uipc_socket2.c revision 1.90
1/* $NetBSD: uipc_socket2.c,v 1.90 2008/03/01 14:16:51 rmind 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. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)uipc_socket2.c 8.2 (Berkeley) 2/14/95 32 */ 33 34#include <sys/cdefs.h> 35__KERNEL_RCSID(0, "$NetBSD: uipc_socket2.c,v 1.90 2008/03/01 14:16:51 rmind Exp $"); 36 37#include "opt_mbuftrace.h" 38#include "opt_sb_max.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/proc.h> 43#include <sys/file.h> 44#include <sys/buf.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/protosw.h> 48#include <sys/poll.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/signalvar.h> 52#include <sys/kauth.h> 53 54/* 55 * Primitive routines for operating on sockets and socket buffers 56 */ 57 58/* strings for sleep message: */ 59const char netcon[] = "netcon"; 60const char netcls[] = "netcls"; 61const char netio[] = "netio"; 62const char netlck[] = "netlck"; 63 64u_long sb_max = SB_MAX; /* maximum socket buffer size */ 65static u_long sb_max_adj; /* adjusted sb_max */ 66 67/* 68 * Procedures to manipulate state flags of socket 69 * and do appropriate wakeups. Normal sequence from the 70 * active (originating) side is that soisconnecting() is 71 * called during processing of connect() call, 72 * resulting in an eventual call to soisconnected() if/when the 73 * connection is established. When the connection is torn down 74 * soisdisconnecting() is called during processing of disconnect() call, 75 * and soisdisconnected() is called when the connection to the peer 76 * is totally severed. The semantics of these routines are such that 77 * connectionless protocols can call soisconnected() and soisdisconnected() 78 * only, bypassing the in-progress calls when setting up a ``connection'' 79 * takes no time. 80 * 81 * From the passive side, a socket is created with 82 * two queues of sockets: so_q0 for connections in progress 83 * and so_q for connections already made and awaiting user acceptance. 84 * As a protocol is preparing incoming connections, it creates a socket 85 * structure queued on so_q0 by calling sonewconn(). When the connection 86 * is established, soisconnected() is called, and transfers the 87 * socket structure to so_q, making it available to accept(). 88 * 89 * If a socket is closed with sockets on either 90 * so_q0 or so_q, these sockets are dropped. 91 * 92 * If higher level protocols are implemented in 93 * the kernel, the wakeups done here will sometimes 94 * cause software-interrupt process scheduling. 95 */ 96 97void 98soisconnecting(struct socket *so) 99{ 100 101 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 102 so->so_state |= SS_ISCONNECTING; 103} 104 105void 106soisconnected(struct socket *so) 107{ 108 struct socket *head; 109 110 head = so->so_head; 111 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 112 so->so_state |= SS_ISCONNECTED; 113 if (head && soqremque(so, 0)) { 114 soqinsque(head, so, 1); 115 sorwakeup(head); 116 wakeup((void *)&head->so_timeo); 117 } else { 118 wakeup((void *)&so->so_timeo); 119 sorwakeup(so); 120 sowwakeup(so); 121 } 122} 123 124void 125soisdisconnecting(struct socket *so) 126{ 127 128 so->so_state &= ~SS_ISCONNECTING; 129 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 130 wakeup((void *)&so->so_timeo); 131 sowwakeup(so); 132 sorwakeup(so); 133} 134 135void 136soisdisconnected(struct socket *so) 137{ 138 139 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 140 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 141 wakeup((void *)&so->so_timeo); 142 sowwakeup(so); 143 sorwakeup(so); 144} 145 146/* 147 * When an attempt at a new connection is noted on a socket 148 * which accepts connections, sonewconn is called. If the 149 * connection is possible (subject to space constraints, etc.) 150 * then we allocate a new structure, propoerly linked into the 151 * data structure of the original socket, and return this. 152 * Connstatus may be 0, SS_ISCONFIRMING, or SS_ISCONNECTED. 