1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_param.h" 36 37 #include <sys/param.h> 38 #include <sys/aio.h> /* for aio_swake proto */ 39 #include <sys/kernel.h> 40 #include <sys/lock.h> 41 #include <sys/mbuf.h> 42 #include <sys/mutex.h> 43 #include <sys/proc.h> 44 #include <sys/protosw.h> 45 #include <sys/resourcevar.h> 46 #include <sys/signalvar.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/sx.h> 50 #include <sys/sysctl.h> 51 52 /* 53 * Function pointer set by the AIO routines so that the socket buffer code 54 * can call back into the AIO module if it is loaded. 55 */ 56 void (*aio_swake)(struct socket *, struct sockbuf *); 57 58 /* 59 * Primitive routines for operating on socket buffers 60 */ 61 62 u_long sb_max = SB_MAX; 63 u_long sb_max_adj = 64 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 65 66 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 67 68 static void sbdrop_internal(struct sockbuf *sb, int len); 69 static void sbflush_internal(struct sockbuf *sb); 70 71 /* 72 * Socantsendmore indicates that no more data will be sent on the socket; it 73 * would normally be applied to a socket when the user informs the system 74 * that no more data is to be sent, by the protocol code (in case 75 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be 76 * received, and will normally be applied to the socket by a protocol when it 77 * detects that the peer will send no more data. Data queued for reading in 78 * the socket may yet be read. 79 */ 80 void 81 socantsendmore_locked(struct socket *so) 82 { 83 84 SOCKBUF_LOCK_ASSERT(&so->so_snd); 85 86 so->so_snd.sb_state |= SBS_CANTSENDMORE; 87 sowwakeup_locked(so); 88 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); 89 } 90 91 void 92 socantsendmore(struct socket *so) 93 { 94 95 SOCKBUF_LOCK(&so->so_snd); 96 socantsendmore_locked(so); 97 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); 98 } 99 100 void 101 socantrcvmore_locked(struct socket *so) 102 { 103 104 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 105 106 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 107 sorwakeup_locked(so); 108 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); 109 } 110 111 void 112 socantrcvmore(struct socket *so) 113 { 114 115 SOCKBUF_LOCK(&so->so_rcv); 116 socantrcvmore_locked(so); 117 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); 118 } 119 120 /* 121 * Wait for data to arrive at/drain from a socket buffer. 122 */ 123 int 124 sbwait(struct sockbuf *sb) 125 { 126 127 SOCKBUF_LOCK_ASSERT(sb); 128 129 sb->sb_flags |= SB_WAIT; 130 return (msleep(&sb->sb_cc, &sb->sb_mtx, 131 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", 132 sb->sb_timeo)); 133 } 134 135 int 136 sblock(struct sockbuf *sb, int flags) 137 { 138 139 KASSERT((flags & SBL_VALID) == flags, 140 ("sblock: flags invalid (0x%x)", flags)); 141 142 if (flags & SBL_WAIT) { 143 if ((sb->sb_flags & SB_NOINTR) || 144 (flags & SBL_NOINTR)) { 145 sx_xlock(&sb->sb_sx); 146 return (0); 147 } 148 return (sx_xlock_sig(&sb->sb_sx)); 149 } else { 150 if (sx_try_xlock(&sb->sb_sx) == 0) 151 return (EWOULDBLOCK); 152 return (0); 153 } 154 } 155 156 void 157 sbunlock(struct sockbuf *sb) 158 { 159 160 sx_xunlock(&sb->sb_sx); 161 } 162 163 /* 164 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 165 * via SIGIO if the socket has the SS_ASYNC flag set. 166 * 167 * Called with the socket buffer lock held; will release the lock by the end 168 * of the function. This allows the caller to acquire the socket buffer lock 169 * while testing for the need for various sorts of wakeup and hold it through 170 * to the point where it's no longer required. We currently hold the lock 171 * through calls out to other subsystems (with the exception of kqueue), and 172 * then release it to avoid lock order issues. It's not clear that's 173 * correct. 