1 /* 2 * Copyright (c) 2005 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 1982, 1986, 1988, 1990, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. All advertising materials mentioning features or use of this software 15 * must display the following acknowledgement: 16 * This product includes software developed by the University of 17 * California, Berkeley and its contributors. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 35 * $FreeBSD: src/sys/kern/uipc_socket2.c,v 1.55.2.17 2002/08/31 19:04:55 dwmalone Exp $ 36 * $DragonFly: src/sys/kern/uipc_socket2.c,v 1.20 2005/05/29 15:52:26 hsu Exp $ 37 */ 38 39 #include "opt_param.h" 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/domain.h> 43 #include <sys/file.h> /* for maxfiles */ 44 #include <sys/kernel.h> 45 #include <sys/proc.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/protosw.h> 49 #include <sys/resourcevar.h> 50 #include <sys/stat.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/signalvar.h> 54 #include <sys/sysctl.h> 55 #include <sys/aio.h> /* for aio_swake proto */ 56 #include <sys/event.h> 57 58 #include <sys/thread2.h> 59 #include <sys/msgport2.h> 60 61 int maxsockets; 62 63 /* 64 * Primitive routines for operating on sockets and socket buffers 65 */ 66 67 u_long sb_max = SB_MAX; 68 u_long sb_max_adj = 69 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 70 71 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 72 73 /* 74 * Procedures to manipulate state flags of socket 75 * and do appropriate wakeups. Normal sequence from the 76 * active (originating) side is that soisconnecting() is 77 * called during processing of connect() call, 78 * resulting in an eventual call to soisconnected() if/when the 79 * connection is established. When the connection is torn down 80 * soisdisconnecting() is called during processing of disconnect() call, 81 * and soisdisconnected() is called when the connection to the peer 82 * is totally severed. The semantics of these routines are such that 83 * connectionless protocols can call soisconnected() and soisdisconnected() 84 * only, bypassing the in-progress calls when setting up a ``connection'' 85 * takes no time. 86 * 87 * From the passive side, a socket is created with 88 * two queues of sockets: so_incomp for connections in progress 89 * and so_comp for connections already made and awaiting user acceptance. 90 * As a protocol is preparing incoming connections, it creates a socket 91 * structure queued on so_incomp by calling sonewconn(). When the connection 92 * is established, soisconnected() is called, and transfers the 93 * socket structure to so_comp, making it available to accept(). 94 * 95 * If a socket is closed with sockets on either 96 * so_incomp or so_comp, these sockets are dropped. 97 * 98 * If higher level protocols are implemented in 99 * the kernel, the wakeups done here will sometimes 100 * cause software-interrupt process scheduling. 101 */ 102 103 void 104 soisconnecting(so) 105 struct socket *so; 106 { 107 108 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 109 so->so_state |= SS_ISCONNECTING; 110 } 111 112 void 113 soisconnected(so) 114 struct socket *so; 115 { 116 struct socket *head = so->so_head; 117 118 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 119 so->so_state |= SS_ISCONNECTED; 120 if (head && (so->so_state & SS_INCOMP)) { 121 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 122 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 123 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 124 so->so_rcv.sb_flags |= SB_UPCALL; 125 so->so_options &= ~SO_ACCEPTFILTER; 126 