153 */ 154struct socket * 155sonewconn(struct socket *head, int connstatus) 156{ 157 struct socket *so; 158 int soqueue; 159 160 soqueue = connstatus ? 1 : 0; 161 if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2) 162 return ((struct socket *)0); 163 so = pool_get(&socket_pool, PR_NOWAIT); 164 if (so == NULL) 165 return (NULL); 166 memset((void *)so, 0, sizeof(*so)); 167 so->so_type = head->so_type; 168 so->so_options = head->so_options &~ SO_ACCEPTCONN; 169 so->so_linger = head->so_linger; 170 so->so_state = head->so_state | SS_NOFDREF; 171 so->so_nbio = head->so_nbio; 172 so->so_proto = head->so_proto; 173 so->so_timeo = head->so_timeo; 174 so->so_pgid = head->so_pgid; 175 so->so_send = head->so_send; 176 so->so_receive = head->so_receive; 177 so->so_uidinfo = head->so_uidinfo; 178#ifdef MBUFTRACE 179 so->so_mowner = head->so_mowner; 180 so->so_rcv.sb_mowner = head->so_rcv.sb_mowner; 181 so->so_snd.sb_mowner = head->so_snd.sb_mowner; 182#endif 183 selinit(&so->so_rcv.sb_sel); 184 selinit(&so->so_snd.sb_sel); 185 (void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat); 186 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 187 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 188 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 189 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 190 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 191 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 192 soqinsque(head, so, soqueue); 193 if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH, 194 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 195 (struct lwp *)0)) { 196 (void) soqremque(so, soqueue); 197 seldestroy(&so->so_rcv.sb_sel); 198 seldestroy(&so->so_snd.sb_sel); 199 pool_put(&socket_pool, so); 200 return (NULL); 201 } 202 if (connstatus) { 203 sorwakeup(head); 204 wakeup((void *)&head->so_timeo); 205 so->so_state |= connstatus; 206 } 207 return (so); 208} 209 210void 211soqinsque(struct socket *head, struct socket *so, int q) 212{ 213 214#ifdef DIAGNOSTIC 215 if (so->so_onq != NULL) 216 panic("soqinsque"); 217#endif 218 219 so->so_head = head; 220 if (q == 0) { 221 head->so_q0len++; 222 so->so_onq = &head->so_q0; 223 } else { 224 head->so_qlen++; 225 so->so_onq = &head->so_q; 226 } 227 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe); 228} 229 230int 231soqremque(struct socket *so, int q) 232{ 233 struct socket *head; 234 235 head = so->so_head; 236 if (q == 0) { 237 if (so->so_onq != &head->so_q0) 238 return (0); 239 head->so_q0len--; 240 } else { 241 if (so->so_onq != &head->so_q) 242 return (0); 243 head->so_qlen--; 244 } 245 TAILQ_REMOVE(so->so_onq, so, so_qe); 246 so->so_onq = NULL; 247 so->so_head = NULL; 248 return (1); 249} 250 251/* 252 * Socantsendmore indicates that no more data will be sent on the 253 * socket; it would normally be applied to a socket when the user 254 * informs the system that no more data is to be sent, by the protocol 255 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 256 * will be received, and will normally be applied to the socket by a 257 * protocol when it detects that the peer will send no more data. 258 * Data queued for reading in the socket may yet be read. 259 */ 260 261void 262socantsendmore(struct socket *so) 263{ 264 265 so->so_state |= SS_CANTSENDMORE; 266 sowwakeup(so); 267} 268 269void 270socantrcvmore(struct socket *so) 271{ 272 273 so->so_state |= SS_CANTRCVMORE; 274 sorwakeup(so); 275} 276 277/* 278 * Wait for data to arrive at/drain from a socket buffer. 