174 */ 175 void 176 sowakeup(struct socket *so, struct sockbuf *sb) 177 { 178 179 SOCKBUF_LOCK_ASSERT(sb); 180 181 selwakeuppri(&sb->sb_sel, PSOCK); 182 if (!SEL_WAITING(&sb->sb_sel)) 183 sb->sb_flags &= ~SB_SEL; 184 if (sb->sb_flags & SB_WAIT) { 185 sb->sb_flags &= ~SB_WAIT; 186 wakeup(&sb->sb_cc); 187 } 188 KNOTE_LOCKED(&sb->sb_sel.si_note, 0); 189 SOCKBUF_UNLOCK(sb); 190 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 191 pgsigio(&so->so_sigio, SIGIO, 0); 192 if (sb->sb_flags & SB_UPCALL) 193 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); 194 if (sb->sb_flags & SB_AIO) 195 aio_swake(so, sb); 196 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED); 197 } 198 199 /* 200 * Socket buffer (struct sockbuf) utility routines. 201 * 202 * Each socket contains two socket buffers: one for sending data and one for 203 * receiving data. Each buffer contains a queue of mbufs, information about 204 * the number of mbufs and amount of data in the queue, and other fields 205 * allowing select() statements and notification on data availability to be 206 * implemented. 207 * 208 * Data stored in a socket buffer is maintained as a list of records. Each 209 * record is a list of mbufs chained together with the m_next field. Records 210 * are chained together with the m_nextpkt field. The upper level routine 211 * soreceive() expects the following conventions to be observed when placing 212 * information in the receive buffer: 213 * 214 * 1. If the protocol requires each message be preceded by the sender's name, 215 * then a record containing that name must be present before any 216 * associated data (mbuf's must be of type MT_SONAME). 217 * 2. If the protocol supports the exchange of ``access rights'' (really just 218 * additional data associated with the message), and there are ``rights'' 219 * to be received, then a record containing this data should be present 220 * (mbuf's must be of type MT_RIGHTS). 221 * 3. If a name or rights record exists, then it must be followed by a data 222 * record, perhaps of zero length. 223 * 224 * Before using a new socket structure it is first necessary to reserve 225 * buffer space to the socket, by calling sbreserve(). This should commit 226 * some of the available buffer space in the system buffer pool for the 227 * socket (currently, it does nothing but enforce limits). The space should 228 * be released by calling sbrelease() when the socket is destroyed. 229 */ 230 int 231 soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 232 { 233 struct thread *td = curthread; 234 235 SOCKBUF_LOCK(&so->so_snd); 236 SOCKBUF_LOCK(&so->so_rcv); 237 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0) 238 goto bad; 239 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0) 240 goto bad2; 241 if (so->so_rcv.sb_lowat == 0) 242 so->so_rcv.sb_lowat = 1; 243 if (so->so_snd.sb_lowat == 0) 244 so->so_snd.sb_lowat = MCLBYTES; 245 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 246 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 247 SOCKBUF_UNLOCK(&so->so_rcv); 