so->so_upcall(so, so->so_upcallarg, 0); 127 return; 128 } 129 TAILQ_REMOVE(&head->so_incomp, so, so_list); 130 head->so_incqlen--; 131 so->so_state &= ~SS_INCOMP; 132 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 133 head->so_qlen++; 134 so->so_state |= SS_COMP; 135 sorwakeup(head); 136 wakeup_one(&head->so_timeo); 137 } else { 138 wakeup(&so->so_timeo); 139 sorwakeup(so); 140 sowwakeup(so); 141 } 142 } 143 144 void 145 soisdisconnecting(so) 146 struct socket *so; 147 { 148 149 so->so_state &= ~SS_ISCONNECTING; 150 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 151 wakeup((caddr_t)&so->so_timeo); 152 sowwakeup(so); 153 sorwakeup(so); 154 } 155 156 void 157 soisdisconnected(so) 158 struct socket *so; 159 { 160 161 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 162 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 163 wakeup((caddr_t)&so->so_timeo); 164 sbdrop(&so->so_snd, so->so_snd.sb_cc); 165 sowwakeup(so); 166 sorwakeup(so); 167 } 168 169 /* 170 * When an attempt at a new connection is noted on a socket 171 * which accepts connections, sonewconn is called. If the 172 * connection is possible (subject to space constraints, etc.) 173 * then we allocate a new structure, propoerly linked into the 174 * data structure of the original socket, and return this. 175 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 176 */ 177 struct socket * 178 sonewconn(struct socket *head, int connstatus) 179 { 180 struct socket *so; 181 struct pru_attach_info ai; 182 183 if (head->so_qlen > 3 * head->so_qlimit / 2) 184 return ((struct socket *)0); 185 so = soalloc(0); 186 if (so == NULL) 187 return ((struct socket *)0); 188 if ((head->so_options & SO_ACCEPTFILTER) != 0) 189 connstatus = 0; 190 so->so_head = head; 191 so->so_type = head->so_type; 192 so->so_options = head->so_options &~ SO_ACCEPTCONN; 193 so->so_linger = head->so_linger; 194 so->so_state = head->so_state | SS_NOFDREF; 195 so->so_proto = head->so_proto; 196 so->so_timeo = head->so_timeo; 197 so->so_cred = crhold(head->so_cred); 198 ai.sb_rlimit = NULL; 199 ai.p_ucred = NULL; 200 ai.fd_rdir = NULL; /* jail code cruft XXX JH */ 201 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat, NULL) || 202 /* Directly call function since we're already at protocol level. */ 203 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, &ai)) { 204 sodealloc(so); 205 return ((struct socket *)0); 206 } 207 208 if (connstatus) { 209 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 210 so->so_state |= SS_COMP; 211 head->so_qlen++; 212 } else { 213 if (head->so_incqlen > head->so_qlimit) { 214 struct socket *sp; 215 sp = TAILQ_FIRST(&head->so_incomp); 216 (void) soabort(sp); 217 } 218 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 219 so->so_state |= SS_INCOMP; 220 head->so_incqlen++; 221 } 222 if (connstatus) { 223 sorwakeup(head); 224 wakeup((caddr_t)&head->so_timeo); 225 so->so_state |= connstatus; 226 } 227 return (so); 228 } 229 230 /* 231 * Socantsendmore indicates that no more data will be sent on the 232 * socket; it would normally be applied to a socket when the user 233 * informs the system that no more data is to be sent, by the protocol 234 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 235 * will be received, and will normally be applied to the socket by a 236 * protocol when it detects that the peer will send no more data. 237 * Data queued for reading in the socket may yet be read. 238 */ 239 240 void 241 socantsendmore(so) 242 struct socket *so; 243 { 244 245 so->so_state |= SS_CANTSENDMORE; 246 sowwakeup(so); 247 } 248 249 void 250 socantrcvmore(so) 251 struct socket *so; 252 { 253 254 so->so_state |= SS_CANTRCVMORE; 255 sorwakeup(so); 256 } 257 258 /* 259 * Wait for data to arrive at/drain from a socket buffer. 