1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 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 8.1 (Berkeley) 6/10/93 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_param.h" 38 39 #include <sys/param.h> 40 #include <sys/aio.h> /* for aio_swake proto */ 41 #include <sys/kernel.h> 42 #include <sys/lock.h> 43 #include <sys/malloc.h> 44 #include <sys/mbuf.h> 45 #include <sys/mutex.h> 46 #include <sys/proc.h> 47 #include <sys/protosw.h> 48 #include <sys/resourcevar.h> 49 #include <sys/signalvar.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/sx.h> 53 #include <sys/sysctl.h> 54 55 /* 56 * Function pointer set by the AIO routines so that the socket buffer code 57 * can call back into the AIO module if it is loaded. 58 */ 59 void (*aio_swake)(struct socket *, struct sockbuf *); 60 61 /* 62 * Primitive routines for operating on socket buffers 63 */ 64 65 u_long sb_max = SB_MAX; 66 u_long sb_max_adj = 67 (quad_t)SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 68 69 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 70 71 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len); 72 static void sbflush_internal(struct sockbuf *sb); 73 74 /* 75 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY. 76 */ 77 static void 78 sbm_clrprotoflags(struct mbuf *m, int flags) 79 { 80 int mask; 81 82 mask = ~M_PROTOFLAGS; 83 if (flags & PRUS_NOTREADY) 84 mask |= M_NOTREADY; 85 while (m) { 86 m->m_flags &= mask; 87 m = m->m_next; 88 } 89 } 90 91 /* 92 * Compress M_NOTREADY mbufs after they have been readied by sbready(). 93 * 94 * sbcompress() skips M_NOTREADY mbufs since the data is not available to 95 * be copied at the time of sbcompress(). This function combines small 96 * mbufs similar to sbcompress() once mbufs are ready. 'm0' is the first 97 * mbuf sbready() marked ready, and 'end' is the first mbuf still not 98 * ready. 99 */ 100 static void 101 sbready_compress(struct sockbuf *sb, struct mbuf *m0, struct mbuf *end) 102 { 103 struct mbuf *m, *n; 104 int ext_size; 105 106 SOCKBUF_LOCK_ASSERT(sb); 107 108 if ((sb->sb_flags & SB_NOCOALESCE) != 0) 109 return; 110 111 for (m = m0; m != end; m = m->m_next) { 112 MPASS((m->m_flags & M_NOTREADY) == 0); 113 114 /* Compress small unmapped mbufs into plain mbufs. */ 115 if ((m->m_flags & M_NOMAP) && m->m_len <= MLEN) { 116 MPASS(m->m_flags & M_EXT); 117 ext_size = m->m_ext.ext_size; 118 if (mb_unmapped_compress(m) == 0) { 119 sb->sb_mbcnt -= ext_size; 120 sb->sb_ccnt -= 1; 121 } 122 } 123 124 /* 125 * NB: In sbcompress(), 'n' is the last mbuf in the 126 * socket buffer and 'm' is the new mbuf being copied 127 * into the trailing space of 'n'. Here, the roles 128 * are reversed and 'n' is the next mbuf after 'm' 129 * that is being copied into the trailing space of 130 * 'm'. 131 */ 132 n = m->m_next; 133 while ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 && 134 M_WRITABLE(m) && 135 (m->m_flags & M_NOMAP) == 0 && 136 n->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 137 n->m_len <= M_TRAILINGSPACE(m) && 138 m->m_type == n->m_type) { 139 KASSERT(sb->sb_lastrecord != n, 140 ("%s: merging start of record (%p) into previous mbuf (%p)", 141 __func__, n, m)); 142 m_copydata(n, 0, n->m_len, mtodo(m, m->m_len)); 143 m->m_len += n->m_len; 144 m->m_next = n->m_next; 145 m->m_flags |= n->m_flags & M_EOR; 146 if (sb->sb_mbtail == n) 147 sb->sb_mbtail = m; 148 149 sb->sb_mbcnt -= MSIZE; 150 sb->sb_mcnt -= 1; 151 if (n->m_flags & M_EXT) { 152 sb->sb_mbcnt -= n->m_ext.ext_size; 153 sb->sb_ccnt -= 1; 154 } 155 m_free(n); 156 n = m->m_next; 157 } 158 } 159 SBLASTRECORDCHK(sb); 160 SBLASTMBUFCHK(sb); 161 } 162 163 /* 164 * Mark ready "count" units of I/O starting with "m". Most mbufs 165 * count as a single unit of I/O except for EXT_PGS-backed mbufs which 166 * can be backed by multiple pages. 