uipc_socket2.c revision 1.47
1/* $NetBSD: uipc_socket2.c,v 1.47 2002/09/27 15:37:47 provos Exp $ */ 2 3/* 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)uipc_socket2.c 8.2 (Berkeley) 2/14/95 36 */ 37 38#include <sys/cdefs.h> 39__KERNEL_RCSID(0, "$NetBSD: uipc_socket2.c,v 1.47 2002/09/27 15:37:47 provos Exp $"); 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/proc.h> 44#include <sys/file.h> 45#include <sys/buf.h> 46#include <sys/malloc.h> 47#include <sys/mbuf.h> 48#include <sys/protosw.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/signalvar.h> 52 53/* 54 * Primitive routines for operating on sockets and socket buffers 55 */ 56 57/* strings for sleep message: */ 58const char netcon[] = "netcon"; 59const char netcls[] = "netcls"; 60const char netio[] = "netio"; 61const char netlck[] = "netlck"; 62 63/* 64 * Procedures to manipulate state flags of socket 65 * and do appropriate wakeups. Normal sequence from the 66 * active (originating) side is that soisconnecting() is 67 * called during processing of connect() call, 68 * resulting in an eventual call to soisconnected() if/when the 69 * connection is established. When the connection is torn down 70 * soisdisconnecting() is called during processing of disconnect() call, 71 * and soisdisconnected() is called when the connection to the peer 72 * is totally severed. The semantics of these routines are such that 73 * connectionless protocols can call soisconnected() and soisdisconnected() 74 * only, bypassing the in-progress calls when setting up a ``connection'' 75 * takes no time. 76 * 77 * From the passive side, a socket is created with 78 * two queues of sockets: so_q0 for connections in progress 79 * and so_q for connections already made and awaiting user acceptance. 80 * As a protocol is preparing incoming connections, it creates a socket 81 * structure queued on so_q0 by calling sonewconn(). When the connection 82 * is established, soisconnected() is called, and transfers the 83 * socket structure to so_q, making it available to accept(). 84 * 85 * If a socket is closed with sockets on either 86 * so_q0 or so_q, these sockets are dropped. 87 * 88 * If higher level protocols are implemented in 89 * the kernel, the wakeups done here will sometimes 90 * cause software-interrupt process scheduling. 91 */ 92 93void 94soisconnecting(struct socket *so) 95{ 96 97 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 98 so->so_state |= SS_ISCONNECTING; 99} 100 101void 102soisconnected(struct socket *so) 103{ 104 struct socket *head; 105 106 head = so->so_head; 107 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 108 so->so_state |= SS_ISCONNECTED; 109 if (head && soqremque(so, 0)) { 110 soqinsque(head, so, 1); 111 sorwakeup(head); 112 wakeup((caddr_t)&head->so_timeo); 113 } else { 114 wakeup((caddr_t)&so->so_timeo); 115 sorwakeup(so); 116 sowwakeup(so); 117 } 118} 119 120void 121soisdisconnecting(struct socket *so) 122{ 123 124 so->so_state &= ~SS_ISCONNECTING; 125 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 126 wakeup((caddr_t)&so->so_timeo); 127 sowwakeup(so); 128 sorwakeup(so); 129} 130 131void 132soisdisconnected(struct socket *so) 133{ 134 135 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 136 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 137 