260 */ 261 int 262 sbwait(sb) 263 struct sockbuf *sb; 264 { 265 266 sb->sb_flags |= SB_WAIT; 267 return (tsleep((caddr_t)&sb->sb_cc, 268 ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), 269 "sbwait", 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 */ 277 int 278 sb_lock(sb) 279 struct sockbuf *sb; 280 { 281 int error; 282 283 while (sb->sb_flags & SB_LOCK) { 284 sb->sb_flags |= SB_WANT; 285 error = tsleep((caddr_t)&sb->sb_flags, 286 ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), 287 "sblock", 0); 288 if (error) 289 return (error); 290 } 291 sb->sb_flags |= SB_LOCK; 292 return (0); 293 } 294 295 /* 296 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 297 * via SIGIO if the socket has the SS_ASYNC flag set. 298 */ 299 void 300 sowakeup(so, sb) 301 struct socket *so; 302 struct sockbuf *sb; 303 { 304 struct selinfo *selinfo = &sb->sb_sel; 305 306 selwakeup(selinfo); 307 sb->sb_flags &= ~SB_SEL; 308 if (sb->sb_flags & SB_WAIT) { 309 sb->sb_flags &= ~SB_WAIT; 310 wakeup((caddr_t)&sb->sb_cc); 311 } 312 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 313 pgsigio(so->so_sigio, SIGIO, 0); 314 if (sb->sb_flags & SB_UPCALL) 315 (*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT); 316 if (sb->sb_flags & SB_AIO) 317 aio_swake(so, sb); 318 KNOTE(&selinfo->si_note, 0); 319 if (sb->sb_flags & SB_MEVENT) { 320 struct netmsg_so_notify *msg, *nmsg; 321 322 TAILQ_FOREACH_MUTABLE(msg, &selinfo->si_mlist, nm_list, nmsg) { 323 if (msg->nm_predicate((struct netmsg *)msg)) { 324 TAILQ_REMOVE(&selinfo->si_mlist, msg, nm_list); 325 lwkt_replymsg(&msg->nm_lmsg, 326 msg->nm_lmsg.ms_error); 327 } 328 } 329 if (TAILQ_EMPTY(&sb->sb_sel.si_mlist)) 330 sb->sb_flags &= ~SB_MEVENT; 331 } 332 } 333 334 /* 335 * Socket buffer (struct sockbuf) utility routines. 336 * 337 * Each socket contains two socket buffers: one for sending data and 338 * one for receiving data. Each buffer contains a queue of mbufs, 339 * information about the number of mbufs and amount of data in the 340 * queue, and other fields allowing select() statements and notification 341 * on data availability to be implemented. 342 * 343 * Data stored in a socket buffer is maintained as a list of records. 344 * Each record is a list of mbufs chained together with the m_next 345 * field. Records are chained together with the m_nextpkt field. The upper 346 * level routine soreceive() expects the following conventions to be 347 * observed when placing information in the receive buffer: 348 * 349 * 1. If the protocol requires each message be preceded by the sender's 350 * name, then a record containing that name must be present before 351 * any associated data (mbuf's must be of type MT_SONAME). 352 * 2. If the protocol supports the exchange of ``access rights'' (really 353 * just additional data associated with the message), and there are 354 * ``rights'' to be received, then a record containing this data 355 * should be present (mbuf's must be of type MT_RIGHTS). 356 * 3. If a name or rights record exists, then it must be followed by 357 * a data record, perhaps of zero length. 358 * 359 * Before using a new socket structure it is first necessary to reserve 360 * buffer space to the socket, by calling sbreserve(). This should commit 361 * some of the available buffer space in the system buffer pool for the 362 * socket (currently, it does nothing but enforce limits). The space 363 * should be released by calling sbrelease() when the socket is destroyed. 