167 */ 168 int 169 sbready(struct sockbuf *sb, struct mbuf *m0, int count) 170 { 171 struct mbuf *m; 172 u_int blocker; 173 174 SOCKBUF_LOCK_ASSERT(sb); 175 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb)); 176 KASSERT(count > 0, ("%s: invalid count %d", __func__, count)); 177 178 m = m0; 179 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0; 180 181 while (count > 0) { 182 KASSERT(m->m_flags & M_NOTREADY, 183 ("%s: m %p !M_NOTREADY", __func__, m)); 184 if ((m->m_flags & M_EXT) != 0 && 185 m->m_ext.ext_type == EXT_PGS) { 186 if (count < m->m_ext.ext_pgs->nrdy) { 187 m->m_ext.ext_pgs->nrdy -= count; 188 count = 0; 189 break; 190 } 191 count -= m->m_ext.ext_pgs->nrdy; 192 m->m_ext.ext_pgs->nrdy = 0; 193 } else 194 count--; 195 196 m->m_flags &= ~(M_NOTREADY | blocker); 197 if (blocker) 198 sb->sb_acc += m->m_len; 199 m = m->m_next; 200 } 201 202 /* 203 * If the first mbuf is still not fully ready because only 204 * some of its backing pages were readied, no further progress 205 * can be made. 206 */ 207 if (m0 == m) { 208 MPASS(m->m_flags & M_NOTREADY); 209 return (EINPROGRESS); 210 } 211 212 if (!blocker) { 213 sbready_compress(sb, m0, m); 214 return (EINPROGRESS); 215 } 216 217 /* This one was blocking all the queue. */ 218 for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) { 219 KASSERT(m->m_flags & M_BLOCKED, 220 ("%s: m %p !M_BLOCKED", __func__, m)); 221 m->m_flags &= ~M_BLOCKED; 222 sb->sb_acc += m->m_len; 223 } 224 225 sb->sb_fnrdy = m; 226 sbready_compress(sb, m0, m); 227 228 return (0); 229 } 230 231 /* 232 * Adjust sockbuf state reflecting allocation of m. 233 */ 234 void 235 sballoc(struct sockbuf *sb, struct mbuf *m) 236 { 237 238 SOCKBUF_LOCK_ASSERT(sb); 239 240 sb->sb_ccc += m->m_len; 241 242 if (sb->sb_fnrdy == NULL) { 243 if (m->m_flags & M_NOTREADY) 244 sb->sb_fnrdy = m; 245 else 246 sb->sb_acc += m->m_len; 247 } else 248 m->m_flags |= M_BLOCKED; 249 250 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 251 sb->sb_ctl += m->m_len; 252 253 sb->sb_mbcnt += MSIZE; 254 sb->sb_mcnt += 1; 255 256 if (m->m_flags & M_EXT) { 257 sb->sb_mbcnt += m->m_ext.ext_size; 258 sb->sb_ccnt += 1; 259 } 260 } 261 262 /* 263 * Adjust sockbuf state reflecting freeing of m. 264 */ 265 void 266 sbfree(struct sockbuf *sb, struct mbuf *m) 267 { 268 269 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ 270 SOCKBUF_LOCK_ASSERT(sb); 271 #endif 272 273 sb->sb_ccc -= m->m_len; 274 275 if (!(m->m_flags & M_NOTAVAIL)) 276 sb->sb_acc -= m->m_len; 277 278 if (m == sb->sb_fnrdy) { 279 struct mbuf *n; 280 281 KASSERT(m->m_flags & M_NOTREADY, 282 ("%s: m %p !M_NOTREADY", __func__, m)); 283 284 n = m->m_next; 285 while (n != NULL && !