wakeup((caddr_t)&so->so_timeo); 138 sowwakeup(so); 139 sorwakeup(so); 140} 141 142/* 143 * When an attempt at a new connection is noted on a socket 144 * which accepts connections, sonewconn is called. If the 145 * connection is possible (subject to space constraints, etc.) 146 * then we allocate a new structure, propoerly linked into the 147 * data structure of the original socket, and return this. 148 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 149 * 150 * Currently, sonewconn() is defined as sonewconn1() in socketvar.h 151 * to catch calls that are missing the (new) second parameter. 152 */ 153struct socket * 154sonewconn1(struct socket *head, int connstatus) 155{ 156 struct socket *so; 157 int soqueue; 158 159 soqueue = connstatus ? 1 : 0; 160 if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2) 161 return ((struct socket *)0); 162 so = pool_get(&socket_pool, PR_NOWAIT); 163 if (so == NULL) 164 return (NULL); 165 memset((caddr_t)so, 0, sizeof(*so)); 166 so->so_type = head->so_type; 167 so->so_options = head->so_options &~ SO_ACCEPTCONN; 168 so->so_linger = head->so_linger; 169 so->so_state = head->so_state | SS_NOFDREF; 170 so->so_proto = head->so_proto; 171 so->so_timeo = head->so_timeo; 172 so->so_pgid = head->so_pgid; 173 so->so_send = head->so_send; 174 so->so_receive = head->so_receive; 175 so->so_uid = head->so_uid; 176 (void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat); 177 soqinsque(head, so, soqueue); 178 if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH, 179 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 180 (struct proc *)0)) { 181 (void) soqremque(so, soqueue); 182 pool_put(&socket_pool, so); 183 return (NULL); 184 } 185 if (connstatus) { 186 sorwakeup(head); 187 wakeup((caddr_t)&head->so_timeo); 188 so->so_state |= connstatus; 189 } 190 return (so); 191} 192 193void 194soqinsque(struct socket *head, struct socket *so, int q) 195{ 196 197#ifdef DIAGNOSTIC 198 if (so->so_onq != NULL) 199 panic("soqinsque"); 200#endif 201 202 so->so_head = head; 203 if (q == 0) { 204 head->so_q0len++; 205 so->so_onq = &head->so_q0; 206 } else { 207 head->so_qlen++; 208 so->so_onq = &head->so_q; 209 } 210 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe); 211} 212 213int 214soqremque(struct socket *so, int q) 215{ 216 struct socket *head; 217 218 head = so->so_head; 219 if (q == 0) { 220 if (so->so_onq != &head->so_q0) 221 return (0); 222 head->so_q0len--; 223 } else { 224 if (so->so_onq != &head->so_q) 225 return (0); 226 head->so_qlen--; 227 } 228 TAILQ_REMOVE(so->so_onq, so, so_qe); 229 so->so_onq = NULL; 230 so->so_head = NULL; 231 return (1); 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 244void 245socantsendmore(struct socket *so) 246{ 247 248 so->so_state |= SS_CANTSENDMORE; 249 sowwakeup(so); 250} 251 252void 253socantrcvmore(struct socket *so) 254{ 255 256 so->so_state |= SS_CANTRCVMORE; 257 sorwakeup(so); 258} 259 260/* 261 * Wait for data to arrive at/drain from a socket buffer. 262 */ 263int 264sbwait(struct sockbuf *sb) 265{ 266 267 sb->sb_flags |= SB_WAIT; 268 return (tsleep((caddr_t)&sb->sb_cc, 269 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio, 270 sb->sb_timeo)); 271} 272 273/* 274 * Lock a sockbuf already known to be locked; 275 * return any error returned from sleep (EINTR). 