364 */ 365 366 int 367 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl) 368 { 369 if (sbreserve(&so->so_snd, sndcc, so, rl) == 0) 370 goto bad; 371 if (sbreserve(&so->so_rcv, rcvcc, so, rl) == 0) 372 goto bad2; 373 if (so->so_rcv.sb_lowat == 0) 374 so->so_rcv.sb_lowat = 1; 375 if (so->so_snd.sb_lowat == 0) 376 so->so_snd.sb_lowat = MCLBYTES; 377 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 378 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 379 return (0); 380 bad2: 381 sbrelease(&so->so_snd, so); 382 bad: 383 return (ENOBUFS); 384 } 385 386 static int 387 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 388 { 389 int error = 0; 390 u_long old_sb_max = sb_max; 391 392 error = SYSCTL_OUT(req, arg1, sizeof(int)); 393 if (error || !req->newptr) 394 return (error); 395 error = SYSCTL_IN(req, arg1, sizeof(int)); 396 if (error) 397 return (error); 398 if (sb_max < MSIZE + MCLBYTES) { 399 sb_max = old_sb_max; 400 return (EINVAL); 401 } 402 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 403 return (0); 404 } 405 406 /* 407 * Allot mbufs to a sockbuf. 408 * Attempt to scale mbmax so that mbcnt doesn't become limiting 409 * if buffering efficiency is near the normal case. 410 */ 411 int 412 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so, struct rlimit *rl) 413 { 414 415 /* 416 * rl will only be NULL when we're in an interrupt (eg, in tcp_input) 417 * or when called from netgraph (ie, ngd_attach) 418 */ 419 if (cc > sb_max_adj) 420 return (0); 421 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 422 rl ? rl->rlim_cur : RLIM_INFINITY)) { 423 return (0); 424 } 425 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 426 if (sb->sb_lowat > sb->sb_hiwat) 427 sb->sb_lowat = sb->sb_hiwat; 428 return (1); 429 } 430 431 /* 432 * Free mbufs held by a socket, and reserved mbuf space. 433 */ 434 void 435 sbrelease(sb, so) 436 struct sockbuf *sb; 437 struct socket *so; 438 { 439 440 sbflush(sb); 441 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 442 RLIM_INFINITY); 443 sb->sb_mbmax = 0; 444 } 445 446 /* 447 * Routines to add and remove 448 * data from an mbuf queue. 449 * 450 * The routines sbappend() or sbappendrecord() are normally called to 451 * append new mbufs to a socket buffer, after checking that adequate 452 * space is available, comparing the function sbspace() with the amount 453 * of data to be added. sbappendrecord() differs from sbappend() in 454 * that data supplied is treated as the beginning of a new record. 455 * To place a sender's address, optional access rights, and data in a 456 * socket receive buffer, sbappendaddr() should be used. To place 457 * access rights and data in a socket receive buffer, sbappendrights() 458 * should be used. In either case, the new data begins a new record. 459 * Note that unlike sbappend() and sbappendrecord(), these routines check 460 * for the caller that there will be enough space to store the data. 461 * Each fails if there is not enough space, or if it cannot find mbufs 462 * to store additional information in. 463 * 464 * Reliable protocols may use the socket send buffer to hold data 465 * awaiting acknowledgement. Data is normally copied from a socket 466 * send buffer in a protocol with m_copy for output to a peer, 467 * and then removing the data from the socket buffer with sbdrop() 468 * or sbdroprecord() when the data is acknowledged by the peer. 469 */ 470 471 /* 472 * Append mbuf chain m to the last record in the 473 * socket buffer sb. The additional space associated 474 * the mbuf chain is recorded in sb. Empty mbufs are 475 * discarded and mbufs are compacted where possible. 