uipc_socket2.c revision 1.87
1/* $NetBSD: uipc_socket2.c,v 1.87 2008/01/29 09:31:22 yamt 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 1.87 2008/01/29 09:31:22 yamt 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_proto = head->so_proto; 172 so->so_timeo = head->so_timeo; 173 so->so_pgid = head->so_pgid; 174 so->so_send = head->so_send; 175 so->so_receive = head->so_receive; 176 so->so_uidinfo = head->so_uidinfo; 177#ifdef MBUFTRACE 178 so->so_mowner = head->so_mowner; 179 so->so_rcv.sb_mowner = head->so_rcv.sb_mowner; 180 so->so_snd.sb_mowner = head->so_snd.sb_mowner; 181#endif 182 selinit(&so->so_rcv.sb_sel); 183 selinit(&so->so_snd.sb_sel); 184 (void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat); 185 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 186 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 187 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 188 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 189 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 190 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 191 soqinsque(head, so, soqueue); 192 if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH, 193 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 194 (struct lwp *)0)) { 195 (void) soqremque(so, soqueue); 196 seldestroy(&so->so_rcv.sb_sel); 197 seldestroy(&so->so_snd.sb_sel); 198 pool_put(&socket_pool, so); 199 return (NULL); 200 } 201 if (connstatus) { 202 sorwakeup(head); 203 wakeup((void *)&head->so_timeo); 204 so->so_state |= connstatus; 205 } 206 return (so); 207} 208 209void 210soqinsque(struct socket *head, struct socket *so, int q) 211{ 212 213#ifdef DIAGNOSTIC 214 if (so->so_onq != NULL) 215 panic("soqinsque"); 216#endif 217 218 so->so_head = head; 219 if (q == 0) { 220 head->so_q0len++; 221 so->so_onq = &head->so_q0; 222 } else { 223 head->so_qlen++; 224 so->so_onq = &head->so_q; 225 } 226 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe); 227} 228 229int 230soqremque(struct socket *so, int q) 231{ 232 struct socket *head; 233 234 head = so->so_head; 235 if (q == 0) { 236 if (so->so_onq != &head->so_q0) 237 return (0); 238 head->so_q0len--; 239 } else { 240 if (so->so_onq != &head->so_q) 241 return (0); 242 head->so_qlen--; 243 } 244 TAILQ_REMOVE(so->so_onq, so, so_qe); 245 so->so_onq = NULL; 246 so->so_head = NULL; 247 return (1); 248} 249 250/* 251 * Socantsendmore indicates that no more data will be sent on the 252 * socket; it would normally be applied to a socket when the user 253 * informs the system that no more data is to be sent, by the protocol 254 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 255 * will be received, and will normally be applied to the socket by a 256 * protocol when it detects that the peer will send no more data. 257 * Data queued for reading in the socket may yet be read. 258 */ 259 260void 261socantsendmore(struct socket *so) 262{ 263 264 so->so_state |= SS_CANTSENDMORE; 265 sowwakeup(so); 266} 267 268void 269socantrcvmore(struct socket *so) 270{ 271 272 so->so_state |= SS_CANTRCVMORE; 273 sorwakeup(so); 274} 275 276/* 277 * Wait for data to arrive at/drain from a socket buffer. 278 */ 279int 280sbwait(struct sockbuf *sb) 281{ 282 283 sb->sb_flags |= SB_WAIT; 284 return (tsleep((void *)&sb->sb_cc, 285 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio, 286 sb->sb_timeo)); 287} 288 289/* 290 * Lock a sockbuf already known to be locked; 291 * return any error returned from sleep (EINTR). 