276 */ 277int 278sb_lock(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) ? PSOCK : PSOCK|PCATCH, 286 netlck, 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. 296 * Do asynchronous notification via SIGIO 297 * if the socket buffer has the SB_ASYNC flag set. 298 */ 299void 300sowakeup(struct socket *so, struct sockbuf *sb) 301{ 302 struct proc *p; 303 304 selwakeup(&sb->sb_sel); 305 sb->sb_flags &= ~SB_SEL; 306 if (sb->sb_flags & SB_WAIT) { 307 sb->sb_flags &= ~SB_WAIT; 308 wakeup((caddr_t)&sb->sb_cc); 309 } 310 if (sb->sb_flags & SB_ASYNC) { 311 if (so->so_pgid < 0) 312 gsignal(-so->so_pgid, SIGIO); 313 else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0) 314 psignal(p, SIGIO); 315 } 316 if (sb->sb_flags & SB_UPCALL) 317 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); 318} 319 320/* 321 * Socket buffer (struct sockbuf) utility routines. 322 * 323 * Each socket contains two socket buffers: one for sending data and 324 * one for receiving data. Each buffer contains a queue of mbufs, 325 * information about the number of mbufs and amount of data in the 326 * queue, and other fields allowing poll() statements and notification 327 * on data availability to be implemented. 328 * 329 * Data stored in a socket buffer is maintained as a list of records. 330 * Each record is a list of mbufs chained together with the m_next 331 * field. Records are chained together with the m_nextpkt field. The upper 332 * level routine soreceive() expects the following conventions to be 333 * observed when placing information in the receive buffer: 334 * 335 * 1. If the protocol requires each message be preceded by the sender's 336 * name, then a record containing that name must be present before 337 * any associated data (mbuf's must be of type MT_SONAME). 338 * 2. If the protocol supports the exchange of ``access rights'' (really 339 * just additional data associated with the message), and there are 340 * ``rights'' to be received, then a record containing this data 341 * should be present (mbuf's must be of type MT_CONTROL). 342 * 3. If a name or rights record exists, then it must be followed by 343 * a data record, perhaps of zero length. 344 * 345 * Before using a new socket structure it is first necessary to reserve 346 * buffer space to the socket, by calling sbreserve(). This should commit 347 * some of the available buffer space in the system buffer pool for the 348 * socket (currently, it does nothing but enforce limits). The space 349 * should be released by calling sbrelease() when the socket is destroyed. 350 */ 351 352int 353soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 354{ 355 356 if (sbreserve(&so->so_snd, sndcc) == 0) 357 goto bad; 358 if (sbreserve(&so->so_rcv, rcvcc) == 0) 359 goto bad2; 360 if (so->so_rcv.sb_lowat == 0) 361 so->so_rcv.sb_lowat = 1; 362 if (so->so_snd.sb_lowat == 0) 363 so->so_snd.sb_lowat = MCLBYTES; 364 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 365 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 366 return (0); 367 bad2: 368 sbrelease(&so->so_snd); 369 bad: 370 return (ENOBUFS); 371} 372 373/* 374 * Allot mbufs to a sockbuf. 