248 SOCKBUF_UNLOCK(&so->so_snd); 249 return (0); 250 bad2: 251 sbrelease_locked(&so->so_snd, so); 252 bad: 253 SOCKBUF_UNLOCK(&so->so_rcv); 254 SOCKBUF_UNLOCK(&so->so_snd); 255 return (ENOBUFS); 256 } 257 258 static int 259 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 260 { 261 int error = 0; 262 u_long tmp_sb_max = sb_max; 263 264 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); 265 if (error || !req->newptr) 266 return (error); 267 if (tmp_sb_max < MSIZE + MCLBYTES) 268 return (EINVAL); 269 sb_max = tmp_sb_max; 270 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 271 return (0); 272 } 273 274 /* 275 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't 276 * become limiting if buffering efficiency is near the normal case. 277 */ 278 int 279 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so, 280 struct thread *td) 281 { 282 rlim_t sbsize_limit; 283 284 SOCKBUF_LOCK_ASSERT(sb); 285 286 /* 287 * When a thread is passed, we take into account the thread's socket 288 * buffer size limit. The caller will generally pass curthread, but 289 * in the TCP input path, NULL will be passed to indicate that no 290 * appropriate thread resource limits are available. In that case, 291 * we don't apply a process limit. 292 */ 293 if (cc > sb_max_adj) 294 return (0); 295 if (td != NULL) { 296 PROC_LOCK(td->td_proc); 297 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE); 298 PROC_UNLOCK(td->td_proc); 299 } else 300 sbsize_limit = RLIM_INFINITY; 301 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 302 sbsize_limit)) 303 return (0); 304 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 305 if (sb->sb_lowat > sb->sb_hiwat) 306 sb->sb_lowat = sb->sb_hiwat; 307 return (1); 308 } 309 310 int 311 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so, 312 struct thread *td) 313 { 314 int error; 315 316 SOCKBUF_LOCK(sb); 317 error = sbreserve_locked(sb, cc, so, td); 318 SOCKBUF_UNLOCK(sb); 319 return (error); 320 } 321 322 /* 323 * Free mbufs held by a socket, and reserved mbuf space. 324 */ 325 void 326 sbrelease_internal(struct sockbuf *sb, struct socket *so) 327 { 328 329 sbflush_internal(sb); 330 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 331 RLIM_INFINITY); 332 sb->sb_mbmax = 0; 333 } 334 335 void 336 sbrelease_locked(struct sockbuf *sb, struct socket *so) 337 { 338 339 SOCKBUF_LOCK_ASSERT(sb); 340 341 sbrelease_internal(sb, so); 342 } 343 344 void 345 sbrelease(struct sockbuf *sb, struct socket *so) 346 { 347 348 SOCKBUF_LOCK(sb); 349 sbrelease_locked(sb, so); 350 SOCKBUF_UNLOCK(sb); 351 } 352 353 void 354 sbdestroy(struct sockbuf *sb, struct socket *so) 355 { 356 357 sbrelease_internal(sb, so); 358 } 359 360 /* 361 * Routines to add and remove data from an mbuf queue. 362 * 363 * The routines sbappend() or sbappendrecord() are normally called to append 364 * new mbufs to a socket buffer, after checking that adequate space is 365 * available, comparing the function sbspace() with the amount of data to be 366 * added. sbappendrecord() differs from sbappend() in that data supplied is 367 * treated as the beginning of a new record. To place a sender's address, 368 * optional access rights, and data in a socket receive buffer, 369 * sbappendaddr() should be used. To place access rights and data in a 370 * socket receive buffer, sbappendrights() should be used. In either case, 371 * the new data begins a new record. Note that unlike sbappend() and 372 * sbappendrecord(), these routines check for the caller that there will be 373 * enough space to store the data. Each fails if there is not enough space, 374 * or if it cannot find mbufs to store additional information in. 375 * 376 * Reliable protocols may use the socket send buffer to hold data awaiting 377 * acknowledgement. Data is normally copied from a socket send buffer in a 378 * protocol with m_copy for output to a peer, and then removing the data from 379 * the socket buffer with sbdrop() or sbdroprecord() when the data is 380 * acknowledged by the peer. 381 */ 382 #ifdef SOCKBUF_DEBUG 383 void 384 sblastrecordchk(struct sockbuf *sb, const char *file, int line) 385 { 386 struct mbuf *m = sb->sb_mb; 387 388 SOCKBUF_LOCK_ASSERT(sb); 389 390 while (m && m->m_nextpkt) 391 m = m->m_nextpkt; 392 393 if (m != sb->sb_lastrecord) { 394 printf("%s: sb_mb %p sb_lastrecord %p last %p\n", 395 __func__, sb->sb_mb, sb->sb_lastrecord, m); 396 printf("packet chain:\n"); 397 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 398 printf("\t%p\n", m); 399 panic("%s from %s:%u", __func__, file, line); 400 } 401 } 402 403 void 404 sblastmbufchk(struct sockbuf *sb, const char *file, int line) 405 { 406 struct mbuf *m = sb->sb_mb; 407 struct mbuf *n; 408 409 SOCKBUF_LOCK_ASSERT(sb); 410 411 while (m && m->m_nextpkt) 412 m = m->m_nextpkt; 413 414 while (m && m->m_next) 415 m = m->m_next; 416 417 if (m != sb->sb_mbtail) { 418 printf("%s: sb_mb %p sb_mbtail %p last %p\n", 419 __func__, sb->sb_mb, sb->sb_mbtail, m); 420 printf("packet tree:\n"); 421 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 422 printf("\t"); 423 for (n = m; n != NULL; n = n->m_next) 424 printf("%p ", n); 425 printf("\n"); 426 } 427 panic("%s from %s:%u", __func__, file, line); 428 } 429 } 430 #endif /* SOCKBUF_DEBUG */ 431 432 #define SBLINKRECORD(sb, m0) do { \ 433 SOCKBUF_LOCK_ASSERT(sb); \ 434 if ((sb)->sb_lastrecord != NULL) \ 435 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 436 else \ 437 (sb)->sb_mb = (m0); \ 438 (sb)->sb_lastrecord = (m0); \ 439 } while (/*CONSTCOND*/0) 440 441 /* 442 * Append mbuf chain m to the last record in the socket buffer sb. The 443 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 444 * are discarded and mbufs are compacted where possible. 445 */ 446 void 447 sbappend_locked(struct sockbuf *sb, struct mbuf *m) 448 { 449 struct mbuf *n; 450 451 SOCKBUF_LOCK_ASSERT(sb); 452 453 if (m == 0) 454 return; 455 456 SBLASTRECORDCHK(sb); 457 n = sb->sb_mb; 458 if (n) { 459 while (n->m_nextpkt) 460 n = n->m_nextpkt; 461 do { 462 if (n->m_flags & M_EOR) { 463 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 464 return; 465 } 466 } while (n->m_next && (n = n->m_next)); 467 } else { 468 /* 469 * XXX Would like to simply use sb_mbtail here, but 470 * XXX I need to verify that I won't miss an EOR that 471 * XXX way. 472 */ 473 if ((n = sb->sb_lastrecord) != NULL) { 474 do { 475 if (n->m_flags & M_EOR) { 476 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 477 return; 478 } 479 } while (n->m_next && (n = n->m_next)); 480 } else { 481 /* 482 * If this is the first record in the socket buffer, 483 * it's also the last record. 484 */ 485 sb->sb_lastrecord = m; 486 } 487 } 488 sbcompress(sb, m, n); 489 SBLASTRECORDCHK(sb); 490 } 491 492 /* 493 * Append mbuf chain m to the last record in the socket buffer sb. The 494 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 495 * are discarded and mbufs are compacted where possible. 