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 * $FreeBSD$ 35 */ 36 37 #include "opt_mac.h" 38 #include "opt_param.h" 39 40 #include <sys/param.h> 41 #include <sys/aio.h> /* for aio_swake proto */ 42 #include <sys/domain.h> 43 #include <sys/event.h> 44 #include <sys/file.h> /* for maxfiles */ 45 #include <sys/kernel.h> 46 #include <sys/lock.h> 47 #include <sys/mac.h> 48 #include <sys/malloc.h> 49 #include <sys/mbuf.h> 50 #include <sys/mutex.h> 51 #include <sys/proc.h> 52 #include <sys/protosw.h> 53 #include <sys/resourcevar.h> 54 #include <sys/signalvar.h> 55 #include <sys/socket.h> 56 #include <sys/socketvar.h> 57 #include <sys/stat.h> 58 #include <sys/sysctl.h> 59 #include <sys/systm.h> 60 61 int maxsockets; 62 63 void (*aio_swake)(struct socket *, struct sockbuf *); 64 65 /* 66 * Primitive routines for operating on sockets and socket buffers 67 */ 68 69 u_long sb_max = SB_MAX; 70 u_long sb_max_adj = 71 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 72 73 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 74 75 /* 76 * Procedures to manipulate state flags of socket 77 * and do appropriate wakeups. Normal sequence from the 78 * active (originating) side is that soisconnecting() is 79 * called during processing of connect() call, 80 * resulting in an eventual call to soisconnected() if/when the 81 * connection is established. When the connection is torn down 82 * soisdisconnecting() is called during processing of disconnect() call, 83 * and soisdisconnected() is called when the connection to the peer 84 * is totally severed. The semantics of these routines are such that 85 * connectionless protocols can call soisconnected() and soisdisconnected() 86 * only, bypassing the in-progress calls when setting up a ``connection'' 87 * takes no time. 88 * 89 * From the passive side, a socket is created with 90 * two queues of sockets: so_incomp for connections in progress 91 * and so_comp for connections already made and awaiting user acceptance. 92 * As a protocol is preparing incoming connections, it creates a socket 93 * structure queued on so_incomp by calling sonewconn(). When the connection 94 * is established, soisconnected() is called, and transfers the 95 * socket structure to so_comp, making it available to accept(). 96 * 97 * If a socket is closed with sockets on either 98 * so_incomp or so_comp, these sockets are dropped. 99 * 100 * If higher level protocols are implemented in 101 * the kernel, the wakeups done here will sometimes 102 * cause software-interrupt process scheduling. 103 */ 104 105 void 106 soisconnecting(so) 107 register struct socket *so; 108 { 109 110 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 111 so->so_state |= SS_ISCONNECTING; 112 } 113 114 void 115 soisconnected(so) 116 struct socket *so; 117 { 118 struct socket *head = so->so_head; 119 120 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 121 so->so_state |= SS_ISCONNECTED; 122 if (head && (so->so_state & SS_INCOMP)) { 123 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 124 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 125 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 126 so->so_rcv.sb_flags |= SB_UPCALL; 127 so->so_options &= ~SO_ACCEPTFILTER; 128 so->so_upcall(so, so->so_upcallarg, 0); 129 return; 130 } 131 TAILQ_REMOVE(&head->so_incomp, so, so_list); 132 head->so_incqlen--; 133 so->so_state &= ~SS_INCOMP; 134 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 135 head->so_qlen++; 136 so->so_state |= SS_COMP; 137 sorwakeup(head); 138 wakeup_one(&head->so_timeo); 139 } else { 140 wakeup(&so->so_timeo); 141 sorwakeup(so); 142 sowwakeup(so); 143 } 144 } 145 146 void 147 soisdisconnecting(so) 148 register struct socket *so; 149 { 150 151 so->so_state &= ~SS_ISCONNECTING; 152 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 153 wakeup(&so->so_timeo); 154 sowwakeup(so); 155 sorwakeup(so); 156 } 157 158 void 159 soisdisconnected(so) 160 register struct socket *so; 161 { 162 163 