292 */ 293int 294sb_lock(struct sockbuf *sb) 295{ 296 int error; 297 298 while (sb->sb_flags & SB_LOCK) { 299 sb->sb_flags |= SB_WANT; 300 error = tsleep((void *)&sb->sb_flags, 301 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH, 302 netlck, 0); 303 if (error) 304 return (error); 305 } 306 sb->sb_flags |= SB_LOCK; 307 return (0); 308} 309 310/* 311 * Wakeup processes waiting on a socket buffer. 312 * Do asynchronous notification via SIGIO 313 * if the socket buffer has the SB_ASYNC flag set. 314 */ 315void 316sowakeup(struct socket *so, struct sockbuf *sb, int code) 317{ 318 selnotify(&sb->sb_sel, 0); 319 sb->sb_flags &= ~SB_SEL; 320 if (sb->sb_flags & SB_WAIT) { 321 sb->sb_flags &= ~SB_WAIT; 322 wakeup((void *)&sb->sb_cc); 323 } 324 if (sb->sb_flags & SB_ASYNC) { 325 int band; 326 if (code == POLL_IN) 327 band = POLLIN|POLLRDNORM; 328 else 329 band = POLLOUT|POLLWRNORM; 330 fownsignal(so->so_pgid, SIGIO, code, band, so); 331 } 332 if (sb->sb_flags & SB_UPCALL) 333 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); 334} 335 336/* 337 * Socket buffer (struct sockbuf) utility routines. 338 * 339 * Each socket contains two socket buffers: one for sending data and 340 * one for receiving data. Each buffer contains a queue of mbufs, 341 * information about the number of mbufs and amount of data in the 342 * queue, and other fields allowing poll() statements and notification 343 * on data availability to be implemented. 344 * 345 * Data stored in a socket buffer is maintained as a list of records. 346 * Each record is a list of mbufs chained together with the m_next 347 * field. Records are chained together with the m_nextpkt field. The upper 348 * level routine soreceive() expects the following conventions to be 349 * observed when placing information in the receive buffer: 350 * 351 * 1. If the protocol requires each message be preceded by the sender's 352 * name, then a record containing that name must be present before 353 * any associated data (mbuf's must be of type MT_SONAME). 354 * 2. If the protocol supports the exchange of ``access rights'' (really 355 * just additional data associated with the message), and there are 356 * ``rights'' to be received, then a record containing this data 357 * should be present (mbuf's must be of type MT_CONTROL). 358 * 3. If a name or rights record exists, then it must be followed by 359 * a data record, perhaps of zero length. 360 * 361 * Before using a new socket structure it is first necessary to reserve 362 * buffer space to the socket, by calling sbreserve(). This should commit 363 * some of the available buffer space in the system buffer pool for the 364 * socket (currently, it does nothing but enforce limits). The space 365 * should be released by calling sbrelease() when the socket is destroyed. 366 */ 367 368int 369sb_max_set(u_long new_sbmax) 370{ 371 int s; 372 373 if (new_sbmax < (16 * 1024)) 374 return (EINVAL); 375 376 s = splsoftnet(); 377 sb_max = new_sbmax; 378 sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES); 379 splx(s); 380 381 return (0); 382} 383 384int 385soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 386{ 387 /* 388 * there's at least one application (a configure script of screen) 389 * which expects a fifo is writable even if it has "some" bytes 390 * in its buffer. 391 * so we want to make sure (hiwat - lowat) >= (some bytes). 392 * 393 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above. 394 * we expect it's large enough for such applications. 395 */ 396 u_long lowat = MAX(sock_loan_thresh, MCLBYTES); 397 u_long hiwat = lowat + PIPE_BUF; 398 399 if (sndcc < hiwat) 400 sndcc = hiwat; 401 if (sbreserve(&so->so_snd, sndcc, so) == 0) 402 goto bad; 403 if (sbreserve(&so->so_rcv, rcvcc, so) == 0) 404 goto bad2; 405 if (so->so_rcv.sb_lowat == 0) 406 so->so_rcv.sb_lowat = 1; 407 if (so->so_snd.sb_lowat == 0) 408 so->so_snd.sb_lowat = lowat; 409 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 410 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 411 return (0); 412 bad2: 413 sbrelease(&so->so_snd, so); 414 bad: 415 return (ENOBUFS); 416} 417 418/* 419 * Allot mbufs to a sockbuf. 