375 * Attempt to scale mbmax so that mbcnt doesn't become limiting 376 * if buffering efficiency is near the normal case. 377 */ 378int 379sbreserve(struct sockbuf *sb, u_long cc) 380{ 381 382 if (cc == 0 || 383 (u_quad_t) cc > (u_quad_t) sb_max * MCLBYTES / (MSIZE + MCLBYTES)) 384 return (0); 385 sb->sb_hiwat = cc; 386 sb->sb_mbmax = min(cc * 2, sb_max); 387 if (sb->sb_lowat > sb->sb_hiwat) 388 sb->sb_lowat = sb->sb_hiwat; 389 return (1); 390} 391 392/* 393 * Free mbufs held by a socket, and reserved mbuf space. 394 */ 395void 396sbrelease(struct sockbuf *sb) 397{ 398 399 sbflush(sb); 400 sb->sb_hiwat = sb->sb_mbmax = 0; 401} 402 403/* 404 * Routines to add and remove 405 * data from an mbuf queue. 406 * 407 * The routines sbappend() or sbappendrecord() are normally called to 408 * append new mbufs to a socket buffer, after checking that adequate 409 * space is available, comparing the function sbspace() with the amount 410 * of data to be added. sbappendrecord() differs from sbappend() in 411 * that data supplied is treated as the beginning of a new record. 412 * To place a sender's address, optional access rights, and data in a 413 * socket receive buffer, sbappendaddr() should be used. To place 414 * access rights and data in a socket receive buffer, sbappendrights() 415 * should be used. In either case, the new data begins a new record. 416 * Note that unlike sbappend() and sbappendrecord(), these routines check 417 * for the caller that there will be enough space to store the data. 418 * Each fails if there is not enough space, or if it cannot find mbufs 419 * to store additional information in. 420 * 421 * Reliable protocols may use the socket send buffer to hold data 422 * awaiting acknowledgement. Data is normally copied from a socket 423 * send buffer in a protocol with m_copy for output to a peer, 424 * and then removing the data from the socket buffer with sbdrop() 425 * or sbdroprecord() when the data is acknowledged by the peer. 426 */ 427 428#ifdef SOCKBUF_DEBUG 429void 430sblastrecordchk(struct sockbuf *sb, const char *where) 431{ 432 struct mbuf *m = sb->sb_mb; 433 434 while (m && m->m_nextpkt) 435 m = m->m_nextpkt; 436 437 if (m != sb->sb_lastrecord) { 438 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n", 439 sb->sb_mb, sb->sb_lastrecord, m); 440 printf("packet chain:\n"); 441 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 442 printf("\t%p\n", m); 443 panic("sblastrecordchk from %s", where); 444 } 445} 446 447void 448sblastmbufchk(struct sockbuf *sb, const char *where) 449{ 450 struct mbuf *m = sb->sb_mb; 451 struct mbuf *n; 452 453 while (m && m->m_nextpkt) 454 m = m->m_nextpkt; 455 456 while (m && m->m_next) 457 m = m->m_next; 458 459 if (m != sb->sb_mbtail) { 460 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n", 461 sb->sb_mb, sb->sb_mbtail, m); 462 printf("packet tree:\n"); 463 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 464 printf("\t"); 465 for (n = m; n != NULL; n = n->m_next) 466 printf("%p ", n); 467 printf("\n"); 468 } 469 panic("sblastmbufchk from %s", where); 470 } 471} 472#endif /* SOCKBUF_DEBUG */ 473 474#define SBLINKRECORD(sb, m0) \ 475do { \ 476 if ((sb)->sb_lastrecord != NULL) \ 477 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 478 else \ 479 (sb)->sb_mb = (m0); \ 480 (sb)->sb_lastrecord = (m0); \ 481} while (/*CONSTCOND*/0) 482 483/* 484 * Append mbuf chain m to the last record in the 485 * socket buffer sb. The additional space associated 486 * the mbuf chain is recorded in sb. Empty mbufs are 487 * discarded and mbufs are compacted where possible. 