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 164 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 165 wakeup(&so->so_timeo); 166 sbdrop(&so->so_snd, so->so_snd.sb_cc); 167 sowwakeup(so); 168 sorwakeup(so); 169 } 170 171 /* 172 * When an attempt at a new connection is noted on a socket 173 * which accepts connections, sonewconn is called. If the 174 * connection is possible (subject to space constraints, etc.) 175 * then we allocate a new structure, propoerly linked into the 176 * data structure of the original socket, and return this. 177 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 178 * 179 * note: the ref count on the socket is 0 on return 180 */ 181 struct socket * 182 sonewconn(head, connstatus) 183 register struct socket *head; 184 int connstatus; 185 { 186 register struct socket *so; 187 188 if (head->so_qlen > 3 * head->so_qlimit / 2) 189 return ((struct socket *)0); 190 so = soalloc(0); 191 if (so == NULL) 192 return ((struct socket *)0); 193 if ((head->so_options & SO_ACCEPTFILTER) != 0) 194 connstatus = 0; 195 so->so_head = head; 196 so->so_type = head->so_type; 197 so->so_options = head->so_options &~ SO_ACCEPTCONN; 198 so->so_linger = head->so_linger; 199 so->so_state = head->so_state | SS_NOFDREF; 200 so->so_proto = head->so_proto; 201 so->so_timeo = head->so_timeo; 202 so->so_cred = crhold(head->so_cred); 203 #ifdef MAC 204 mac_create_socket_from_socket(head, so); 205 #endif 206 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) || 207 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 208 sotryfree(so); 209 return ((struct socket *)0); 210 } 211 212 if (connstatus) { 213 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 214 so->so_state |= SS_COMP; 215 head->so_qlen++; 216 } else { 217 if (head->so_incqlen > head->so_qlimit) { 218 struct socket *sp; 219 sp = TAILQ_FIRST(&head->so_incomp); 220 (void) soabort(sp); 221 } 222 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 223 so->so_state |= SS_INCOMP; 224 head->so_incqlen++; 225 } 226 if (connstatus) { 227 sorwakeup(head); 228 wakeup(&head->so_timeo); 229 so->so_state |= connstatus; 230 } 231 return (so); 232 } 233 234 /* 235 * Socantsendmore indicates that no more data will be sent on the 236 * socket; it would normally be applied to a socket when the user 237 * informs the system that no more data is to be sent, by the protocol 238 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 239 * will be received, and will normally be applied to the socket by a 240 * protocol when it detects that the peer will send no more data. 241 * Data queued for reading in the socket may yet be read. 242 */ 243 244 void 245 socantsendmore(so) 246 struct socket *so; 247 { 248 249 so->so_state |= SS_CANTSENDMORE; 250 sowwakeup(so); 251 } 252 253 void 254 socantrcvmore(so) 255 struct socket *so; 256 { 257 258 so->so_state |= SS_CANTRCVMORE; 259 sorwakeup(so); 260 } 261 262 /* 263 * Wait for data to arrive at/drain from a socket buffer. 264 */ 265 int 266 sbwait(sb) 267 struct sockbuf *sb; 268 { 269 270 sb->sb_flags |= SB_WAIT; 271 return (tsleep(&sb->sb_cc, 272 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", 273 sb->sb_timeo)); 274 } 275 276 /* 277 * Lock a sockbuf already known to be locked; 278 * return any error returned from sleep (EINTR). 279 */ 280 int 281 sb_lock(sb) 282 register struct sockbuf *sb; 283 { 284 int error; 285 286 while (sb->sb_flags & SB_LOCK) { 287 sb->sb_flags |= SB_WANT; 288 error = tsleep(&sb->sb_flags, 289 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH, 290 "sblock", 0); 291 if (error) 292 return (error); 293 } 294 sb->sb_flags |= SB_LOCK; 295 return (0); 296 } 297 298 /* 299 * Wakeup processes waiting on a socket buffer. 