420 * Attempt to scale mbmax so that mbcnt doesn't become limiting 421 * if buffering efficiency is near the normal case. 422 */ 423int 424sbreserve(struct sockbuf *sb, u_long cc, struct socket *so) 425{ 426 struct lwp *l = curlwp; /* XXX */ 427 rlim_t maxcc; 428 struct uidinfo *uidinfo; 429 430 KDASSERT(sb_max_adj != 0); 431 if (cc == 0 || cc > sb_max_adj) 432 return (0); 433 if (so) { 434 if (l && kauth_cred_geteuid(l->l_cred) == so->so_uidinfo->ui_uid) 435 maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur; 436 else 437 maxcc = RLIM_INFINITY; 438 uidinfo = so->so_uidinfo; 439 } else { 440 uidinfo = uid_find(0); /* XXX: nothing better */ 441 maxcc = RLIM_INFINITY; 442 } 443 if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc)) 444 return 0; 445 sb->sb_mbmax = min(cc * 2, sb_max); 446 if (sb->sb_lowat > sb->sb_hiwat) 447 sb->sb_lowat = sb->sb_hiwat; 448 return (1); 449} 450 451/* 452 * Free mbufs held by a socket, and reserved mbuf space. 453 */ 454void 455sbrelease(struct sockbuf *sb, struct socket *so) 456{ 457 458 sbflush(sb); 459 (void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); 460 sb->sb_mbmax = 0; 461} 462 463/* 464 * Routines to add and remove 465 * data from an mbuf queue. 466 * 467 * The routines sbappend() or sbappendrecord() are normally called to 468 * append new mbufs to a socket buffer, after checking that adequate 469 * space is available, comparing the function sbspace() with the amount 470 * of data to be added. sbappendrecord() differs from sbappend() in 471 * that data supplied is treated as the beginning of a new record. 472 * To place a sender's address, optional access rights, and data in a 473 * socket receive buffer, sbappendaddr() should be used. To place 474 * access rights and data in a socket receive buffer, sbappendrights() 475 * should be used. In either case, the new data begins a new record. 476 * Note that unlike sbappend() and sbappendrecord(), these routines check 477 * for the caller that there will be enough space to store the data. 478 * Each fails if there is not enough space, or if it cannot find mbufs 479 * to store additional information in. 480 * 481 * Reliable protocols may use the socket send buffer to hold data 482 * awaiting acknowledgement. Data is normally copied from a socket 483 * send buffer in a protocol with m_copy for output to a peer, 484 * and then removing the data from the socket buffer with sbdrop() 485 * or sbdroprecord() when the data is acknowledged by the peer. 486 */ 487 488#ifdef SOCKBUF_DEBUG 489void 490sblastrecordchk(struct sockbuf *sb, const char *where) 491{ 492 struct mbuf *m = sb->sb_mb; 493 494 while (m && m->m_nextpkt) 495 m = m->m_nextpkt; 496 497 if (m != sb->sb_lastrecord) { 498 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n", 499 sb->sb_mb, sb->sb_lastrecord, m); 500 printf("packet chain:\n"); 501 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 502 printf("\t%p\n", m); 503 panic("sblastrecordchk from %s", where); 504 } 505} 506 507void 508sblastmbufchk(struct sockbuf *sb, const char *where) 509{ 510 struct mbuf *m = sb->sb_mb; 511 struct mbuf *n; 512 513 while (m && m->m_nextpkt) 514 m = m->m_nextpkt; 515 516 while (m && m->m_next) 517 m = m->m_next; 518 519 if (m != sb->sb_mbtail) { 520 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n", 521 sb->sb_mb, sb->sb_mbtail, m); 