279 */ 280int 281sbwait(struct sockbuf *sb) 282{ 283 284 sb->sb_flags |= SB_WAIT; 285 return (tsleep((void *)&sb->sb_cc, 286 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio, 287 sb->sb_timeo)); 288} 289 290/* 291 * Lock a sockbuf already known to be locked; 292 * return any error returned from sleep (EINTR). 293 */ 294int 295sb_lock(struct sockbuf *sb) 296{ 297 int error; 298 299 while (sb->sb_flags & SB_LOCK) { 300 sb->sb_flags |= SB_WANT; 301 error = tsleep((void *)&sb->sb_flags, 302 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH, 303 netlck, 0); 304 if (error) 305 return (error); 306 } 307 sb->sb_flags |= SB_LOCK; 308 return (0); 309} 310 311/* 312 * Wakeup processes waiting on a socket buffer. 313 * Do asynchronous notification via SIGIO 314 * if the socket buffer has the SB_ASYNC flag set. 315 */ 316void 317sowakeup(struct socket *so, struct sockbuf *sb, int code) 318{ 319 int band; 320 321 if (code == POLL_IN) 322 band = POLLIN|POLLRDNORM; 323 else 324 band = POLLOUT|POLLWRNORM; 325 selnotify(&sb->sb_sel, band, 0); 326 327 sb->sb_flags &= ~SB_SEL; 328 if (sb->sb_flags & SB_WAIT) { 329 sb->sb_flags &= ~SB_WAIT; 330 wakeup((void *)&sb->sb_cc); 331 } 332 if (sb->sb_flags & SB_ASYNC) 333 fownsignal(so->so_pgid, SIGIO, code, band, so); 334 if (sb->sb_flags & SB_UPCALL) 335 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); 336} 337 338/* 339 * Socket buffer (struct sockbuf) utility routines. 340 * 341 * Each socket contains two socket buffers: one for sending data and 342 * one for receiving data. Each buffer contains a queue of mbufs, 343 * information about the number of mbufs and amount of data in the 344 * queue, and other fields allowing poll() statements and notification 345 * on data availability to be implemented. 346 * 347 * Data stored in a socket buffer is maintained as a list of records. 348 * Each record is a list of mbufs chained together with the m_next 349 * field. Records are chained together with the m_nextpkt field. The upper 350 * level routine soreceive() expects the following conventions to be 351 * observed when placing information in the receive buffer: 352 * 353 * 1. If the protocol requires each message be preceded by the sender's 354 * name, then a record containing that name must be present before 355 * any associated data (mbuf's must be of type MT_SONAME). 356 * 2. If the protocol supports the exchange of ``access rights'' (really 357 * just additional data associated with the message), and there are 358 * ``rights'' to be received, then a record containing this data 359 * should be present (mbuf's must be of type MT_CONTROL). 360 * 3. If a name or rights record exists, then it must be followed by 361 * a data record, perhaps of zero length. 362 * 363 * Before using a new socket structure it is first necessary to reserve 364 * buffer space to the socket, by calling sbreserve(). This should commit 365 * some of the available buffer space in the system buffer pool for the 366 * socket (currently, it does nothing but enforce limits). The space 367 * should be released by calling sbrelease() when the socket is destroyed. 368 */ 369 370int 371sb_max_set(u_long new_sbmax) 372{ 373 int s; 374 375 if (new_sbmax < (16 * 1024)) 376 return (EINVAL); 377 378 s = splsoftnet(); 379 sb_max = new_sbmax; 380 sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES); 381 splx(s); 382 383 return (0); 384} 385 386int 387soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 388{ 389 /* 390 * there's at least one application (a configure script of screen) 391 * which expects a fifo is writable even if it has "some" bytes 392 * in its buffer. 393 * so we want to make sure (hiwat - lowat) >= (some bytes). 