300 * Do asynchronous notification via SIGIO 301 * if the socket has the SS_ASYNC flag set. 302 */ 303 void 304 sowakeup(so, sb) 305 register struct socket *so; 306 register struct sockbuf *sb; 307 { 308 309 selwakeup(&sb->sb_sel); 310 sb->sb_flags &= ~SB_SEL; 311 if (sb->sb_flags & SB_WAIT) { 312 sb->sb_flags &= ~SB_WAIT; 313 wakeup(&sb->sb_cc); 314 } 315 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 316 pgsigio(&so->so_sigio, SIGIO, 0); 317 if (sb->sb_flags & SB_UPCALL) 318 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); 319 if (sb->sb_flags & SB_AIO) 320 aio_swake(so, sb); 321 KNOTE(&sb->sb_sel.si_note, 0); 322 } 323 324 /* 325 * Socket buffer (struct sockbuf) utility routines. 326 * 327 * Each socket contains two socket buffers: one for sending data and 328 * one for receiving data. Each buffer contains a queue of mbufs, 329 * information about the number of mbufs and amount of data in the 330 * queue, and other fields allowing select() statements and notification 331 * on data availability to be implemented. 332 * 333 * Data stored in a socket buffer is maintained as a list of records. 334 * Each record is a list of mbufs chained together with the m_next 335 * field. Records are chained together with the m_nextpkt field. The upper 336 * level routine soreceive() expects the following conventions to be 337 * observed when placing information in the receive buffer: 338 * 339 * 1. If the protocol requires each message be preceded by the sender's 340 * name, then a record containing that name must be present before 341 * any associated data (mbuf's must be of type MT_SONAME). 342 * 2. If the protocol supports the exchange of ``access rights'' (really 343 * just additional data associated with the message), and there are 344 * ``rights'' to be received, then a record containing this data 345 * should be present (mbuf's must be of type MT_RIGHTS). 346 * 3. If a name or rights record exists, then it must be followed by 347 * a data record, perhaps of zero length. 348 * 349 * Before using a new socket structure it is first necessary to reserve 350 * buffer space to the socket, by calling sbreserve(). This should commit 351 * some of the available buffer space in the system buffer pool for the 352 * socket (currently, it does nothing but enforce limits). The space 353 * should be released by calling sbrelease() when the socket is destroyed. 354 */ 355 356 int 357 soreserve(so, sndcc, rcvcc) 358 register struct socket *so; 359 u_long sndcc, rcvcc; 360 { 361 struct thread *td = curthread; 362 363 if (sbreserve(&so->so_snd, sndcc, so, td) == 0) 364 goto bad; 365 if (sbreserve(&so->so_rcv, rcvcc, so, td) == 0) 366 goto bad2; 367 if (so->so_rcv.sb_lowat == 0) 368 so->so_rcv.sb_lowat = 1; 369 if (so->so_snd.sb_lowat == 0) 370 so->so_snd.sb_lowat = MCLBYTES; 371 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 372 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 373 return (0); 374 bad2: 375 sbrelease(&so->so_snd, so); 376 bad: 377 return (ENOBUFS); 378 } 379 380 static int 381 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 382 { 383 int error = 0; 384 u_long old_sb_max = sb_max; 385 386 error = SYSCTL_OUT(req, arg1, sizeof(int)); 387 if (error || !req->newptr) 388 return (error); 389 error = SYSCTL_IN(req, arg1, sizeof(int)); 390 if (error) 391 return (error); 392 if (sb_max < MSIZE + MCLBYTES) { 393 sb_max = old_sb_max; 394 return (EINVAL); 395 } 396 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 397 return (0); 398 } 399 400 /* 401 * Allot mbufs to a sockbuf. 402 * Attempt to scale mbmax so that mbcnt doesn't become limiting 403 * if buffering efficiency is near the normal case. 404 */ 405 int 406 sbreserve(sb, cc, so, td) 407 struct sockbuf *sb; 408 u_long cc; 409 struct socket *so; 410 struct thread *td; 411 { 412 413 /* 414 * td will only be NULL when we're in an interrupt 415 * (e.g. in tcp_input()) 416 */ 417 if (cc > sb_max_adj) 418 return (0); 419 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 420 td ? td->td_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur : RLIM_INFINITY)) { 421 return (0); 422 } 423 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 424 if (sb->sb_lowat > sb->sb_hiwat) 425 sb->sb_lowat = sb->sb_hiwat; 426 return (1); 427 } 428 429 /* 430 * Free mbufs held by a socket, and reserved mbuf space. 