476 */ 477 void 478 sbappend(struct sockbuf *sb, struct mbuf *m) 479 { 480 struct mbuf *n; 481 482 if (m == NULL) 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 if (n == NULL) 497 sb->sb_lastrecord = sb->sb_mb; 498 } 499 500 /* 501 * sbappendstream() is an optimized form of sbappend() for protocols 502 * such as TCP that only have one record in the socket buffer, are 503 * not PR_ATOMIC, nor allow MT_CONTROL data. A protocol that uses 504 * sbappendstream() must use sbappendstream() exclusively. 505 */ 506 void 507 sbappendstream(struct sockbuf *sb, struct mbuf *m) 508 { 509 KKASSERT(m->m_nextpkt == NULL); 510 sbcompress(sb, m, sb->sb_lastmbuf); 511 } 512 513 #ifdef SOCKBUF_DEBUG 514 void 515 sbcheck(sb) 516 struct sockbuf *sb; 517 { 518 struct mbuf *m; 519 struct mbuf *n = 0; 520 u_long len = 0, mbcnt = 0; 521 522 for (m = sb->sb_mb; m; m = n) { 523 n = m->m_nextpkt; 524 for (; m; m = m->m_next) { 525 len += m->m_len; 526 mbcnt += MSIZE; 527 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 528 mbcnt += m->m_ext.ext_size; 529 } 530 } 531 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 532 printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc, 533 mbcnt, sb->sb_mbcnt); 534 panic("sbcheck"); 535 } 536 } 537 #endif 538 539 /* 540 * Same as sbappend(), except the mbuf chain begins a new record. 541 */ 542 void 543 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 544 { 545 struct mbuf *firstmbuf; 546 struct mbuf *secondmbuf; 547 548 if (m0 == NULL) 549 return; 550 551 /* 552 * Break the first mbuf off from the rest of the mbuf chain. 553 */ 554 firstmbuf = m0; 555 secondmbuf = m0->m_next; 556 m0->m_next = NULL; 557 558 /* 559 * Insert the first mbuf of the m0 mbuf chain as the last record of 560 * the sockbuf. Note this permits zero length records! 561 */ 562 if (sb->sb_mb == NULL) 563 sb->sb_mb = firstmbuf; 564 else 565 sb->sb_lastrecord->m_nextpkt = firstmbuf; 566 sb->sb_lastrecord = firstmbuf; /* update hint for new last record */ 567 568 if ((firstmbuf->m_flags & M_EOR) && (secondmbuf != NULL)) { 569 /* propagate the EOR flag */ 570 firstmbuf->m_flags &= ~M_EOR; 571 secondmbuf->m_flags |= M_EOR; 572 } 573 574 /* 575 * The succeeding call to sbcompress() omits accounting for 576 * the first mbuf, so do it here. 577 */ 578 sballoc(sb, firstmbuf); 579 580 /* Compact the rest of the mbuf chain in after the first mbuf. */ 581 sbcompress(sb, secondmbuf, firstmbuf); 582 } 583 584 /* 585 * As above except that OOB data is inserted at the beginning of the sockbuf, 586 * but after any other OOB data. 587 */ 588 void 589 sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 590 { 591 struct mbuf *m; 592 struct mbuf **mp; 593 594 if (m0 == 0) 595 return; 596 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) { 597 m = *mp; 598 again: 599 switch (m->m_type) { 600 601 case MT_OOBDATA: 602 continue; /* WANT next train */ 603 604 case MT_CONTROL: 605 m = m->m_next; 606 if (m) 607 goto again; /* inspect THIS train further */ 608 } 609 break; 610 } 611 /* 612 * Put the first mbuf on the queue. 613 * Note this permits zero length records. 614 */ 615 sballoc(sb, m0); 616 m0->m_nextpkt = *mp; 617 *mp = m0; 618 if (m0->m_nextpkt == NULL) 619 sb->sb_lastrecord = m0; 620 621 m = m0->m_next; 622 m0->m_next = 0; 623 if (m && (m0->m_flags & M_EOR)) { 624 m0->m_flags &= ~M_EOR; 625 m->m_flags |= M_EOR; 626 } 627 sbcompress(sb, m, m0); 628 } 629 630 /* 631 * Append address and data, and optionally, control (ancillary) data 632 * to the receive queue of a socket. If present, 633 * m0 must include a packet header with total length. 