522 printf("packet tree:\n"); 523 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 524 printf("\t"); 525 for (n = m; n != NULL; n = n->m_next) 526 printf("%p ", n); 527 printf("\n"); 528 } 529 panic("sblastmbufchk from %s", where); 530 } 531} 532#endif /* SOCKBUF_DEBUG */ 533 534/* 535 * Link a chain of records onto a socket buffer 536 */ 537#define SBLINKRECORDCHAIN(sb, m0, mlast) \ 538do { \ 539 if ((sb)->sb_lastrecord != NULL) \ 540 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 541 else \ 542 (sb)->sb_mb = (m0); \ 543 (sb)->sb_lastrecord = (mlast); \ 544} while (/*CONSTCOND*/0) 545 546 547#define SBLINKRECORD(sb, m0) \ 548 SBLINKRECORDCHAIN(sb, m0, m0) 549 550/* 551 * Append mbuf chain m to the last record in the 552 * socket buffer sb. The additional space associated 553 * the mbuf chain is recorded in sb. Empty mbufs are 554 * discarded and mbufs are compacted where possible. 555 */ 556void 557sbappend(struct sockbuf *sb, struct mbuf *m) 558{ 559 struct mbuf *n; 560 561 if (m == 0) 562 return; 563 564#ifdef MBUFTRACE 565 m_claimm(m, sb->sb_mowner); 566#endif 567 568 SBLASTRECORDCHK(sb, "sbappend 1"); 569 570 if ((n = sb->sb_lastrecord) != NULL) { 571 /* 572 * XXX Would like to simply use sb_mbtail here, but 573 * XXX I need to verify that I won't miss an EOR that 574 * XXX way. 575 */ 576 do { 577 if (n->m_flags & M_EOR) { 578 sbappendrecord(sb, m); /* XXXXXX!!!! */ 579 return; 580 } 581 } while (n->m_next && (n = n->m_next)); 582 } else { 583 /* 584 * If this is the first record in the socket buffer, it's 585 * also the last record. 586 */ 587 sb->sb_lastrecord = m; 588 } 589 sbcompress(sb, m, n); 590 SBLASTRECORDCHK(sb, "sbappend 2"); 591} 592 593/* 594 * This version of sbappend() should only be used when the caller 595 * absolutely knows that there will never be more than one record 596 * in the socket buffer, that is, a stream protocol (such as TCP). 597 */ 598void 599sbappendstream(struct sockbuf *sb, struct mbuf *m) 600{ 601 602 KDASSERT(m->m_nextpkt == NULL); 603 KASSERT(sb->sb_mb == sb->sb_lastrecord); 604 605 SBLASTMBUFCHK(sb, __func__); 606 607#ifdef MBUFTRACE 608 m_claimm(m, sb->sb_mowner); 609#endif 610 611 sbcompress(sb, m, sb->sb_mbtail); 612 613 sb->sb_lastrecord = sb->sb_mb; 614 SBLASTRECORDCHK(sb, __func__); 615} 616 617#ifdef SOCKBUF_DEBUG 618void 619sbcheck(struct sockbuf *sb) 620{ 621 struct mbuf *m; 622 u_long len, mbcnt; 623 624 len = 0; 625 mbcnt = 0; 626 for (m = sb->sb_mb; m; m = m->m_next) { 627 len += m->m_len; 628 mbcnt += MSIZE; 629 if (m->m_flags & M_EXT) 630 mbcnt += m->m_ext.ext_size; 631 if (m->m_nextpkt) 632 panic("sbcheck nextpkt"); 633 } 634 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 635 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc, 636 mbcnt, sb->sb_mbcnt); 637 panic("sbcheck"); 638 } 639} 640#endif 641 642/* 643 * As above, except the mbuf chain 644 * begins a new record. 645 */ 646void 647sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 648{ 649 struct mbuf *m; 650 651 if (m0 == 0) 652 return; 653 654#ifdef MBUFTRACE 655 m_claimm(m0, sb->sb_mowner); 656#endif 657 /* 658 * Put the first mbuf on the queue. 659 * Note this permits zero length records. 