(n->m_flags & M_NOTREADY)) { 286 n->m_flags &= ~M_BLOCKED; 287 sb->sb_acc += n->m_len; 288 n = n->m_next; 289 } 290 sb->sb_fnrdy = n; 291 } 292 293 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 294 sb->sb_ctl -= m->m_len; 295 296 sb->sb_mbcnt -= MSIZE; 297 sb->sb_mcnt -= 1; 298 if (m->m_flags & M_EXT) { 299 sb->sb_mbcnt -= m->m_ext.ext_size; 300 sb->sb_ccnt -= 1; 301 } 302 303 if (sb->sb_sndptr == m) { 304 sb->sb_sndptr = NULL; 305 sb->sb_sndptroff = 0; 306 } 307 if (sb->sb_sndptroff != 0) 308 sb->sb_sndptroff -= m->m_len; 309 } 310 311 /* 312 * Socantsendmore indicates that no more data will be sent on the socket; it 313 * would normally be applied to a socket when the user informs the system 314 * that no more data is to be sent, by the protocol code (in case 315 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be 316 * received, and will normally be applied to the socket by a protocol when it 317 * detects that the peer will send no more data. Data queued for reading in 318 * the socket may yet be read. 319 */ 320 void 321 socantsendmore_locked(struct socket *so) 322 { 323 324 SOCKBUF_LOCK_ASSERT(&so->so_snd); 325 326 so->so_snd.sb_state |= SBS_CANTSENDMORE; 327 sowwakeup_locked(so); 328 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); 329 } 330 331 void 332 socantsendmore(struct socket *so) 333 { 334 335 SOCKBUF_LOCK(&so->so_snd); 336 socantsendmore_locked(so); 337 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); 338 } 339 340 void 341 socantrcvmore_locked(struct socket *so) 342 { 343 344 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 345 346 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 347 sorwakeup_locked(so); 348 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); 349 } 350 351 void 352 socantrcvmore(struct socket *so) 353 { 354 355 SOCKBUF_LOCK(&so->so_rcv); 356 socantrcvmore_locked(so); 357 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); 358 } 359 360 /* 361 * Wait for data to arrive at/drain from a socket buffer. 362 */ 363 int 364 sbwait(struct sockbuf *sb) 365 { 366 367 SOCKBUF_LOCK_ASSERT(sb); 368 369 sb->sb_flags |= SB_WAIT; 370 return (msleep_sbt(&sb->sb_acc, &sb->sb_mtx, 371 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", 372 sb->sb_timeo, 0, 0)); 373 } 374 375 int 376 sblock(struct sockbuf *sb, int flags) 377 { 378 379 KASSERT((flags & SBL_VALID) == flags, 380 ("sblock: flags invalid (0x%x)", flags)); 381 382 if (flags & SBL_WAIT) { 383 if ((sb->sb_flags & SB_NOINTR) || 384 (flags & SBL_NOINTR)) { 385 sx_xlock(&sb->sb_sx); 386 return (0); 387 } 388 return (sx_xlock_sig(&sb->sb_sx)); 389 } else { 390 if (sx_try_xlock(&sb->sb_sx) == 0) 391 return (EWOULDBLOCK); 392 return (0); 393 } 394 } 395 396 void 397 sbunlock(struct sockbuf *sb) 398 { 399 400 sx_xunlock(&sb->sb_sx); 401 } 402 403 /* 404 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 405 * via SIGIO if the socket has the SS_ASYNC flag set. 406 * 407 * Called with the socket buffer lock held; will release the lock by the end 408 * of the function. This allows the caller to acquire the socket buffer lock 409 * while testing for the need for various sorts of wakeup and hold it through 410 * to the point where it's no longer required. We currently hold the lock 411 * through calls out to other subsystems (with the exception of kqueue), and 412 * then release it to avoid lock order issues. It's not clear that's 413 * correct. 