394 * 395 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above. 396 * we expect it's large enough for such applications. 397 */ 398 u_long lowat = MAX(sock_loan_thresh, MCLBYTES); 399 u_long hiwat = lowat + PIPE_BUF; 400 401 if (sndcc < hiwat) 402 sndcc = hiwat; 403 if (sbreserve(&so->so_snd, sndcc, so) == 0) 404 goto bad; 405 if (sbreserve(&so->so_rcv, rcvcc, so) == 0) 406 goto bad2; 407 if (so->so_rcv.sb_lowat == 0) 408 so->so_rcv.sb_lowat = 1; 409 if (so->so_snd.sb_lowat == 0) 410 so->so_snd.sb_lowat = lowat; 411 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 412 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 413 return (0); 414 bad2: 415 sbrelease(&so->so_snd, so); 416 bad: 417 return (ENOBUFS); 418} 419 420/* 421 * Allot mbufs to a sockbuf. 422 * Attempt to scale mbmax so that mbcnt doesn't become limiting 423 * if buffering efficiency is near the normal case. 424 */ 425int 426sbreserve(struct sockbuf *sb, u_long cc, struct socket *so) 427{ 428 struct lwp *l = curlwp; /* XXX */ 429 rlim_t maxcc; 430 struct uidinfo *uidinfo; 431 432 KDASSERT(sb_max_adj != 0); 433 if (cc == 0 || cc > sb_max_adj) 434 return (0); 435 if (so) { 436 if (kauth_cred_geteuid(l->l_cred) == so->so_uidinfo->ui_uid) 437 maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur; 438 else 439 maxcc = RLIM_INFINITY; 440 uidinfo = so->so_uidinfo; 441 } else { 442 uidinfo = uid_find(0); /* XXX: nothing better */ 443 maxcc = RLIM_INFINITY; 444 } 445 if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc)) 446 return 0; 447 sb->sb_mbmax = min(cc * 2, sb_max); 448 if (sb->sb_lowat > sb->sb_hiwat) 449 sb->sb_lowat = sb->sb_hiwat; 450 return (1); 451} 452 453/* 454 * Free mbufs held by a socket, and reserved mbuf space. 455 */ 456void 457sbrelease(struct sockbuf *sb, struct socket *so) 458{ 459 460 sbflush(sb); 461 (void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); 462 sb->sb_mbmax = 0; 463} 464 465/* 466 * Routines to add and remove 467 * data from an mbuf queue. 468 * 469 * The routines sbappend() or sbappendrecord() are normally called to 470 * append new mbufs to a socket buffer, after checking that adequate 471 * space is available, comparing the function sbspace() with the amount 472 * of data to be added. sbappendrecord() differs from sbappend() in 473 * that data supplied is treated as the beginning of a new record. 474 * To place a sender's address, optional access rights, and data in a 475 * socket receive buffer, sbappendaddr() should be used. To place 476 * access rights and data in a socket receive buffer, sbappendrights() 477 * should be used. In either case, the new data begins a new record. 478 * Note that unlike sbappend() and sbappendrecord(), these routines check 479 * for the caller that there will be enough space to store the data. 480 * Each fails if there is not enough space, or if it cannot find mbufs 481 * to store additional information in. 482 * 483 * Reliable protocols may use the socket send buffer to hold data 484 * awaiting acknowledgement. Data is normally copied from a socket 485 * send buffer in a protocol with m_copy for output to a peer, 486 * and then removing the data from the socket buffer with sbdrop() 487 * or sbdroprecord() when the data is acknowledged by the peer. 488 */ 489 490#ifdef SOCKBUF_DEBUG 491void 492sblastrecordchk(struct sockbuf *sb, const char *where) 493{ 494 struct mbuf *m = sb->sb_mb; 495 496 while (m && m->m_nextpkt) 497 m = m->m_nextpkt; 498 499 if (m != sb->sb_lastrecord) { 500 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n", 501 sb->sb_mb, sb->sb_lastrecord, m); 502 printf("packet chain:\n"); 503 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 504 printf("\t%p\n", m); 505 panic("sblastrecordchk from %s", where); 506 } 507} 508 509void 510sblastmbufchk(struct sockbuf *sb, const char *where) 511{ 512 struct mbuf *m = sb->sb_mb; 513 struct mbuf *n; 514 515 while (m && m->m_nextpkt) 516 m = m->m_nextpkt; 517 518 while (m && m->m_next) 519 m = m->m_next; 520 521 if (m != sb->sb_mbtail) { 522 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n", 523 sb->sb_mb, sb->sb_mbtail, m); 524 printf("packet tree:\n"); 525 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 526 printf("\t"); 527 for (n = m; n != NULL; n = n->m_next) 528 printf("%p ", n); 529 printf("\n"); 530 } 531 panic("sblastmbufchk from %s", where); 532 } 533} 534#endif /* SOCKBUF_DEBUG */ 535 536/* 537 * Link a chain of records onto a socket buffer 538 */ 539#define SBLINKRECORDCHAIN(sb, m0, mlast) \ 540do { \ 541 if ((sb)->sb_lastrecord != NULL) \ 542 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 543 else \ 544 (sb)->sb_mb = (m0); \ 545 (sb)->sb_lastrecord = (mlast); \ 546} while (/*CONSTCOND*/0) 547 548 549#define SBLINKRECORD(sb, m0) \ 550 SBLINKRECORDCHAIN(sb, m0, m0) 551 552/* 553 * Append mbuf chain m to the last record in the 554 * socket buffer sb. The additional space associated 555 * the mbuf chain is recorded in sb. Empty mbufs are 556 * discarded and mbufs are compacted where possible. 557 */ 558void 559sbappend(struct sockbuf *sb, struct mbuf *m) 560{ 561 struct mbuf *n; 562 563 if (m == 0) 564 return; 565 566#ifdef MBUFTRACE 567 m_claimm(m, sb->sb_mowner); 568#endif 569 570 SBLASTRECORDCHK(sb, "sbappend 1"); 571 572 if ((n = sb->sb_lastrecord) != NULL) { 573 /* 574 * XXX Would like to simply use sb_mbtail here, but 575 * XXX I need to verify that I won't miss an EOR that 576 * XXX way. 577 */ 578 do { 579 if (n->m_flags & M_EOR) { 580 sbappendrecord(sb, m); /* XXXXXX!!!! */ 581 return; 582 } 583 } while (n->m_next && (n = n->m_next)); 584 } else { 585 /* 586 * If this is the first record in the socket buffer, it's 587 * also the last record. 588 */ 589 sb->sb_lastrecord = m; 590 } 591 sbcompress(sb, m, n); 592 SBLASTRECORDCHK(sb, "sbappend 2"); 593} 594 595/* 596 * This version of sbappend() should only be used when the caller 597 * absolutely knows that there will never be more than one record 598 * in the socket buffer, that is, a stream protocol (such as TCP). 599 */ 600void 601sbappendstream(struct sockbuf *sb, struct mbuf *m) 602{ 603 604 KDASSERT(m->m_nextpkt == NULL); 605 KASSERT(sb->sb_mb == sb->sb_lastrecord); 606 607 SBLASTMBUFCHK(sb, __func__); 608 609#ifdef MBUFTRACE 610 m_claimm(m, sb->sb_mowner); 611#endif 612 613 sbcompress(sb, m, sb->sb_mbtail); 614 615 sb->sb_lastrecord = sb->sb_mb; 616 SBLASTRECORDCHK(sb, __func__); 617} 618 619#ifdef SOCKBUF_DEBUG 620void 621sbcheck(struct sockbuf *sb) 622{ 623 struct mbuf *m; 624 u_long len, mbcnt; 625 626 len = 0; 627 mbcnt = 0; 628 for (m = sb->sb_mb; m; m = m->m_next) { 629 len += m->m_len; 630 mbcnt += MSIZE; 631 if (m->m_flags & M_EXT) 632 mbcnt += m->m_ext.ext_size; 633 if (m->m_nextpkt) 634 panic("sbcheck nextpkt"); 635 } 636 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 637 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc, 638 mbcnt, sb->sb_mbcnt); 639 panic("sbcheck"); 640 } 641} 642#endif 643 644/* 645 * As above, except the mbuf chain 646 * begins a new record. 647 */ 648void 649sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 650{ 651 struct mbuf *m; 652 653 if (m0 == 0) 654 return; 655 656#ifdef MBUFTRACE 657 m_claimm(m0, sb->sb_mowner); 658#endif 659 /* 660 * Put the first mbuf on the queue. 661 * Note this permits zero length records. 