488 */ 489void 490sbappend(struct sockbuf *sb, struct mbuf *m) 491{ 492 struct mbuf *n; 493 494 if (m == 0) 495 return; 496 497 SBLASTRECORDCHK(sb, "sbappend 1"); 498 499 if ((n = sb->sb_lastrecord) != NULL) { 500 /* 501 * XXX Would like to simply use sb_mbtail here, but 502 * XXX I need to verify that I won't miss an EOR that 503 * XXX way. 504 */ 505 do { 506 if (n->m_flags & M_EOR) { 507 sbappendrecord(sb, m); /* XXXXXX!!!! */ 508 return; 509 } 510 } while (n->m_next && (n = n->m_next)); 511 } else { 512 /* 513 * If this is the first record in the socket buffer, it's 514 * also the last record. 515 */ 516 sb->sb_lastrecord = m; 517 } 518 sbcompress(sb, m, n); 519 SBLASTRECORDCHK(sb, "sbappend 2"); 520} 521 522/* 523 * This version of sbappend() should only be used when the caller 524 * absolutely knows that there will never be more than one record 525 * in the socket buffer, that is, a stream protocol (such as TCP). 526 */ 527void 528sbappendstream(struct sockbuf *sb, struct mbuf *m) 529{ 530 531 KDASSERT(m->m_nextpkt == NULL); 532 KASSERT(sb->sb_mb == sb->sb_lastrecord); 533 534 SBLASTMBUFCHK(sb, __func__); 535 536 sbcompress(sb, m, sb->sb_mbtail); 537 538 sb->sb_lastrecord = sb->sb_mb; 539 SBLASTRECORDCHK(sb, __func__); 540} 541 542#ifdef SOCKBUF_DEBUG 543void 544sbcheck(struct sockbuf *sb) 545{ 546 struct mbuf *m; 547 u_long len, mbcnt; 548 549 len = 0; 550 mbcnt = 0; 551 for (m = sb->sb_mb; m; m = m->m_next) { 552 len += m->m_len; 553 mbcnt += MSIZE; 554 if (m->m_flags & M_EXT) 555 mbcnt += m->m_ext.ext_size; 556 if (m->m_nextpkt) 557 panic("sbcheck nextpkt"); 558 } 559 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 560 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc, 561 mbcnt, sb->sb_mbcnt); 562 panic("sbcheck"); 563 } 564} 565#endif 566 567/* 568 * As above, except the mbuf chain 569 * begins a new record. 570 */ 571void 572sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 573{ 574 struct mbuf *m; 575 576 if (m0 == 0) 577 return; 578 579 /* 580 * Put the first mbuf on the queue. 581 * Note this permits zero length records. 582 */ 583 sballoc(sb, m0); 584 SBLASTRECORDCHK(sb, "sbappendrecord 1"); 585 SBLINKRECORD(sb, m0); 586 m = m0->m_next; 587 m0->m_next = 0; 588 if (m && (m0->m_flags & M_EOR)) { 589 m0->m_flags &= ~M_EOR; 590 m->m_flags |= M_EOR; 591 } 592 sbcompress(sb, m, m0); 593 SBLASTRECORDCHK(sb, "sbappendrecord 2"); 594} 595 596/* 597 * As above except that OOB data 598 * is inserted at the beginning of the sockbuf, 599 * but after any other OOB data. 600 */ 601void 602sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 603{ 604 struct mbuf *m, **mp; 605 606 if (m0 == 0) 607 return; 608 609 SBLASTRECORDCHK(sb, "sbinsertoob 1"); 610 611 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) { 612 again: 613 switch (m->m_type) { 614 615 case MT_OOBDATA: 616 continue; /* WANT next train */ 617 618 case MT_CONTROL: 619 if ((m = m->m_next) != NULL) 620 goto again; /* inspect THIS train further */ 621 } 622 break; 623 } 624 /* 625 * Put the first mbuf on the queue. 626 * Note this permits zero length records. 