660 */ 661 sballoc(sb, m0); 662 SBLASTRECORDCHK(sb, "sbappendrecord 1"); 663 SBLINKRECORD(sb, m0); 664 m = m0->m_next; 665 m0->m_next = 0; 666 if (m && (m0->m_flags & M_EOR)) { 667 m0->m_flags &= ~M_EOR; 668 m->m_flags |= M_EOR; 669 } 670 sbcompress(sb, m, m0); 671 SBLASTRECORDCHK(sb, "sbappendrecord 2"); 672} 673 674/* 675 * As above except that OOB data 676 * is inserted at the beginning of the sockbuf, 677 * but after any other OOB data. 678 */ 679void 680sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 681{ 682 struct mbuf *m, **mp; 683 684 if (m0 == 0) 685 return; 686 687 SBLASTRECORDCHK(sb, "sbinsertoob 1"); 688 689 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) { 690 again: 691 switch (m->m_type) { 692 693 case MT_OOBDATA: 694 continue; /* WANT next train */ 695 696 case MT_CONTROL: 697 if ((m = m->m_next) != NULL) 698 goto again; /* inspect THIS train further */ 699 } 700 break; 701 } 702 /* 703 * Put the first mbuf on the queue. 704 * Note this permits zero length records. 705 */ 706 sballoc(sb, m0); 707 m0->m_nextpkt = *mp; 708 if (*mp == NULL) { 709 /* m0 is actually the new tail */ 710 sb->sb_lastrecord = m0; 711 } 712 *mp = m0; 713 m = m0->m_next; 714 m0->m_next = 0; 715 if (m && (m0->m_flags & M_EOR)) { 716 m0->m_flags &= ~M_EOR; 717 m->m_flags |= M_EOR; 718 } 719 sbcompress(sb, m, m0); 720 SBLASTRECORDCHK(sb, "sbinsertoob 2"); 721} 722 723/* 724 * Append address and data, and optionally, control (ancillary) data 725 * to the receive queue of a socket. If present, 726 * m0 must include a packet header with total length. 727 * Returns 0 if no space in sockbuf or insufficient mbufs. 728 */ 729int 730sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, 731 struct mbuf *control) 732{ 733 struct mbuf *m, *n, *nlast; 734 int space, len; 735 736 space = asa->sa_len; 737 738 if (m0 != NULL) { 739 if ((m0->m_flags & M_PKTHDR) == 0) 740 panic("sbappendaddr"); 741 space += m0->m_pkthdr.len; 742#ifdef MBUFTRACE 743 m_claimm(m0, sb->sb_mowner); 744#endif 745 } 746 for (n = control; n; n = n->m_next) { 747 space += n->m_len; 748 MCLAIM(n, sb->sb_mowner); 749 if (n->m_next == 0) /* keep pointer to last control buf */ 750 break; 751 } 752 if (space > sbspace(sb)) 753 return (0); 754 MGET(m, M_DONTWAIT, MT_SONAME); 755 if (m == 0) 756 return (0); 757 MCLAIM(m, sb->sb_mowner); 758 /* 759 * XXX avoid 'comparison always true' warning which isn't easily 760 * avoided. 761 */ 762 len = asa->sa_len; 763 if (len > MLEN) { 764 MEXTMALLOC(m, asa->sa_len, M_NOWAIT); 765 if ((m->m_flags & M_EXT) == 0) { 766 m_free(m); 767 return (0); 768 } 769 } 770 m->m_len = asa->sa_len; 771 memcpy(mtod(m, void *), asa, asa->sa_len); 772 if (n) 773 n->m_next = m0; /* concatenate data to control */ 774 else 775 control = m0; 776 m->m_next = control; 777 778 SBLASTRECORDCHK(sb, "sbappendaddr 1"); 779 780 for (n = m; n->m_next != NULL; n = n->m_next) 781 sballoc(sb, n); 782 sballoc(sb, n); 783 nlast = n; 784 SBLINKRECORD(sb, m); 785 786 sb->sb_mbtail = nlast; 787 SBLASTMBUFCHK(sb, "sbappendaddr"); 788 789 SBLASTRECORDCHK(sb, "sbappendaddr 2"); 790 791 return (1); 792} 793 794/* 795 * Helper for sbappendchainaddr: prepend a struct sockaddr* to 796 * an mbuf chain. 797 */ 798static inline struct mbuf * 799m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0, 800 const struct sockaddr *asa) 801{ 802 struct mbuf *m; 803 const int salen = asa->sa_len; 804 805 /* only the first in each chain need be a pkthdr */ 806 MGETHDR(m, M_DONTWAIT, MT_SONAME); 807 if (m == 0) 808 return (0); 809 MCLAIM(m, sb->sb_mowner); 810#ifdef notyet 811 if (salen > MHLEN) { 812 MEXTMALLOC(m, salen, M_NOWAIT); 813 if ((m->m_flags & M_EXT) == 0) { 814 m_free(m); 815 return (0); 816 } 817 } 818#else 819 KASSERT(salen <= MHLEN); 820#endif 821 m->m_len = salen; 822 memcpy(mtod(m, void *), asa, salen); 823 m->m_next = m0; 824 m->m_pkthdr.len = salen + m0->m_pkthdr.len; 825 826 return m; 827} 828 829int 830sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa, 831 struct mbuf *m0, int sbprio) 832{ 833 int space; 834 struct mbuf *m, *n, *n0, *nlast; 835 int error; 836 837 /* 838 * XXX sbprio reserved for encoding priority of this* request: 839 * SB_PRIO_NONE --> honour normal sb limits 840 * SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space, 841 * take whole chain. Intended for large requests 842 * that should be delivered atomically (all, or none). 