496 */ 497 void 498 sbappend(struct sockbuf *sb, struct mbuf *m) 499 { 500 501 SOCKBUF_LOCK(sb); 502 sbappend_locked(sb, m); 503 SOCKBUF_UNLOCK(sb); 504 } 505 506 /* 507 * This version of sbappend() should only be used when the caller absolutely 508 * knows that there will never be more than one record in the socket buffer, 509 * that is, a stream protocol (such as TCP). 510 */ 511 void 512 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m) 513 { 514 SOCKBUF_LOCK_ASSERT(sb); 515 516 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); 517 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); 518 519 SBLASTMBUFCHK(sb); 520 521 sbcompress(sb, m, sb->sb_mbtail); 522 523 sb->sb_lastrecord = sb->sb_mb; 524 SBLASTRECORDCHK(sb); 525 } 526 527 /* 528 * This version of sbappend() should only be used when the caller absolutely 529 * knows that there will never be more than one record in the socket buffer, 530 * that is, a stream protocol (such as TCP). 531 */ 532 void 533 sbappendstream(struct sockbuf *sb, struct mbuf *m) 534 { 535 536 SOCKBUF_LOCK(sb); 537 sbappendstream_locked(sb, m); 538 SOCKBUF_UNLOCK(sb); 539 } 540 541 #ifdef SOCKBUF_DEBUG 542 void 543 sbcheck(struct sockbuf *sb) 544 { 545 struct mbuf *m; 546 struct mbuf *n = 0; 547 u_long len = 0, mbcnt = 0; 548 549 SOCKBUF_LOCK_ASSERT(sb); 550 551 for (m = sb->sb_mb; m; m = n) { 552 n = m->m_nextpkt; 553 for (; m; m = m->m_next) { 554 len += m->m_len; 555 mbcnt += MSIZE; 556 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 557 mbcnt += m->m_ext.ext_size; 558 } 559 } 560 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 561 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc, 562 mbcnt, sb->sb_mbcnt); 563 panic("sbcheck"); 564 } 565 } 566 #endif 567 568 /* 569 * As above, except the mbuf chain begins a new record. 570 */ 571 void 572 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) 573 { 574 struct mbuf *m; 575 576 SOCKBUF_LOCK_ASSERT(sb); 577 578 if (m0 == 0) 579 return; 580 m = sb->sb_mb; 581 if (m) 582 while (m->m_nextpkt) 583 m = m->m_nextpkt; 584 /* 585 * Put the first mbuf on the queue. Note this permits zero length 586 * records. 587 */ 588 sballoc(sb, m0); 589 SBLASTRECORDCHK(sb); 590 SBLINKRECORD(sb, m0); 591 if (m) 592 m->m_nextpkt = m0; 593 else 594 sb->sb_mb = m0; 595 m = m0->m_next; 596 m0->m_next = 0; 597 if (m && (m0->m_flags & M_EOR)) { 598 m0->m_flags &= ~M_EOR; 599 m->m_flags |= M_EOR; 600 } 601 sbcompress(sb, m, m0); 602 } 603 604 /* 605 * As above, except the mbuf chain begins a new record. 606 */ 607 void 608 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 609 { 610 611 SOCKBUF_LOCK(sb); 612 sbappendrecord_locked(sb, m0); 613 SOCKBUF_UNLOCK(sb); 614 } 615 616 /* 617 * Append address and data, and optionally, control (ancillary) data to the 618 * receive queue of a socket. If present, m0 must include a packet header 619 * with total length. Returns 0 if no space in sockbuf or insufficient 620 * mbufs. 