627 */ 628 sballoc(sb, m0); 629 m0->m_nextpkt = *mp; 630 if (*mp == NULL) { 631 /* m0 is actually the new tail */ 632 sb->sb_lastrecord = m0; 633 } 634 *mp = m0; 635 m = m0->m_next; 636 m0->m_next = 0; 637 if (m && (m0->m_flags & M_EOR)) { 638 m0->m_flags &= ~M_EOR; 639 m->m_flags |= M_EOR; 640 } 641 sbcompress(sb, m, m0); 642 SBLASTRECORDCHK(sb, "sbinsertoob 2"); 643} 644 645/* 646 * Append address and data, and optionally, control (ancillary) data 647 * to the receive queue of a socket. If present, 648 * m0 must include a packet header with total length. 649 * Returns 0 if no space in sockbuf or insufficient mbufs. 650 */ 651int 652sbappendaddr(struct sockbuf *sb, struct sockaddr *asa, struct mbuf *m0, 653 struct mbuf *control) 654{ 655 struct mbuf *m, *n, *nlast; 656 int space; 657 658 space = asa->sa_len; 659 660 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 661 panic("sbappendaddr"); 662 if (m0) 663 space += m0->m_pkthdr.len; 664 for (n = control; n; n = n->m_next) { 665 space += n->m_len; 666 if (n->m_next == 0) /* keep pointer to last control buf */ 667 break; 668 } 669 if (space > sbspace(sb)) 670 return (0); 671 MGET(m, M_DONTWAIT, MT_SONAME); 672 if (m == 0) 673 return (0); 674 if (asa->sa_len > MLEN) { 675 MEXTMALLOC(m, asa->sa_len, M_NOWAIT); 676 if ((m->m_flags & M_EXT) == 0) { 677 m_free(m); 678 return (0); 679 } 680 } 681 m->m_len = asa->sa_len; 682 memcpy(mtod(m, caddr_t), (caddr_t)asa, asa->sa_len); 683 if (n) 684 n->m_next = m0; /* concatenate data to control */ 685 else 686 control = m0; 687 m->m_next = control; 688 689 SBLASTRECORDCHK(sb, "sbappendaddr 1"); 690 691 for (n = m; n->m_next != NULL; n = n->m_next) 692 sballoc(sb, n); 693 sballoc(sb, n); 694 nlast = n; 695 SBLINKRECORD(sb, m); 696 697 sb->sb_mbtail = nlast; 698 SBLASTMBUFCHK(sb, "sbappendaddr"); 699 700 SBLASTRECORDCHK(sb, "sbappendaddr 2"); 701 702 return (1); 703} 704 705int 706sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 707{ 708 struct mbuf *m, *mlast, *n; 709 int space; 710 711 space = 0; 712 if (control == 0) 713 panic("sbappendcontrol"); 714 for (m = control; ; m = m->m_next) { 715 space += m->m_len; 716 if (m->m_next == 0) 717 break; 718 } 719 n = m; /* save pointer to last control buffer */ 720 for (m = m0; m; m = m->m_next) 721 space += m->m_len; 722 if (space > sbspace(sb)) 723 return (0); 724 n->m_next = m0; /* concatenate data to control */ 725 726 SBLASTRECORDCHK(sb, "sbappendcontrol 1"); 727 728 for (m = control; m->m_next != NULL; m = m->m_next) 729 sballoc(sb, m); 730 sballoc(sb, m); 731 mlast = m; 732 SBLINKRECORD(sb, control); 733 734 sb->sb_mbtail = mlast; 735 SBLASTMBUFCHK(sb, "sbappendcontrol"); 736 737 SBLASTRECORDCHK(sb, "sbappendcontrol 2"); 738 739 return (1); 740} 741 742/* 743 * Compress mbuf chain m into the socket 744 * buffer sb following mbuf n. If n 745 * is null, the buffer is presumed empty. 