634 * Returns 0 if no space in sockbuf or insufficient mbufs. 635 */ 636 int 637 sbappendaddr(sb, asa, m0, control) 638 struct sockbuf *sb; 639 const struct sockaddr *asa; 640 struct mbuf *m0, *control; 641 { 642 struct mbuf *m, *n; 643 int space = asa->sa_len; 644 645 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 646 panic("sbappendaddr"); 647 648 if (m0) 649 space += m0->m_pkthdr.len; 650 for (n = control; n; n = n->m_next) { 651 space += n->m_len; 652 if (n->m_next == 0) /* keep pointer to last control buf */ 653 break; 654 } 655 if (space > sbspace(sb)) 656 return (0); 657 if (asa->sa_len > MLEN) 658 return (0); 659 MGET(m, MB_DONTWAIT, MT_SONAME); 660 if (m == 0) 661 return (0); 662 m->m_len = asa->sa_len; 663 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 664 if (n) 665 n->m_next = m0; /* concatenate data to control */ 666 else 667 control = m0; 668 m->m_next = control; 669 for (n = m; n; n = n->m_next) 670 sballoc(sb, n); 671 672 if (sb->sb_mb == NULL) 673 sb->sb_mb = m; 674 else 675 sb->sb_lastrecord->m_nextpkt = m; 676 sb->sb_lastrecord = m; 677 678 return (1); 679 } 680 681 /* 682 * Append control information followed by data. 683 * control must be non-null. 684 */ 685 int 686 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 687 { 688 struct mbuf *n; 689 u_int length, cmbcnt, m0mbcnt; 690 691 KASSERT(control != NULL, ("sbappendcontrol")); 692 693 length = m_countm(control, &n, &cmbcnt) + m_countm(m0, NULL, &m0mbcnt); 694 if (length > sbspace(sb)) 695 return (0); 696 697 n->m_next = m0; /* concatenate data to control */ 698 699 if (sb->sb_mb == NULL) 700 sb->sb_mb = control; 701 else 702 sb->sb_lastrecord->m_nextpkt = control; 703 sb->sb_lastrecord = control; 704 705 sb->sb_cc += length; 706 sb->sb_mbcnt += cmbcnt + m0mbcnt; 707 708 return (1); 709 } 710 711 /* 712 * Compress mbuf chain m into the socket buffer sb following mbuf tailm. 713 * If tailm is null, the buffer is presumed empty. Also, as a side-effect, 714 * increment the sockbuf counts for each mbuf in the chain. 715 */ 716 void 717 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *tailm) 718 { 719 int eor = 0; 720 721 while (m) { 722 struct mbuf *o; 723 724 eor |= m->m_flags & M_EOR; 725 /* 726 * Disregard empty mbufs as long as we don't encounter 727 * an end-of-record or there is a trailing mbuf of 728 * the same type to propagate the EOR flag to. 729 */ 730 if (m->m_len == 0 && 731 (eor == 0 || 732 (((o = m->m_next) || (o = tailm)) && 733 o->m_type == m->m_type))) { 734 m = m_free(m); 735 continue; 736 } 737 738 /* See if we can coalesce with preceding mbuf. */ 739 if (tailm && !(tailm->m_flags & M_EOR) && M_WRITABLE(tailm) && 740 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 741 m->m_len <= M_TRAILINGSPACE(tailm) && 742 tailm->m_type == m->m_type) { 743 bcopy(mtod(m, caddr_t), 744 mtod(tailm, caddr_t) + tailm->m_len, 745 (unsigned)m->m_len); 746 tailm->m_len += m->m_len; 747 sb->sb_cc += m->m_len; /* update sb counter */ 748 m = m_free(m); 749 continue; 750 } 751 752 /* Insert whole mbuf. */ 753 if (tailm == NULL) { 754 KASSERT(sb->sb_mb == NULL, 755 ("sbcompress: sb_mb not NULL")); 756 sb->sb_mb = m; /* put at front of sockbuf */ 757 } else { 758 tailm->m_next = m; /* tack m on following tailm */ 759 } 760 sb->sb_lastmbuf = m; /* update last mbuf hint */ 761 762 tailm = m; /* just inserted mbuf becomes the new tail */ 763 m = m->m_next; /* advance to next mbuf */ 764 tailm->m_next = NULL; /* split inserted mbuf off from chain */ 765 766 /* update sb counters for just added mbuf */ 767 sballoc(sb, tailm); 768 769 /* clear EOR on intermediate mbufs */ 770 tailm->m_flags &= ~M_EOR; 771 } 772 773 if (eor) { 774 if (tailm) 775 tailm->m_flags |= eor; /* propagate EOR to last mbuf */ 776 else 777 printf("semi-panic: sbcompress"); 778 } 779 } 780 781 /* 782 * Free all mbufs in a sockbuf. 