621 */ 622 int 623 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, 624 struct mbuf *m0, struct mbuf *control) 625 { 626 struct mbuf *m, *n, *nlast; 627 int space = asa->sa_len; 628 629 SOCKBUF_LOCK_ASSERT(sb); 630 631 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 632 panic("sbappendaddr_locked"); 633 if (m0) 634 space += m0->m_pkthdr.len; 635 space += m_length(control, &n); 636 637 if (space > sbspace(sb)) 638 return (0); 639 #if MSIZE <= 256 640 if (asa->sa_len > MLEN) 641 return (0); 642 #endif 643 MGET(m, M_DONTWAIT, MT_SONAME); 644 if (m == 0) 645 return (0); 646 m->m_len = asa->sa_len; 647 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 648 if (n) 649 n->m_next = m0; /* concatenate data to control */ 650 else 651 control = m0; 652 m->m_next = control; 653 for (n = m; n->m_next != NULL; n = n->m_next) 654 sballoc(sb, n); 655 sballoc(sb, n); 656 nlast = n; 657 SBLINKRECORD(sb, m); 658 659 sb->sb_mbtail = nlast; 660 SBLASTMBUFCHK(sb); 661 662 SBLASTRECORDCHK(sb); 663 return (1); 664 } 665 666 /* 667 * Append address and data, and optionally, control (ancillary) data to the 668 * receive queue of a socket. If present, m0 must include a packet header 669 * with total length. Returns 0 if no space in sockbuf or insufficient 670 * mbufs. 671 */ 672 int 673 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, 674 struct mbuf *m0, struct mbuf *control) 675 { 676 int retval; 677 678 SOCKBUF_LOCK(sb); 679 retval = sbappendaddr_locked(sb, asa, m0, control); 680 SOCKBUF_UNLOCK(sb); 681 return (retval); 682 } 683 684 int 685 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, 686 struct mbuf *control) 687 { 688 struct mbuf *m, *n, *mlast; 689 int space; 690 691 SOCKBUF_LOCK_ASSERT(sb); 692 693 if (control == 0) 694 panic("sbappendcontrol_locked"); 695 space = m_length(control, &n) + m_length(m0, NULL); 696 697 if (space > sbspace(sb)) 698 return (0); 699 n->m_next = m0; /* concatenate data to control */ 700 701 SBLASTRECORDCHK(sb); 702 703 for (m = control; m->m_next; m = m->m_next) 704 sballoc(sb, m); 705 sballoc(sb, m); 706 mlast = m; 707 SBLINKRECORD(sb, control); 708 709 sb->sb_mbtail = mlast; 710 SBLASTMBUFCHK(sb); 711 712 SBLASTRECORDCHK(sb); 713 return (1); 714 } 715 716 int 717 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 718 { 719 int retval; 720 721 SOCKBUF_LOCK(sb); 722 retval = sbappendcontrol_locked(sb, m0, control); 723 SOCKBUF_UNLOCK(sb); 724 return (retval); 725 } 726 727 /* 728 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf 729 * (n). If (n) is NULL, the buffer is presumed empty. 730 * 731 * When the data is compressed, mbufs in the chain may be handled in one of 732 * three ways: 733 * 734 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no 735 * record boundary, and no change in data type). 736 * 737 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into 738 * an mbuf already in the socket buffer. This can occur if an 739 * appropriate mbuf exists, there is room, and no merging of data types 740 * will occur. 741 * 742 * (3) The mbuf may be appended to the end of the existing mbuf chain. 743 * 744 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as 745 * end-of-record. 746 */ 747 void 748 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 749 { 750 int eor = 0; 751 struct mbuf *o; 752 753 SOCKBUF_LOCK_ASSERT(sb); 754 755 while (m) { 756 eor |= m->m_flags & M_EOR; 757 if (m->m_len == 0 && 758 (eor == 0 || 759 (((o = m->m_next) || (o = n)) && 760 o->m_type == m->m_type))) { 761 if (sb->sb_lastrecord == m) 762 sb->sb_lastrecord = m->m_next; 763 m = m_free(m); 764 continue; 765 } 766 if (n && (n->m_flags & M_EOR) == 0 && 767 M_WRITABLE(n) && 768 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 769 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 770 m->m_len <= M_TRAILINGSPACE(n) && 771 n->m_type == m->m_type) { 772 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, 773 (unsigned)m->m_len); 774 n->m_len += m->m_len; 775 sb->sb_cc += m->m_len; 776 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 777 /* XXX: Probably don't need.