414 */ 415 void 416 sowakeup(struct socket *so, struct sockbuf *sb) 417 { 418 int ret; 419 420 SOCKBUF_LOCK_ASSERT(sb); 421 422 selwakeuppri(sb->sb_sel, PSOCK); 423 if (!SEL_WAITING(sb->sb_sel)) 424 sb->sb_flags &= ~SB_SEL; 425 if (sb->sb_flags & SB_WAIT) { 426 sb->sb_flags &= ~SB_WAIT; 427 wakeup(&sb->sb_acc); 428 } 429 KNOTE_LOCKED(&sb->sb_sel->si_note, 0); 430 if (sb->sb_upcall != NULL) { 431 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT); 432 if (ret == SU_ISCONNECTED) { 433 KASSERT(sb == &so->so_rcv, 434 ("SO_SND upcall returned SU_ISCONNECTED")); 435 soupcall_clear(so, SO_RCV); 436 } 437 } else 438 ret = SU_OK; 439 if (sb->sb_flags & SB_AIO) 440 sowakeup_aio(so, sb); 441 SOCKBUF_UNLOCK(sb); 442 if (ret == SU_ISCONNECTED) 443 soisconnected(so); 444 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 445 pgsigio(&so->so_sigio, SIGIO, 0); 446 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED); 447 } 448 449 /* 450 * Socket buffer (struct sockbuf) utility routines. 451 * 452 * Each socket contains two socket buffers: one for sending data and one for 453 * receiving data. Each buffer contains a queue of mbufs, information about 454 * the number of mbufs and amount of data in the queue, and other fields 455 * allowing select() statements and notification on data availability to be 456 * implemented. 457 * 458 * Data stored in a socket buffer is maintained as a list of records. Each 459 * record is a list of mbufs chained together with the m_next field. Records 460 * are chained together with the m_nextpkt field. The upper level routine 461 * soreceive() expects the following conventions to be observed when placing 462 * information in the receive buffer: 463 * 464 * 1. If the protocol requires each message be preceded by the sender's name, 465 * then a record containing that name must be present before any 466 * associated data (mbuf's must be of type MT_SONAME). 467 * 2. If the protocol supports the exchange of ``access rights'' (really just 468 * additional data associated with the message), and there are ``rights'' 469 * to be received, then a record containing this data should be present 470 * (mbuf's must be of type MT_RIGHTS). 471 * 3. If a name or rights record exists, then it must be followed by a data 472 * record, perhaps of zero length. 473 * 474 * Before using a new socket structure it is first necessary to reserve 475 * buffer space to the socket, by calling sbreserve(). This should commit 476 * some of the available buffer space in the system buffer pool for the 477 * socket (currently, it does nothing but enforce limits). The space should 478 * be released by calling sbrelease() when the socket is destroyed. 479 */ 480 int 481 soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 482 { 483 struct thread *td = curthread; 484 485 SOCKBUF_LOCK(&so->so_snd); 486 SOCKBUF_LOCK(&so->so_rcv); 487 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0) 488 goto bad; 489 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0) 490 goto bad2; 491 if (so->so_rcv.sb_lowat == 0) 492 so->so_rcv.sb_lowat = 1; 493 if (so->so_snd.sb_lowat == 0) 494 so->so_snd.sb_lowat = MCLBYTES; 495 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 496 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 497 SOCKBUF_UNLOCK(&so->so_rcv); 498 SOCKBUF_UNLOCK(&so->so_snd); 499 return (0); 500 bad2: 501 sbrelease_locked(&so->so_snd, so); 502 bad: 503 SOCKBUF_UNLOCK(&so->so_rcv); 