431 */ 432 void 433 sbrelease(sb, so) 434 struct sockbuf *sb; 435 struct socket *so; 436 { 437 438 sbflush(sb); 439 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 440 RLIM_INFINITY); 441 sb->sb_mbmax = 0; 442 } 443 444 /* 445 * Routines to add and remove 446 * data from an mbuf queue. 447 * 448 * The routines sbappend() or sbappendrecord() are normally called to 449 * append new mbufs to a socket buffer, after checking that adequate 450 * space is available, comparing the function sbspace() with the amount 451 * of data to be added. sbappendrecord() differs from sbappend() in 452 * that data supplied is treated as the beginning of a new record. 453 * To place a sender's address, optional access rights, and data in a 454 * socket receive buffer, sbappendaddr() should be used. To place 455 * access rights and data in a socket receive buffer, sbappendrights() 456 * should be used. In either case, the new data begins a new record. 457 * Note that unlike sbappend() and sbappendrecord(), these routines check 458 * for the caller that there will be enough space to store the data. 459 * Each fails if there is not enough space, or if it cannot find mbufs 460 * to store additional information in. 461 * 462 * Reliable protocols may use the socket send buffer to hold data 463 * awaiting acknowledgement. Data is normally copied from a socket 464 * send buffer in a protocol with m_copy for output to a peer, 465 * and then removing the data from the socket buffer with sbdrop() 466 * or sbdroprecord() when the data is acknowledged by the peer. 467 */ 468 469 /* 470 * Append mbuf chain m to the last record in the 471 * socket buffer sb. The additional space associated 472 * the mbuf chain is recorded in sb. Empty mbufs are 473 * discarded and mbufs are compacted where possible. 474 */ 475 void 476 sbappend(sb, m) 477 struct sockbuf *sb; 478 struct mbuf *m; 479 { 480 register struct mbuf *n; 481 482 if (m == 0) 483 return; 484 n = sb->sb_mb; 485 if (n) { 486 while (n->m_nextpkt) 487 n = n->m_nextpkt; 488 do { 489 if (n->m_flags & M_EOR) { 490 sbappendrecord(sb, m); /* XXXXXX!!!! */ 491 return; 492 } 493 } while (n->m_next && (n = n->m_next)); 494 } 495 sbcompress(sb, m, n); 496 } 497 498 #ifdef SOCKBUF_DEBUG 499 void 500 sbcheck(sb) 501 register struct sockbuf *sb; 502 { 503 register struct mbuf *m; 504 register struct mbuf *n = 0; 505 register u_long len = 0, mbcnt = 0; 506 507 for (m = sb->sb_mb; m; m = n) { 508 n = m->m_nextpkt; 509 for (; m; m = m->m_next) { 510 len += m->m_len; 511 mbcnt += MSIZE; 512 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 513 mbcnt += m->m_ext.ext_size; 514 } 515 } 516 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 517 printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc, 518 mbcnt, sb->sb_mbcnt); 519 panic("sbcheck"); 520 } 521 } 522 #endif 523 524 /* 525 * As above, except the mbuf chain 526 * begins a new record. 527 */ 528 void 529 sbappendrecord(sb, m0) 530 register struct sockbuf *sb; 531 register struct mbuf *m0; 532 { 533 register struct mbuf *m; 534 535 if (m0 == 0) 536 return; 537 m = sb->sb_mb; 538 if (m) 539 while (m->m_nextpkt) 540 m = m->m_nextpkt; 541 /* 542 * Put the first mbuf on the queue. 543 * Note this permits zero length records. 544 */ 545 sballoc(sb, m0); 546 if (m) 547 m->m_nextpkt = m0; 548 else 549 sb->sb_mb = m0; 550 m = m0->m_next; 551 m0->m_next = 0; 552 if (m && (m0->m_flags & M_EOR)) { 553 m0->m_flags &= ~M_EOR; 554 m->m_flags |= M_EOR; 555 } 556 sbcompress(sb, m, m0); 557 } 558 559 /* 560 * As above except that OOB data 561 * is inserted at the beginning of the sockbuf, 562 * but after any other OOB data. 563 */ 564 void 565 sbinsertoob(sb, m0) 566 register struct sockbuf *sb; 567 register struct mbuf *m0; 568 { 569 register struct mbuf *m; 570 register struct mbuf **mp; 571 572 if (m0 == 0) 573 return; 574 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) { 575 m = *mp; 576 again: 577 switch (m->m_type) { 578 579 case MT_OOBDATA: 580 continue; /* WANT next train */ 581 582 case MT_CONTROL: 583 m = m->m_next; 584 if (m) 585 goto again; /* inspect THIS train further */ 586 } 587 break; 588 } 589 /* 590 * Put the first mbuf on the queue. 