snd.sb_hiwat; 905 return 0; 906 } 907 908 /* 909 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 910 */ 911 struct sockaddr * 912 dup_sockaddr(sa, canwait) 913 struct sockaddr *sa; 914 int canwait; 915 { 916 struct sockaddr *sa2; 917 918 MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME, 919 canwait ? M_WAITOK : M_NOWAIT); 920 if (sa2) 921 bcopy(sa, sa2, sa->sa_len); 922 return sa2; 923 } 924 925 /* 926 * Create an external-format (``xsocket'') structure using the information 927 * in the kernel-format socket structure pointed to by so. This is done 928 * to reduce the spew of irrelevant information over this interface, 929 * to isolate user code from changes in the kernel structure, and 930 * potentially to provide information-hiding if we decide that 931 * some of this information should be hidden from users. 932 */ 933 void 934 sotoxsocket(struct socket *so, struct xsocket *xso) 935 { 936 xso->xso_len = sizeof *xso; 937 xso->xso_so = so; 938 xso->so_type = so->so_type; 939 xso->so_options = so->so_options; 940 xso->so_linger = so->so_linger; 941 xso->so_state = so->so_state; 942 xso->so_pcb = so->so_pcb; 943 xso->xso_protocol = so->so_proto->pr_protocol; 944 xso->xso_family = so->so_proto->pr_domain->dom_family; 945 xso->so_qlen = so->so_qlen; 946 xso->so_incqlen = so->so_incqlen; 947 xso->so_qlimit = so->so_qlimit; 948 xso->so_timeo = so->so_timeo; 949 xso->so_error = so->so_error; 950 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 951 xso->so_oobmark = so->so_oobmark; 952 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 953 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 954 xso->so_uid = so->so_cred->cr_uid; 955 } 956 957 /* 958 * This does the same for sockbufs. Note that the xsockbuf structure, 959 * since it is always embedded in a socket, does not include a self 960 * pointer nor a length. We make this entry point public in case 961 * some other mechanism needs it. 962 */ 963 void 964 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 965 { 966 xsb->sb_cc = sb->sb_cc; 967 xsb->sb_hiwat = sb->sb_hiwat; 968 xsb->sb_mbcnt = sb->sb_mbcnt; 969 xsb->sb_mbmax = sb->sb_mbmax; 970 xsb->sb_lowat = sb->sb_lowat; 971 xsb->sb_flags = sb->sb_flags; 972 xsb->sb_timeo = sb->sb_timeo; 973 } 974 975 /* 976 * Here is the definition of some of the basic objects in the kern.ipc 977 * branch of the MIB. 978 */ 979 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 980 981 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 982 static int dummy; 983 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 984 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, 985 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); 986 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, 987 &maxsockets, 0, "Maximum number of sockets avaliable"); 988 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 989 &sb_efficiency, 0, ""); 990 991 /* 992 * Initialise maxsockets 993 */ 994 static void init_maxsockets(void *ignored) 995 { 996 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 997 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 998 } 999 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 1000