504 SOCKBUF_UNLOCK(&so->so_snd); 505 return (ENOBUFS); 506 } 507 508 static int 509 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 510 { 511 int error = 0; 512 u_long tmp_sb_max = sb_max; 513 514 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); 515 if (error || !req->newptr) 516 return (error); 517 if (tmp_sb_max < MSIZE + MCLBYTES) 518 return (EINVAL); 519 sb_max = tmp_sb_max; 520 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 521 return (0); 522 } 523 524 /* 525 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't 526 * become limiting if buffering efficiency is near the normal case. 527 */ 528 int 529 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so, 530 struct thread *td) 531 { 532 rlim_t sbsize_limit; 533 534 SOCKBUF_LOCK_ASSERT(sb); 535 536 /* 537 * When a thread is passed, we take into account the thread's socket 538 * buffer size limit. The caller will generally pass curthread, but 539 * in the TCP input path, NULL will be passed to indicate that no 540 * appropriate thread resource limits are available. In that case, 541 * we don't apply a process limit. 542 */ 543 if (cc > sb_max_adj) 544 return (0); 545 if (td != NULL) { 546 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE); 547 } else 548 sbsize_limit = RLIM_INFINITY; 549 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 550 sbsize_limit)) 551 return (0); 552 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 553 if (sb->sb_lowat > sb->sb_hiwat) 554 sb->sb_lowat = sb->sb_hiwat; 555 return (1); 556 } 557 558 int 559 sbsetopt(struct socket *so, int cmd, u_long cc) 560 { 561 struct sockbuf *sb; 562 short *flags; 563 u_int *hiwat, *lowat; 564 int error; 565 566 sb = NULL; 567 SOCK_LOCK(so); 568 if (SOLISTENING(so)) { 569 switch (cmd) { 570 case SO_SNDLOWAT: 571 case SO_SNDBUF: 572 lowat = &so->sol_sbsnd_lowat; 573 hiwat = &so->sol_sbsnd_hiwat; 574 flags = &so->sol_sbsnd_flags; 575 break; 576 case SO_RCVLOWAT: 577 case SO_RCVBUF: 578 lowat = &so->sol_sbrcv_lowat; 579 hiwat = &so->sol_sbrcv_hiwat; 580 flags = &so->sol_sbrcv_flags; 581 break; 582 } 583 } else { 584 switch (cmd) { 585 case SO_SNDLOWAT: 586 case SO_SNDBUF: 587 sb = &so->so_snd; 588 break; 589 case SO_RCVLOWAT: 590 case SO_RCVBUF: 591 sb = &so->so_rcv; 592 break; 593 } 594 flags = &sb->sb_flags; 595 hiwat = &sb->sb_hiwat; 596 lowat = &sb->sb_lowat; 597 SOCKBUF_LOCK(sb); 598 } 599 600 error = 0; 601 switch (cmd) { 602 case SO_SNDBUF: 603 case SO_RCVBUF: 604 if (SOLISTENING(so)) { 605 if (cc > sb_max_adj) { 606 error = ENOBUFS; 607 break; 608 } 609 *hiwat = cc; 610 if (*lowat > *hiwat) 611 *lowat = *hiwat; 612 } else { 613 if (!sbreserve_locked(sb, cc, so, curthread)) 614 error = ENOBUFS; 615 } 616 if (error == 0) 617 *flags &= ~SB_AUTOSIZE; 618 break; 619 case SO_SNDLOWAT: 620 case SO_RCVLOWAT: 621 /* 622 * Make sure the low-water is never greater than the 623 * high-water. 624 */ 625 *lowat = (cc > *hiwat) ? *hiwat : cc; 626 break; 627 } 628 629 if (!SOLISTENING(so)) 630 SOCKBUF_UNLOCK(sb); 631 SOCK_UNLOCK(so); 632 return (error); 633 } 634 635 /* 636 * Free mbufs held by a socket, and reserved mbuf space. 