591 * Note this permits zero length records. 592 */ 593 sballoc(sb, m0); 594 m0->m_nextpkt = *mp; 595 *mp = m0; 596 m = m0->m_next; 597 m0->m_next = 0; 598 if (m && (m0->m_flags & M_EOR)) { 599 m0->m_flags &= ~M_EOR; 600 m->m_flags |= M_EOR; 601 } 602 sbcompress(sb, m, m0); 603 } 604 605 /* 606 * Append address and data, and optionally, control (ancillary) data 607 * to the receive queue of a socket. If present, 608 * m0 must include a packet header with total length. 609 * Returns 0 if no space in sockbuf or insufficient mbufs. 610 */ 611 int 612 sbappendaddr(sb, asa, m0, control) 613 register struct sockbuf *sb; 614 struct sockaddr *asa; 615 struct mbuf *m0, *control; 616 { 617 register struct mbuf *m, *n; 618 int space = asa->sa_len; 619 620 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 621 panic("sbappendaddr"); 622 if (m0) 623 space += m0->m_pkthdr.len; 624 for (n = control; n; n = n->m_next) { 625 space += n->m_len; 626 if (n->m_next == 0) /* keep pointer to last control buf */ 627 break; 628 } 629 if (space > sbspace(sb)) 630 return (0); 631 if (asa->sa_len > MLEN) 632 return (0); 633 MGET(m, M_DONTWAIT, MT_SONAME); 634 if (m == 0) 635 return (0); 636 m->m_len = asa->sa_len; 637 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 638 if (n) 639 n->m_next = m0; /* concatenate data to control */ 640 else 641 control = m0; 642 m->m_next = control; 643 for (n = m; n; n = n->m_next) 644 sballoc(sb, n); 645 n = sb->sb_mb; 646 if (n) { 647 while (n->m_nextpkt) 648 n = n->m_nextpkt; 649 n->m_nextpkt = m; 650 } else 651 sb->sb_mb = m; 652 return (1); 653 } 654 655 int 656 sbappendcontrol(sb, m0, control) 657 struct sockbuf *sb; 658 struct mbuf *control, *m0; 659 { 660 register struct mbuf *m, *n; 661 int space = 0; 662 663 if (control == 0) 664 panic("sbappendcontrol"); 665 for (m = control; ; m = m->m_next) { 666 space += m->m_len; 667 if (m->m_next == 0) 668 break; 669 } 670 n = m; /* save pointer to last control buffer */ 671 for (m = m0; m; m = m->m_next) 672 space += m->m_len; 673 if (space > sbspace(sb)) 674 return (0); 675 n->m_next = m0; /* concatenate data to control */ 676 for (m = control; m; m = m->m_next) 677 sballoc(sb, m); 678 n = sb->sb_mb; 679 if (n) { 680 while (n->m_nextpkt) 681 n = n->m_nextpkt; 682 n->m_nextpkt = control; 683 } else 684 sb->sb_mb = control; 685 return (1); 686 } 687 688 /* 689 * Compress mbuf chain m into the socket 690 * buffer sb following mbuf n. If n 691 * is null, the buffer is presumed empty. 692 */ 693 void 694 sbcompress(sb, m, n) 695 register struct sockbuf *sb; 696 register struct mbuf *m, *n; 697 { 698 register int eor = 0; 699 register struct mbuf *o; 700 701 while (m) { 702 eor |= m->m_flags & M_EOR; 703 if (m->m_len == 0 && 704 (eor == 0 || 705 (((o = m->m_next) || (o = n)) && 706 o->m_type == m->m_type))) { 707 m = m_free(m); 708 continue; 709 } 710 if (n && (n->m_flags & M_EOR) == 0 && 711 M_WRITABLE(n) && 712 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 713 m->m_len <= M_TRAILINGSPACE(n) && 714 n->m_type == m->m_type) { 715 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, 716 (unsigned)m->m_len); 717 n->m_len += m->m_len; 718 sb->sb_cc += m->m_len; 719 m = m_free(m); 720 continue; 721 } 722 if (n) 723 n->m_next = m; 724 else 725 sb->sb_mb = m; 726 sballoc(sb, m); 727 n = m; 728 m->m_flags &= ~M_EOR; 729 m = m->m_next; 730 n->m_next = 0; 731 } 732 if (eor) { 733 if (n) 734 n->m_flags |= eor; 735 else 736 printf("semi-panic: sbcompress\n"); 737 } 738 } 739 740 /* 741 * Free all mbufs in a sockbuf. 742 * Check that all resources are reclaimed. 743 */ 744 void 745 sbflush(sb) 746 register struct sockbuf *sb; 747 { 748 749 if (sb->sb_flags & SB_LOCK) 750 panic("sbflush: locked"); 751 while (sb->sb_mbcnt) { 752 /* 753 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty: 754 * we would loop forever. Panic instead. 