662 */ 663 sballoc(sb, m0); 664 SBLASTRECORDCHK(sb, "sbappendrecord 1"); 665 SBLINKRECORD(sb, m0); 666 m = m0->m_next; 667 m0->m_next = 0; 668 if (m && (m0->m_flags & M_EOR)) { 669 m0->m_flags &= ~M_EOR; 670 m->m_flags |= M_EOR; 671 } 672 sbcompress(sb, m, m0); 673 SBLASTRECORDCHK(sb, "sbappendrecord 2"); 674} 675 676/* 677 * As above except that OOB data 678 * is inserted at the beginning of the sockbuf, 679 * but after any other OOB data. 680 */ 681void 682sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 683{ 684 struct mbuf *m, **mp; 685 686 if (m0 == 0) 687 return; 688 689 SBLASTRECORDCHK(sb, "sbinsertoob 1"); 690 691 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) { 692 again: 693 switch (m->m_type) { 694 695 case MT_OOBDATA: 696 continue; /* WANT next train */ 697 698 case MT_CONTROL: 699 if ((m = m->m_next) != NULL) 700 goto again; /* inspect THIS train further */ 701 } 702 break; 703 } 704 /* 705 * Put the first mbuf on the queue. 706 * Note this permits zero length records. 707 */ 708 sballoc(sb, m0); 709 m0->m_nextpkt = *mp; 710 if (*mp == NULL) { 711 /* m0 is actually the new tail */ 712 sb->sb_lastrecord = m0; 713 } 714 *mp = m0; 715 m = m0->m_next; 716 m0->m_next = 0; 717 if (m && (m0->m_flags & M_EOR)) { 718 m0->m_flags &= ~M_EOR; 719 m->m_flags |= M_EOR; 720 } 721 sbcompress(sb, m, m0); 722 SBLASTRECORDCHK(sb, "sbinsertoob 2"); 723} 724 725/* 726 * Append address and data, and optionally, control (ancillary) data 727 * to the receive queue of a socket. If present, 728 * m0 must include a packet header with total length. 729 * Returns 0 if no space in sockbuf or insufficient mbufs. 730 */ 731int 732sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, 733 struct mbuf *control) 734{ 735 struct mbuf *m, *n, *nlast; 736 int space, len; 737 738 space = asa->sa_len; 739 740 if (m0 != NULL) { 741 if ((m0->m_flags & M_PKTHDR) == 0) 742 panic("sbappendaddr"); 743 space += m0->m_pkthdr.len; 744#ifdef MBUFTRACE 745 m_claimm(m0, sb->sb_mowner); 746#endif 747 } 748 for (n = control; n; n = n->m_next) { 749 space += n->m_len; 750 MCLAIM(n, sb->sb_mowner); 751 if (n->m_next == 0) /* keep pointer to last control buf */ 752 break; 753 } 754 if (space > sbspace(sb)) 755 return (0); 756 MGET(m, M_DONTWAIT, MT_SONAME); 757 if (m == 0) 758 return (0); 759 MCLAIM(m, sb->sb_mowner); 760 /* 761 * XXX avoid 'comparison always true' warning which isn't easily 762 * avoided. 763 */ 764 len = asa->sa_len; 765 if (len > MLEN) { 766 MEXTMALLOC(m, asa->sa_len, M_NOWAIT); 767 if ((m->m_flags & M_EXT) == 0) { 768 m_free(m); 769 return (0); 770 } 771 } 772 m->m_len = asa->sa_len; 773 memcpy(mtod(m, void *), asa, asa->sa_len); 774 if (n) 775 n->m_next = m0; /* concatenate data to control */ 776 else 777 control = m0; 778 m->m_next = control; 779 780 SBLASTRECORDCHK(sb, "sbappendaddr 1"); 781 782 for (n = m; n->m_next != NULL; n = n->m_next) 783 sballoc(sb, n); 784 sballoc(sb, n); 785 nlast = n; 786 SBLINKRECORD(sb, m); 787 788 sb->sb_mbtail = nlast; 789 SBLASTMBUFCHK(sb, "sbappendaddr"); 790 791 SBLASTRECORDCHK(sb, "sbappendaddr 2"); 792 793 return (1); 794} 795 796/* 797 * Helper for sbappendchainaddr: prepend a struct sockaddr* to 798 * an mbuf chain. 799 */ 800static inline struct mbuf * 801m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0, 802 const struct sockaddr *asa) 803{ 804 struct mbuf *m; 805 const int salen = asa->sa_len; 806 807 /* only the first in each chain need be a pkthdr */ 808 MGETHDR(m, M_DONTWAIT, MT_SONAME); 809 if (m == 0) 810 return (0); 811 MCLAIM(m, sb->sb_mowner); 812#ifdef notyet 813 if (salen > MHLEN) { 814 MEXTMALLOC(m, salen, M_NOWAIT); 815 if ((m->m_flags & M_EXT) == 0) { 816 m_free(m); 817 return (0); 818 } 819 } 820#else 821 KASSERT(salen <= MHLEN); 822#endif 823 m->m_len = salen; 824 memcpy(mtod(m, void *), asa, salen); 825 m->m_next = m0; 826 m->m_pkthdr.len = salen + m0->m_pkthdr.len; 827 828 return m; 829} 830 831int 832sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa, 833 struct mbuf *m0, int sbprio) 834{ 835 int space; 836 struct mbuf *m, *n, *n0, *nlast; 837 int error; 838 839 /* 840 * XXX sbprio reserved for encoding priority of this* request: 841 * SB_PRIO_NONE --> honour normal sb limits 842 * SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space, 843 * take whole chain. Intended for large requests 844 * that should be delivered atomically (all, or none). 