*/ 778 sb->sb_ctl += m->m_len; 779 m = m_free(m); 780 continue; 781 } 782 if (n) 783 n->m_next = m; 784 else 785 sb->sb_mb = m; 786 sb->sb_mbtail = m; 787 sballoc(sb, m); 788 n = m; 789 m->m_flags &= ~M_EOR; 790 m = m->m_next; 791 n->m_next = 0; 792 } 793 if (eor) { 794 KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); 795 n->m_flags |= eor; 796 } 797 SBLASTMBUFCHK(sb); 798 } 799 800 /* 801 * Free all mbufs in a sockbuf. Check that all resources are reclaimed. 802 */ 803 static void 804 sbflush_internal(struct sockbuf *sb) 805 { 806 807 while (sb->sb_mbcnt) { 808 /* 809 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty: 810 * we would loop forever. Panic instead. 811 */ 812 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 813 break; 814 sbdrop_internal(sb, (int)sb->sb_cc); 815 } 816 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt) 817 panic("sbflush_internal: cc %u || mb %p || mbcnt %u", 818 sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt); 819 } 820 821 void 822 sbflush_locked(struct sockbuf *sb) 823 { 824 825 SOCKBUF_LOCK_ASSERT(sb); 826 sbflush_internal(sb); 827 } 828 829 void 830 sbflush(struct sockbuf *sb) 831 { 832 833 SOCKBUF_LOCK(sb); 834 sbflush_locked(sb); 835 SOCKBUF_UNLOCK(sb); 836 } 837 838 /* 839 * Drop data from (the front of) a sockbuf. 840 */ 841 static void 842 sbdrop_internal(struct sockbuf *sb, int len) 843 { 844 struct mbuf *m; 845 struct mbuf *next; 846 847 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 848 while (len > 0) { 849 if (m == 0) { 850 if (next == 0) 851 panic("sbdrop"); 852 m = next; 853 next = m->m_nextpkt; 854 continue; 855 } 856 if (m->m_len > len) { 857 m->m_len -= len; 858 m->m_data += len; 859 sb->sb_cc -= len; 860 if (sb->sb_sndptroff != 0) 861 sb->sb_sndptroff -= len; 862 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 863 sb->sb_ctl -= len; 864 break; 865 } 866 len -= m->m_len; 867 sbfree(sb, m); 868 m = m_free(m); 869 } 870 while (m && m->m_len == 0) { 871 sbfree(sb, m); 872 m = m_free(m); 873 } 874 if (m) { 875 sb->sb_mb = m; 876 m->m_nextpkt = next; 877 } else 878 sb->sb_mb = next; 879 /* 880 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure 881 * sb_lastrecord is up-to-date if we dropped part of the last record. 882 */ 883 m = sb->sb_mb; 884 if (m == NULL) { 885 sb->sb_mbtail = NULL; 886 sb->sb_lastrecord = NULL; 887 } else if (m->m_nextpkt == NULL) { 888 sb->sb_lastrecord = m; 889 } 890 } 891 892 /* 893 * Drop data from (the front of) a sockbuf. 894 */ 895 void 896 sbdrop_locked(struct sockbuf *sb, int len) 897 { 898 899 SOCKBUF_LOCK_ASSERT(sb); 900 901 sbdrop_internal(sb, len); 902 } 903 904 void 905 sbdrop(struct sockbuf *sb, int len) 906 { 907 908 SOCKBUF_LOCK(sb); 909 sbdrop_locked(sb, len); 910 SOCKBUF_UNLOCK(sb); 911 } 912 913 /* 914 * Maintain a pointer and offset pair into the socket buffer mbuf chain to 915 * avoid traversal of the entire socket buffer for larger offsets. 916 */ 917 struct mbuf * 918 sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff) 919 { 920 struct mbuf *m, *ret; 921 922 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 923 KASSERT(off + len <= sb->sb_cc, ("%s: beyond sb", __func__)); 924 KASSERT(sb->sb_sndptroff <= sb->sb_cc, ("%s: sndptroff broken", __func__)); 925 926 /* 927 * Is off below stored offset? Happens on retransmits. 