755 */ 756 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 757 break; 758 sbdrop(sb, (int)sb->sb_cc); 759 } 760 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt) 761 panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt); 762 } 763 764 /* 765 * Drop data from (the front of) a sockbuf. 766 */ 767 void 768 sbdrop(sb, len) 769 register struct sockbuf *sb; 770 register int len; 771 { 772 register struct mbuf *m; 773 struct mbuf *next; 774 775 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 776 while (len > 0) { 777 if (m == 0) { 778 if (next == 0) 779 panic("sbdrop"); 780 m = next; 781 next = m->m_nextpkt; 782 continue; 783 } 784 if (m->m_len > len) { 785 m->m_len -= len; 786 m->m_data += len; 787 sb->sb_cc -= len; 788 break; 789 } 790 len -= m->m_len; 791 sbfree(sb, m); 792 m = m_free(m); 793 } 794 while (m && m->m_len == 0) { 795 sbfree(sb, m); 796 m = m_free(m); 797 } 798 if (m) { 799 sb->sb_mb = m; 800 m->m_nextpkt = next; 801 } else 802 sb->sb_mb = next; 803 } 804 805 /* 806 * Drop a record off the front of a sockbuf 807 * and move the next record to the front. 808 */ 809 void 810 sbdroprecord(sb) 811 register struct sockbuf *sb; 812 { 813 register struct mbuf *m; 814 815 m = sb->sb_mb; 816 if (m) { 817 sb->sb_mb = m->m_nextpkt; 818 do { 819 sbfree(sb, m); 820 m = m_free(m); 821 } while (m); 822 } 823 } 824 825 /* 826 * Create a "control" mbuf containing the specified data 827 * with the specified type for presentation on a socket buffer. 828 */ 829 struct mbuf * 830 sbcreatecontrol(p, size, type, level) 831 caddr_t p; 832 register int size; 833 int type, level; 834 { 835 register struct cmsghdr *cp; 836 struct mbuf *m; 837 838 if (CMSG_SPACE((u_int)size) > MCLBYTES) 839 return ((struct mbuf *) NULL); 840 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL) 841 return ((struct mbuf *) NULL); 842 if (CMSG_SPACE((u_int)size) > MLEN) { 843 MCLGET(m, M_DONTWAIT); 844 if ((m->m_flags & M_EXT) == 0) { 845 m_free(m); 846 return ((struct mbuf *) NULL); 847 } 848 } 849 cp = mtod(m, struct cmsghdr *); 850 m->m_len = 0; 851 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), 852 ("sbcreatecontrol: short mbuf")); 853 if (p != NULL) 854 (void)memcpy(CMSG_DATA(cp), p, size); 855 m->m_len = CMSG_SPACE(size); 856 cp->cmsg_len = CMSG_LEN(size); 857 cp->cmsg_level = level; 858 cp->cmsg_type = type; 859 return (m); 860 } 861 862 /* 863 * Some routines that return EOPNOTSUPP for entry points that are not 864 * supported by a protocol. Fill in as needed. 865 */ 866 int 867 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 868 { 869 return EOPNOTSUPP; 870 } 871 872 int 873 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 874 { 875 return EOPNOTSUPP; 876 } 877 878 int 879 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 880 { 881 return EOPNOTSUPP; 882 } 883 884 int 885 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 886 struct ifnet *ifp, struct thread *td) 887 { 888 return EOPNOTSUPP; 889 } 890 891 int 892 pru_listen_notsupp(struct socket *so, struct thread *td) 893 { 894 return EOPNOTSUPP; 895 } 896 897 int 898 pru_rcvd_notsupp(struct socket *so, int flags) 899 { 900 return EOPNOTSUPP; 901 } 902 903 int 904 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 905 { 906 return EOPNOTSUPP; 907 } 908 909 /* 910 * This isn't really a ``null'' operation, but it's the default one 911 * and doesn't do anything destructive. 912 */ 913 int 914 pru_sense_null(struct socket *so, struct stat *sb) 915 { 916 sb->st_blksize = so->so_snd.sb_hiwat; 917 return 0; 918 } 919 920 /* 921 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 922 */ 923 struct sockaddr * 924 dup_sockaddr(sa, canwait) 925 struct sockaddr *sa; 926 int canwait; 927 { 928 struct sockaddr *sa2; 929 930 MALLOC(sa2, struct sockaddr *, sa->sa_len, M_SONAME, 931 canwait ? M_WAITOK : M_NOWAIT); 932 if (sa2) 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