845 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow 846 * over normal socket limits, for messages indicating 847 * buffer overflow in earlier normal/lower-priority messages 848 * SB_PRIO_BESTEFFORT --> ignore limits entirely. 849 * Intended for kernel-generated messages only. 850 * Up to generator to avoid total mbuf resource exhaustion. 851 */ 852 (void)sbprio; 853 854 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 855 panic("sbappendaddrchain"); 856 857 space = sbspace(sb); 858 859#ifdef notyet 860 /* 861 * Enforce SB_PRIO_* limits as described above. 862 */ 863#endif 864 865 n0 = NULL; 866 nlast = NULL; 867 for (m = m0; m; m = m->m_nextpkt) { 868 struct mbuf *np; 869 870#ifdef MBUFTRACE 871 m_claimm(m, sb->sb_mowner); 872#endif 873 874 /* Prepend sockaddr to this record (m) of input chain m0 */ 875 n = m_prepend_sockaddr(sb, m, asa); 876 if (n == NULL) { 877 error = ENOBUFS; 878 goto bad; 879 } 880 881 /* Append record (asa+m) to end of new chain n0 */ 882 if (n0 == NULL) { 883 n0 = n; 884 } else { 885 nlast->m_nextpkt = n; 886 } 887 /* Keep track of last record on new chain */ 888 nlast = n; 889 890 for (np = n; np; np = np->m_next) 891 sballoc(sb, np); 892 } 893 894 SBLASTRECORDCHK(sb, "sbappendaddrchain 1"); 895 896 /* Drop the entire chain of (asa+m) records onto the socket */ 897 SBLINKRECORDCHAIN(sb, n0, nlast); 898 899 SBLASTRECORDCHK(sb, "sbappendaddrchain 2"); 900 901 for (m = nlast; m->m_next; m = m->m_next) 902 ; 903 sb->sb_mbtail = m; 904 SBLASTMBUFCHK(sb, "sbappendaddrchain"); 905 906 return (1); 907 908bad: 909 /* 910 * On error, free the prepended addreseses. For consistency 911 * with sbappendaddr(), leave it to our caller to free 912 * the input record chain passed to us as m0. 913 */ 914 while ((n = n0) != NULL) { 915 struct mbuf *np; 916 917 /* Undo the sballoc() of this record */ 918 for (np = n; np; np = np->m_next) 919 sbfree(sb, np); 920 921 n0 = n->m_nextpkt; /* iterate at next prepended address */ 922 MFREE(n, np); /* free prepended address (not data) */ 923 } 924 return 0; 925} 926 927 928int 929sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 930{ 931 struct mbuf *m, *mlast, *n; 932 int space; 933 934 space = 0; 935 if (control == 0) 936 panic("sbappendcontrol"); 937 for (m = control; ; m = m->m_next) { 938 space += m->m_len; 939 MCLAIM(m, sb->sb_mowner); 940 if (m->m_next == 0) 941 break; 942 } 943 n = m; /* save pointer to last control buffer */ 944 for (m = m0; m; m = m->m_next) { 945 MCLAIM(m, sb->sb_mowner); 946 space += m->m_len; 947 } 948 if (space > sbspace(sb)) 949 return (0); 950 n->m_next = m0; /* concatenate data to control */ 951 952 SBLASTRECORDCHK(sb, "sbappendcontrol 1"); 953 954 for (m = control; m->m_next != NULL; m = m->m_next) 955 sballoc(sb, m); 956 sballoc(sb, m); 957 mlast = m; 958 SBLINKRECORD(sb, control); 959 960 sb->sb_mbtail = mlast; 961 SBLASTMBUFCHK(sb, "sbappendcontrol"); 962 963 SBLASTRECORDCHK(sb, "sbappendcontrol 2"); 964 965 return (1); 966} 967 968/* 969 * Compress mbuf chain m into the socket 970 * buffer sb following mbuf n. If n 971 * is null, the buffer is presumed empty. 