928 * Just return, we can't help here. 929 */ 930 if (sb->sb_sndptroff > off) { 931 *moff = off; 932 return (sb->sb_mb); 933 } 934 935 /* Return closest mbuf in chain for current offset. */ 936 *moff = off - sb->sb_sndptroff; 937 m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb; 938 939 /* Advance by len to be as close as possible for the next transmit. */ 940 for (off = off - sb->sb_sndptroff + len - 1; 941 off > 0 && m != NULL && off >= m->m_len; 942 m = m->m_next) { 943 sb->sb_sndptroff += m->m_len; 944 off -= m->m_len; 945 } 946 if (off > 0 && m == NULL) 947 panic("%s: sockbuf %p and mbuf %p clashing", __func__, sb, ret); 948 sb->sb_sndptr = m; 949 950 return (ret); 951 } 952 953 /* 954 * Drop a record off the front of a sockbuf and move the next record to the 955 * front. 956 */ 957 void 958 sbdroprecord_locked(struct sockbuf *sb) 959 { 960 struct mbuf *m; 961 962 SOCKBUF_LOCK_ASSERT(sb); 963 964 m = sb->sb_mb; 965 if (m) { 966 sb->sb_mb = m->m_nextpkt; 967 do { 968 sbfree(sb, m); 969 m = m_free(m); 970 } while (m); 971 } 972 SB_EMPTY_FIXUP(sb); 973 } 974 975 /* 976 * Drop a record off the front of a sockbuf and move the next record to the 977 * front. 978 */ 979 void 980 sbdroprecord(struct sockbuf *sb) 981 { 982 983 SOCKBUF_LOCK(sb); 984 sbdroprecord_locked(sb); 985 SOCKBUF_UNLOCK(sb); 986 } 987 988 /* 989 * Create a "control" mbuf containing the specified data with the specified 990 * type for presentation on a socket buffer. 991 */ 992 struct mbuf * 993 sbcreatecontrol(caddr_t p, int size, int type, int level) 994 { 995 struct cmsghdr *cp; 996 struct mbuf *m; 997 998 if (CMSG_SPACE((u_int)size) > MCLBYTES) 999 return ((struct mbuf *) NULL); 1000 if (CMSG_SPACE((u_int)size) > MLEN) 1001 m = m_getcl(M_DONTWAIT, MT_CONTROL, 0); 1002 else 1003 m = m_get(M_DONTWAIT, MT_CONTROL); 1004 if (m == NULL) 1005 return ((struct mbuf *) NULL); 1006 cp = mtod(m, struct cmsghdr *); 1007 m->m_len = 0; 1008 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), 1009 ("sbcreatecontrol: short mbuf")); 1010 if (p != NULL) 1011 (void)memcpy(CMSG_DATA(cp), p, size); 1012 m->m_len = CMSG_SPACE(size); 1013 cp->cmsg_len = CMSG_LEN(size); 1014 cp->cmsg_level = level; 1015 cp->cmsg_type = type; 1016 return (m); 1017 } 1018 1019 /* 1020 * This does the same for socket buffers that sotoxsocket does for sockets: 1021 * generate an user-format data structure describing the socket buffer. Note 1022 * that the xsockbuf structure, since it is always embedded in a socket, does 1023 * not include a self pointer nor a length. We make this entry point public 1024 * in case some other mechanism needs it. 1025 */ 1026 void 1027 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 1028 { 1029 1030 xsb->sb_cc = sb->sb_cc; 1031 xsb->sb_hiwat = sb->sb_hiwat; 1032 xsb->sb_mbcnt = sb->sb_mbcnt; 1033 xsb->sb_mcnt = sb->sb_mcnt; 1034 xsb->sb_ccnt = sb->sb_ccnt; 1035 xsb->sb_mbmax = sb->sb_mbmax; 1036 xsb->sb_lowat = sb->sb_lowat; 1037 xsb->sb_flags = sb->sb_flags; 1038 xsb->sb_timeo = sb->sb_timeo; 1039 } 1040 1041 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 1042 static int dummy; 1043 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 1044 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW, 1045 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size"); 1046 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1047 &sb_efficiency, 0, ""); 1048