843 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow 844 * over normal socket limits, for messages indicating 845 * buffer overflow in earlier normal/lower-priority messages 846 * SB_PRIO_BESTEFFORT --> ignore limits entirely. 847 * Intended for kernel-generated messages only. 848 * Up to generator to avoid total mbuf resource exhaustion. 849 */ 850 (void)sbprio; 851 852 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 853 panic("sbappendaddrchain"); 854 855 space = sbspace(sb); 856 857#ifdef notyet 858 /* 859 * Enforce SB_PRIO_* limits as described above. 860 */ 861#endif 862 863 n0 = NULL; 864 nlast = NULL; 865 for (m = m0; m; m = m->m_nextpkt) { 866 struct mbuf *np; 867 868#ifdef MBUFTRACE 869 m_claimm(m, sb->sb_mowner); 870#endif 871 872 /* Prepend sockaddr to this record (m) of input chain m0 */ 873 n = m_prepend_sockaddr(sb, m, asa); 874 if (n == NULL) { 875 error = ENOBUFS; 876 goto bad; 877 } 878 879 /* Append record (asa+m) to end of new chain n0 */ 880 if (n0 == NULL) { 881 n0 = n; 882 } else { 883 nlast->m_nextpkt = n; 884 } 885 /* Keep track of last record on new chain */ 886 nlast = n; 887 888 for (np = n; np; np = np->m_next) 889 sballoc(sb, np); 890 } 891 892 SBLASTRECORDCHK(sb, "sbappendaddrchain 1"); 893 894 /* Drop the entire chain of (asa+m) records onto the socket */ 895 SBLINKRECORDCHAIN(sb, n0, nlast); 896 897 SBLASTRECORDCHK(sb, "sbappendaddrchain 2"); 898 899 for (m = nlast; m->m_next; m = m->m_next) 900 ; 901 sb->sb_mbtail = m; 902 SBLASTMBUFCHK(sb, "sbappendaddrchain"); 903 904 return (1); 905 906bad: 907 /* 908 * On error, free the prepended addreseses. For consistency 909 * with sbappendaddr(), leave it to our caller to free 910 * the input record chain passed to us as m0. 911 */ 912 while ((n = n0) != NULL) { 913 struct mbuf *np; 914 915 /* Undo the sballoc() of this record */ 916 for (np = n; np; np = np->m_next) 917 sbfree(sb, np); 918 919 n0 = n->m_nextpkt; /* iterate at next prepended address */ 920 MFREE(n, np); /* free prepended address (not data) */ 921 } 922 return 0; 923} 924 925 926int 927sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 928{ 929 struct mbuf *m, *mlast, *n; 930 int space; 931 932 space = 0; 933 if (control == 0) 934 panic("sbappendcontrol"); 935 for (m = control; ; m = m->m_next) { 936 space += m->m_len; 937 MCLAIM(m, sb->sb_mowner); 938 if (m->m_next == 0) 939 break; 940 } 941 n = m; /* save pointer to last control buffer */ 942 for (m = m0; m; m = m->m_next) { 943 MCLAIM(m, sb->sb_mowner); 944 space += m->m_len; 945 } 946 if (space > sbspace(sb)) 947 return (0); 948 n->m_next = m0; /* concatenate data to control */ 949 950 SBLASTRECORDCHK(sb, "sbappendcontrol 1"); 951 952 for (m = control; m->m_next != NULL; m = m->m_next) 953 sballoc(sb, m); 954 sballoc(sb, m); 955 mlast = m; 956 SBLINKRECORD(sb, control); 957 958 sb->sb_mbtail = mlast; 959 SBLASTMBUFCHK(sb, "sbappendcontrol"); 960 961 SBLASTRECORDCHK(sb, "sbappendcontrol 2"); 962 963 return (1); 964} 965 966/* 967 * Compress mbuf chain m into the socket 968 * buffer sb following mbuf n. If n 969 * is null, the buffer is presumed empty. 