783 * Check that all resources are reclaimed. 784 */ 785 void 786 sbflush(sb) 787 struct sockbuf *sb; 788 { 789 790 if (sb->sb_flags & SB_LOCK) 791 panic("sbflush: locked"); 792 while (sb->sb_mbcnt) { 793 /* 794 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty: 795 * we would loop forever. Panic instead. 796 */ 797 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 798 break; 799 sbdrop(sb, (int)sb->sb_cc); 800 } 801 KASSERT(!(sb->sb_cc || sb->sb_mb || sb->sb_mbcnt || sb->sb_lastmbuf), 802 ("sbflush: cc %ld || mb %p || mbcnt %ld || lastmbuf %p", 803 sb->sb_cc, sb->sb_mb, sb->sb_mbcnt, sb->sb_lastmbuf)); 804 } 805 806 /* 807 * Drop data from (the front of) a sockbuf. 808 */ 809 void 810 sbdrop(sb, len) 811 struct sockbuf *sb; 812 int len; 813 { 814 struct mbuf *m; 815 struct mbuf *nextpkt; 816 817 m = sb->sb_mb; 818 nextpkt = (m != NULL) ? m->m_nextpkt : NULL; 819 while (len > 0) { 820 if (m == NULL) { 821 if (nextpkt == NULL) 822 panic("sbdrop"); 823 m = nextpkt; 824 nextpkt = m->m_nextpkt; 825 m->m_nextpkt = NULL; 826 continue; 827 } 828 if (m->m_len > len) { 829 m->m_len -= len; 830 m->m_data += len; 831 sb->sb_cc -= len; 832 break; 833 } 834 len -= m->m_len; 835 sbfree(sb, m); 836 m = m_free(m); 837 } 838 while (m && m->m_len == 0) { 839 sbfree(sb, m); 840 m = m_free(m); 841 } 842 if (m != NULL) { 843 sb->sb_mb = m; 844 m->m_nextpkt = nextpkt; 845 } else { 846 sb->sb_mb = nextpkt; 847 sb->sb_lastmbuf = NULL; /* invalidate hint */ 848 } 849 } 850 851 /* 852 * Drop a record off the front of a sockbuf 853 * and move the next record to the front. 854 */ 855 void 856 sbdroprecord(sb) 857 struct sockbuf *sb; 858 { 859 struct mbuf *m; 860 861 m = sb->sb_mb; 862 if (m) { 863 sb->sb_mb = m->m_nextpkt; 864 do { 865 sbfree(sb, m); 866 m = m_free(m); 867 } while (m); 868 } 869 } 870 871 /* 872 * Create a "control" mbuf containing the specified data 873 * with the specified type for presentation on a socket buffer. 874 */ 875 struct mbuf * 876 sbcreatecontrol(p, size, type, level) 877 caddr_t p; 878 int size; 879 int type, level; 880 { 881 struct cmsghdr *cp; 882 struct mbuf *m; 883 884 if (CMSG_SPACE((u_int)size) > MCLBYTES) 885 return (NULL); 886 m = m_getl(CMSG_SPACE((u_int)size), MB_DONTWAIT, MT_CONTROL, 0, NULL); 887 if (m == NULL) 888 return (NULL); 889 m->m_len = CMSG_SPACE(size); 890 cp = mtod(m, struct cmsghdr *); 891 if (p != NULL) 892 memcpy(CMSG_DATA(cp), p, size); 893 cp->cmsg_len = CMSG_LEN(size); 894 cp->cmsg_level = level; 895 cp->cmsg_type = type; 896 return (m); 897 } 898 899 /* 900 * Some routines that return EOPNOTSUPP for entry points that are not 901 * supported by a protocol. Fill in as needed. 902 */ 903 int 904 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 905 { 906 return EOPNOTSUPP; 907 } 908 909 int 910 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 911 { 912 return EOPNOTSUPP; 913 } 914 915 int 916 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 917 { 918 return EOPNOTSUPP; 919 } 920 921 int 922 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 923 struct ifnet *ifp, struct thread *td) 924 { 925 return EOPNOTSUPP; 926 } 927 928 int 929 pru_listen_notsupp(struct socket *so, struct thread *td) 930 { 931 return EOPNOTSUPP; 932 } 933 934 int 935 pru_rcvd_notsupp(struct socket *so, int flags) 936 { 937 return EOPNOTSUPP; 938 } 939 940 int 941 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 942 { 943 return EOPNOTSUPP; 944 } 945 946 /* 947 * This isn't really a ``null'' operation, but it's the default one 948 * and doesn't do anything destructive. 