637 */ 638 void 639 sbrelease_internal(struct sockbuf *sb, struct socket *so) 640 { 641 642 sbflush_internal(sb); 643 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 644 RLIM_INFINITY); 645 sb->sb_mbmax = 0; 646 } 647 648 void 649 sbrelease_locked(struct sockbuf *sb, struct socket *so) 650 { 651 652 SOCKBUF_LOCK_ASSERT(sb); 653 654 sbrelease_internal(sb, so); 655 } 656 657 void 658 sbrelease(struct sockbuf *sb, struct socket *so) 659 { 660 661 SOCKBUF_LOCK(sb); 662 sbrelease_locked(sb, so); 663 SOCKBUF_UNLOCK(sb); 664 } 665 666 void 667 sbdestroy(struct sockbuf *sb, struct socket *so) 668 { 669 670 sbrelease_internal(sb, so); 671 } 672 673 /* 674 * Routines to add and remove data from an mbuf queue. 675 * 676 * The routines sbappend() or sbappendrecord() are normally called to append 677 * new mbufs to a socket buffer, after checking that adequate space is 678 * available, comparing the function sbspace() with the amount of data to be 679 * added. sbappendrecord() differs from sbappend() in that data supplied is 680 * treated as the beginning of a new record. To place a sender's address, 681 * optional access rights, and data in a socket receive buffer, 682 * sbappendaddr() should be used. To place access rights and data in a 683 * socket receive buffer, sbappendrights() should be used. In either case, 684 * the new data begins a new record. Note that unlike sbappend() and 685 * sbappendrecord(), these routines check for the caller that there will be 686 * enough space to store the data. Each fails if there is not enough space, 687 * or if it cannot find mbufs to store additional information in. 688 * 689 * Reliable protocols may use the socket send buffer to hold data awaiting 690 * acknowledgement. Data is normally copied from a socket send buffer in a 691 * protocol with m_copy for output to a peer, and then removing the data from 692 * the socket buffer with sbdrop() or sbdroprecord() when the data is 693 * acknowledged by the peer. 694 */ 695 #ifdef SOCKBUF_DEBUG 696 void 697 sblastrecordchk(struct sockbuf *sb, const char *file, int line) 698 { 699 struct mbuf *m = sb->sb_mb; 700 701 SOCKBUF_LOCK_ASSERT(sb); 702 703 while (m && m->m_nextpkt) 704 m = m->m_nextpkt; 705 706 if (m != sb->sb_lastrecord) { 707 printf("%s: sb_mb %p sb_lastrecord %p last %p\n", 708 __func__, sb->sb_mb, sb->sb_lastrecord, m); 709 printf("packet chain:\n"); 710 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 711 printf("\t%p\n", m); 712 panic("%s from %s:%u", __func__, file, line); 713 } 714 } 715 716 void 717 sblastmbufchk(struct sockbuf *sb, const char *file, int line) 718 { 719 struct mbuf *m = sb->sb_mb; 720 struct mbuf *n; 721 722 SOCKBUF_LOCK_ASSERT(sb); 723 724 while (m && m->m_nextpkt) 725 m = m->m_nextpkt; 726 727 while (m && m->m_next) 728 m = m->m_next; 729 730 if (m != sb->sb_mbtail) { 731 printf("%s: sb_mb %p sb_mbtail %p last %p\n", 732 __func__, sb->sb_mb, sb->sb_mbtail, m); 733 printf("packet tree:\n"); 734 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 735 printf("\t"); 736 for (n = m; n != NULL; n = n->m_next) 737 printf("%p ", n); 738 printf("\n"); 739 } 740 panic("%s from %s:%u", __func__, file, line); 741 } 742 } 743 #endif /* SOCKBUF_DEBUG */ 744 745 #define SBLINKRECORD(sb, m0) do { \ 746 SOCKBUF_LOCK_ASSERT(sb); \ 747 if ((sb)->sb_lastrecord != NULL) \ 748 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 749 else \ 750 (sb)->sb_mb = (m0); \ 751 (sb)->sb_lastrecord = (m0); \ 752 } while (/*CONSTCOND*/0) 753 754 /* 755 * Append mbuf chain m to the last record in the socket buffer sb. The 756 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 757 * are discarded and mbufs are compacted where possible. 