746 */ 747void 748sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 749{ 750 int eor; 751 struct mbuf *o; 752 753 eor = 0; 754 while (m) { 755 eor |= m->m_flags & M_EOR; 756 if (m->m_len == 0 && 757 (eor == 0 || 758 (((o = m->m_next) || (o = n)) && 759 o->m_type == m->m_type))) { 760 if (sb->sb_lastrecord == m) 761 sb->sb_lastrecord = m->m_next; 762 m = m_free(m); 763 continue; 764 } 765 if (n && (n->m_flags & M_EOR) == 0 && 766 /* M_TRAILINGSPACE() checks buffer writeability */ 767 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */ 768 m->m_len <= M_TRAILINGSPACE(n) && 769 n->m_type == m->m_type) { 770 memcpy(mtod(n, caddr_t) + n->m_len, mtod(m, caddr_t), 771 (unsigned)m->m_len); 772 n->m_len += m->m_len; 773 sb->sb_cc += m->m_len; 774 m = m_free(m); 775 continue; 776 } 777 if (n) 778 n->m_next = m; 779 else 780 sb->sb_mb = m; 781 sb->sb_mbtail = m; 782 sballoc(sb, m); 783 n = m; 784 m->m_flags &= ~M_EOR; 785 m = m->m_next; 786 n->m_next = 0; 787 } 788 if (eor) { 789 if (n) 790 n->m_flags |= eor; 791 else 792 printf("semi-panic: sbcompress\n"); 793 } 794 SBLASTMBUFCHK(sb, __func__); 795} 796 797/* 798 * Free all mbufs in a sockbuf. 799 * Check that all resources are reclaimed. 800 */ 801void 802sbflush(struct sockbuf *sb) 803{ 804 805 KASSERT((sb->sb_flags & SB_LOCK) == 0); 806 807 while (sb->sb_mbcnt) 808 sbdrop(sb, (int)sb->sb_cc); 809 810 KASSERT(sb->sb_cc == 0); 811 KASSERT(sb->sb_mb == NULL); 812 KASSERT(sb->sb_mbtail == NULL); 813 KASSERT(sb->sb_lastrecord == NULL); 814} 815 816/* 817 * Drop data from (the front of) a sockbuf. 818 */ 819void 820sbdrop(struct sockbuf *sb, int len) 821{ 822 struct mbuf *m, *mn, *next; 823 824 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 825 while (len > 0) { 826 if (m == 0) { 827 if (next == 0) 828 panic("sbdrop"); 829 m = next; 830 next = m->m_nextpkt; 831 continue; 832 } 833 if (m->m_len > len) { 834 m->m_len -= len; 835 m->m_data += len; 836 sb->sb_cc -= len; 837 break; 838 } 839 len -= m->m_len; 840 sbfree(sb, m); 841 MFREE(m, mn); 842 m = mn; 843 } 844 while (m && m->m_len == 0) { 845 sbfree(sb, m); 846 MFREE(m, mn); 847 m = mn; 848 } 849 if (m) { 850 sb->sb_mb = m; 851 m->m_nextpkt = next; 852 } else 853 sb->sb_mb = next; 854 /* 855 * First part is an inline SB_EMPTY_FIXUP(). Second part 856 * makes sure sb_lastrecord is up-to-date if we dropped 857 * part of the last record. 858 */ 859 m = sb->sb_mb; 860 if (m == NULL) { 861 sb->sb_mbtail = NULL; 862 sb->sb_lastrecord = NULL; 863 } else if (m->m_nextpkt == NULL) 864 sb->sb_lastrecord = m; 865} 866 867/* 868 * Drop a record off the front of a sockbuf 869 * and move the next record to the front. 870 */ 871void 872sbdroprecord(struct sockbuf *sb) 873{ 874 struct mbuf *m, *mn; 875 876 m = sb->sb_mb; 877 if (m) { 878 sb->sb_mb = m->m_nextpkt; 879 do { 880 sbfree(sb, m); 881 MFREE(m, mn); 882 } while ((m = mn) != NULL); 883 } 884 SB_EMPTY_FIXUP(sb); 885} 886 887/* 888 * Create a "control" mbuf containing the specified data 889 * with the specified type for presentation on a socket buffer. 890 */ 891struct mbuf * 892sbcreatecontrol(caddr_t p, int size, int type, int level) 893{ 894 struct cmsghdr *cp; 895 struct mbuf *m; 896 897 if (CMSG_SPACE(size) > MCLBYTES) { 898 printf("sbcreatecontrol: message too large %d\n", size); 899 return NULL; 900 } 901 902 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL) 903 return ((struct mbuf *) NULL); 904 if (CMSG_SPACE(size) > MLEN) { 905 MCLGET(m, M_DONTWAIT); 906 if ((m->m_flags & M_EXT) == 0) { 907 m_free(m); 908 return NULL; 909 } 910 } 911 cp = mtod(m, struct cmsghdr *); 912 memcpy(CMSG_DATA(cp), p, size); 913 m->m_len = CMSG_SPACE(size); 914 cp->cmsg_len = CMSG_LEN(size); 915 cp->cmsg_level = level; 916 cp->cmsg_type = type; 917 return (m); 918} 919