949 */ 950 int 951 pru_sense_null(struct socket *so, struct stat *sb) 952 { 953 sb->st_blksize = so->so_snd.sb_hiwat; 954 return 0; 955 } 956 957 /* 958 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. Callers 959 * of this routine assume that it always succeeds, so we have to use a 960 * blockable allocation even though we might be called from a critical thread. 961 */ 962 struct sockaddr * 963 dup_sockaddr(const struct sockaddr *sa) 964 { 965 struct sockaddr *sa2; 966 967 sa2 = malloc(sa->sa_len, M_SONAME, M_INTWAIT); 968 bcopy(sa, sa2, sa->sa_len); 969 return (sa2); 970 } 971 972 /* 973 * Create an external-format (``xsocket'') structure using the information 974 * in the kernel-format socket structure pointed to by so. This is done 975 * to reduce the spew of irrelevant information over this interface, 976 * to isolate user code from changes in the kernel structure, and 977 * potentially to provide information-hiding if we decide that 978 * some of this information should be hidden from users. 979 */ 980 void 981 sotoxsocket(struct socket *so, struct xsocket *xso) 982 { 983 xso->xso_len = sizeof *xso; 984 xso->xso_so = so; 985 xso->so_type = so->so_type; 986 xso->so_options = so->so_options; 987 xso->so_linger = so->so_linger; 988 xso->so_state = so->so_state; 989 xso->so_pcb = so->so_pcb; 990 xso->xso_protocol = so->so_proto->pr_protocol; 991 xso->xso_family = so->so_proto->pr_domain->dom_family; 992 xso->so_qlen = so->so_qlen; 993 xso->so_incqlen = so->so_incqlen; 994 xso->so_qlimit = so->so_qlimit; 995 xso->so_timeo = so->so_timeo; 996 xso->so_error = so->so_error; 997 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 998 xso->so_oobmark = so->so_oobmark; 999 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 1000 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 1001 xso->so_uid = so->so_cred->cr_uid; 1002 } 1003 1004 /* 1005 * This does the same for sockbufs. Note that the xsockbuf structure, 1006 * since it is always embedded in a socket, does not include a self 1007 * pointer nor a length. We make this entry point public in case 1008 * some other mechanism needs it. 1009 */ 1010 void 1011 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 1012 { 1013 xsb->sb_cc = sb->sb_cc; 1014 xsb->sb_hiwat = sb->sb_hiwat; 1015 xsb->sb_mbcnt = sb->sb_mbcnt; 1016 xsb->sb_mbmax = sb->sb_mbmax; 1017 xsb->sb_lowat = sb->sb_lowat; 1018 xsb->sb_flags = sb->sb_flags; 1019 xsb->sb_timeo = sb->sb_timeo; 1020 } 1021 1022 /* 1023 * Here is the definition of some of the basic objects in the kern.ipc 1024 * branch of the MIB. 1025 */ 1026 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 1027 1028 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 1029 static int dummy; 1030 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 1031 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, 1032 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); 1033 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, 1034 &maxsockets, 0, "Maximum number of sockets avaliable"); 1035 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1036 &sb_efficiency, 0, ""); 1037 1038 /* 1039 * Initialise maxsockets 1040 */ 1041 static void init_maxsockets(void *ignored) 1042 { 1043 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 1044 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 1045 } 1046 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 1047