970 */ 971void 972sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 973{ 974 int eor; 975 struct mbuf *o; 976 977 eor = 0; 978 while (m) { 979 eor |= m->m_flags & M_EOR; 980 if (m->m_len == 0 && 981 (eor == 0 || 982 (((o = m->m_next) || (o = n)) && 983 o->m_type == m->m_type))) { 984 if (sb->sb_lastrecord == m) 985 sb->sb_lastrecord = m->m_next; 986 m = m_free(m); 987 continue; 988 } 989 if (n && (n->m_flags & M_EOR) == 0 && 990 /* M_TRAILINGSPACE() checks buffer writeability */ 991 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */ 992 m->m_len <= M_TRAILINGSPACE(n) && 993 n->m_type == m->m_type) { 994 memcpy(mtod(n, char *) + n->m_len, mtod(m, void *), 995 (unsigned)m->m_len); 996 n->m_len += m->m_len; 997 sb->sb_cc += m->m_len; 998 m = m_free(m); 999 continue; 1000 } 1001 if (n) 1002 n->m_next = m; 1003 else 1004 sb->sb_mb = m; 1005 sb->sb_mbtail = m; 1006 sballoc(sb, m); 1007 n = m; 1008 m->m_flags &= ~M_EOR; 1009 m = m->m_next; 1010 n->m_next = 0; 1011 } 1012 if (eor) { 1013 if (n) 1014 n->m_flags |= eor; 1015 else 1016 printf("semi-panic: sbcompress\n"); 1017 } 1018 SBLASTMBUFCHK(sb, __func__); 1019} 1020 1021/* 1022 * Free all mbufs in a sockbuf. 1023 * Check that all resources are reclaimed. 1024 */ 1025void 1026sbflush(struct sockbuf *sb) 1027{ 1028 1029 KASSERT((sb->sb_flags & SB_LOCK) == 0); 1030 1031 while (sb->sb_mbcnt) 1032 sbdrop(sb, (int)sb->sb_cc); 1033 1034 KASSERT(sb->sb_cc == 0); 1035 KASSERT(sb->sb_mb == NULL); 1036 KASSERT(sb->sb_mbtail == NULL); 1037 KASSERT(sb->sb_lastrecord == NULL); 1038} 1039 1040/* 1041 * Drop data from (the front of) a sockbuf. 1042 */ 1043void 1044sbdrop(struct sockbuf *sb, int len) 1045{ 1046 struct mbuf *m, *mn, *next; 1047 1048 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1049 while (len > 0) { 1050 if (m == 0) { 1051 if (next == 0) 1052 panic("sbdrop"); 1053 m = next; 1054 next = m->m_nextpkt; 1055 continue; 1056 } 1057 if (m->m_len > len) { 1058 m->m_len -= len; 1059 m->m_data += len; 1060 sb->sb_cc -= len; 1061 break; 1062 } 1063 len -= m->m_len; 1064 sbfree(sb, m); 1065 MFREE(m, mn); 1066 m = mn; 1067 } 1068 while (m && m->m_len == 0) { 1069 sbfree(sb, m); 1070 MFREE(m, mn); 1071 m = mn; 1072 } 1073 if (m) { 1074 sb->sb_mb = m; 1075 m->m_nextpkt = next; 1076 } else 1077 sb->sb_mb = next; 1078 /* 1079 * First part is an inline SB_EMPTY_FIXUP(). Second part 1080 * makes sure sb_lastrecord is up-to-date if we dropped 1081 * part of the last record. 1082 */ 1083 m = sb->sb_mb; 1084 if (m == NULL) { 1085 sb->sb_mbtail = NULL; 1086 sb->sb_lastrecord = NULL; 1087 } else if (m->m_nextpkt == NULL) 1088 sb->sb_lastrecord = m; 1089} 1090 1091/* 1092 * Drop a record off the front of a sockbuf 1093 * and move the next record to the front. 1094 */ 1095void 1096sbdroprecord(struct sockbuf *sb) 1097{ 1098 struct mbuf *m, *mn; 1099 1100 m = sb->sb_mb; 1101 if (m) { 1102 sb->sb_mb = m->m_nextpkt; 1103 do { 1104 sbfree(sb, m); 1105 MFREE(m, mn); 1106 } while ((m = mn) != NULL); 1107 } 1108 SB_EMPTY_FIXUP(sb); 1109} 1110 1111/* 1112 * Create a "control" mbuf containing the specified data 1113 * with the specified type for presentation on a socket buffer. 1114 */ 1115struct mbuf * 1116sbcreatecontrol(void *p, int size, int type, int level) 1117{ 1118 struct cmsghdr *cp; 1119 struct mbuf *m; 1120 1121 if (CMSG_SPACE(size) > MCLBYTES) { 1122 printf("sbcreatecontrol: message too large %d\n", size); 1123 return NULL; 1124 } 1125 1126 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL) 1127 return ((struct mbuf *) NULL); 1128 if (CMSG_SPACE(size) > MLEN) { 1129 MCLGET(m, M_DONTWAIT); 1130 if ((m->m_flags & M_EXT) == 0) { 1131 m_free(m); 1132 return NULL; 1133 } 1134 } 1135 cp = mtod(m, struct cmsghdr *); 1136 memcpy(CMSG_DATA(cp), p, size); 1137 m->m_len = CMSG_SPACE(size); 1138 cp->cmsg_len = CMSG_LEN(size); 1139 cp->cmsg_level = level; 1140 cp->cmsg_type = type; 1141 return (m); 1142} 1143