972 */ 973void 974sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 975{ 976 int eor; 977 struct mbuf *o; 978 979 eor = 0; 980 while (m) { 981 eor |= m->m_flags & M_EOR; 982 if (m->m_len == 0 && 983 (eor == 0 || 984 (((o = m->m_next) || (o = n)) && 985 o->m_type == m->m_type))) { 986 if (sb->sb_lastrecord == m) 987 sb->sb_lastrecord = m->m_next; 988 m = m_free(m); 989 continue; 990 } 991 if (n && (n->m_flags & M_EOR) == 0 && 992 /* M_TRAILINGSPACE() checks buffer writeability */ 993 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */ 994 m->m_len <= M_TRAILINGSPACE(n) && 995 n->m_type == m->m_type) { 996 memcpy(mtod(n, char *) + n->m_len, mtod(m, void *), 997 (unsigned)m->m_len); 998 n->m_len += m->m_len; 999 sb->sb_cc += m->m_len; 1000 m = m_free(m); 1001 continue; 1002 } 1003 if (n) 1004 n->m_next = m; 1005 else 1006 sb->sb_mb = m; 1007 sb->sb_mbtail = m; 1008 sballoc(sb, m); 1009 n = m; 1010 m->m_flags &= ~M_EOR; 1011 m = m->m_next; 1012 n->m_next = 0; 1013 } 1014 if (eor) { 1015 if (n) 1016 n->m_flags |= eor; 1017 else 1018 printf("semi-panic: sbcompress\n"); 1019 } 1020 SBLASTMBUFCHK(sb, __func__); 1021} 1022 1023/* 1024 * Free all mbufs in a sockbuf. 1025 * Check that all resources are reclaimed. 1026 */ 1027void 1028sbflush(struct sockbuf *sb) 1029{ 1030 1031 KASSERT((sb->sb_flags & SB_LOCK) == 0); 1032 1033 while (sb->sb_mbcnt) 1034 sbdrop(sb, (int)sb->sb_cc); 1035 1036 KASSERT(sb->sb_cc == 0); 1037 KASSERT(sb->sb_mb == NULL); 1038 KASSERT(sb->sb_mbtail == NULL); 1039 KASSERT(sb->sb_lastrecord == NULL); 1040} 1041 1042/* 1043 * Drop data from (the front of) a sockbuf. 1044 */ 1045void 1046sbdrop(struct sockbuf *sb, int len) 1047{ 1048 struct mbuf *m, *mn, *next; 1049 1050 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1051 while (len > 0) { 1052 if (m == 0) { 1053 if (next == 0) 1054 panic("sbdrop"); 1055 m = next; 1056 next = m->m_nextpkt; 1057 continue; 1058 } 1059 if (m->m_len > len) { 1060 m->m_len -= len; 1061 m->m_data += len; 1062 sb->sb_cc -= len; 1063 break; 1064 } 1065 len -= m->m_len; 1066 sbfree(sb, m); 1067 MFREE(m, mn); 1068 m = mn; 1069 } 1070 while (m && m->m_len == 0) { 1071 sbfree(sb, m); 1072 MFREE(m, mn); 1073 m = mn; 1074 } 1075 if (m) { 1076 sb->sb_mb = m; 1077 m->m_nextpkt = next; 1078 } else 1079 sb->sb_mb = next; 1080 /* 1081 * First part is an inline SB_EMPTY_FIXUP(). Second part 1082 * makes sure sb_lastrecord is up-to-date if we dropped 1083 * part of the last record. 1084 */ 1085 m = sb->sb_mb; 1086 if (m == NULL) { 1087 sb->sb_mbtail = NULL; 1088 sb->sb_lastrecord = NULL; 1089 } else if (m->m_nextpkt == NULL) 1090 sb->sb_lastrecord = m; 1091} 1092 1093/* 1094 * Drop a record off the front of a sockbuf 1095 * and move the next record to the front. 1096 */ 1097void 1098sbdroprecord(struct sockbuf *sb) 1099{ 1100 struct mbuf *m, *mn; 1101 1102 m = sb->sb_mb; 1103 if (m) { 1104 sb->sb_mb = m->m_nextpkt; 1105 do { 1106 sbfree(sb, m); 1107 MFREE(m, mn); 1108 } while ((m = mn) != NULL); 1109 } 1110 SB_EMPTY_FIXUP(sb); 1111} 1112 1113/* 1114 * Create a "control" mbuf containing the specified data 1115 * with the specified type for presentation on a socket buffer. 1116 */ 1117struct mbuf * 1118sbcreatecontrol(void *p, int size, int type, int level) 1119{ 1120 struct cmsghdr *cp; 1121 struct mbuf *m; 1122 1123 if (CMSG_SPACE(size) > MCLBYTES) { 1124 printf("sbcreatecontrol: message too large %d\n", size); 1125 return NULL; 1126 } 1127 1128 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL) 1129 return ((struct mbuf *) NULL); 1130 if (CMSG_SPACE(size) > MLEN) { 1131 MCLGET(m, M_DONTWAIT); 1132 if ((m->m_flags & M_EXT) == 0) { 1133 m_free(m); 1134 return NULL; 1135 } 1136 } 1137 cp = mtod(m, struct cmsghdr *); 1138 memcpy(CMSG_DATA(cp), p, size); 1139 m->m_len = CMSG_SPACE(size); 1140 cp->cmsg_len = CMSG_LEN(size); 1141 cp->cmsg_level = level; 1142 cp->cmsg_type = type; 1143 return (m); 1144} 1145