758 */ 759 void 760 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags) 761 { 762 struct mbuf *n; 763 764 SOCKBUF_LOCK_ASSERT(sb); 765 766 if (m == NULL) 767 return; 768 sbm_clrprotoflags(m, flags); 769 SBLASTRECORDCHK(sb); 770 n = sb->sb_mb; 771 if (n) { 772 while (n->m_nextpkt) 773 n = n->m_nextpkt; 774 do { 775 if (n->m_flags & M_EOR) { 776 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 777 return; 778 } 779 } while (n->m_next && (n = n->m_next)); 780 } else { 781 /* 782 * XXX Would like to simply use sb_mbtail here, but 783 * XXX I need to verify that I won't miss an EOR that 784 * XXX way. 785 */ 786 if ((n = sb->sb_lastrecord) != NULL) { 787 do { 788 if (n->m_flags & M_EOR) { 789 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 790 return; 791 } 792 } while (n->m_next && (n = n->m_next)); 793 } else { 794 /* 795 * If this is the first record in the socket buffer, 796 * it's also the last record. 797 */ 798 sb->sb_lastrecord = m; 799 } 800 } 801 sbcompress(sb, m, n); 802 SBLASTRECORDCHK(sb); 803 } 804 805 /* 806 * Append mbuf chain m to the last record in the socket buffer sb. The 807 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 808 * are discarded and mbufs are compacted where possible. 809 */ 810 void 811 sbappend(struct sockbuf *sb, struct mbuf *m, int flags) 812 { 813 814 SOCKBUF_LOCK(sb); 815 sbappend_locked(sb, m, flags); 816 SOCKBUF_UNLOCK(sb); 817 } 818 819 /* 820 * This version of sbappend() should only be used when the caller absolutely 821 * knows that there will never be more than one record in the socket buffer, 822 * that is, a stream protocol (such as TCP). 823 */ 824 void 825 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags) 826 { 827 SOCKBUF_LOCK_ASSERT(sb); 828 829 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); 830 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); 831 832 SBLASTMBUFCHK(sb); 833 834 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 835 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0); 836 837 sbcompress(sb, m, sb->sb_mbtail); 838 839 sb->sb_lastrecord = sb->sb_mb; 840 SBLASTRECORDCHK(sb); 841 } 842 843 /* 844 * This version of sbappend() should only be used when the caller absolutely 845 * knows that there will never be more than one record in the socket buffer, 846 * that is, a stream protocol (such as TCP). 847 */ 848 void 849 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags) 850 { 851 852 SOCKBUF_LOCK(sb); 853 sbappendstream_locked(sb, m, flags); 854 SOCKBUF_UNLOCK(sb); 855 } 856 857 #ifdef SOCKBUF_DEBUG 858 void 859 sbcheck(struct sockbuf *sb, const char *file, int line) 860 { 861 struct mbuf *m, *n, *fnrdy; 862 u_long acc, ccc, mbcnt; 863 864 SOCKBUF_LOCK_ASSERT(sb); 865 866 acc = ccc = mbcnt = 0; 867 fnrdy = NULL; 868 869 for (m = sb->sb_mb; m; m = n) { 870 n = m->m_nextpkt; 871 for (; m; m = m->m_next) { 872 if (m->m_len == 0) { 873 printf("sb %p empty mbuf %p\n", sb, m); 874 goto fail; 875 } 876 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) { 877 if (m != sb->sb_fnrdy) { 878 printf("sb %p: fnrdy %p != m %p\n", 879 sb, sb->sb_fnrdy, m); 880 goto fail; 881 } 882 fnrdy = m; 883 } 884 if (fnrdy) { 885 if (!(m->m_flags & M_NOTAVAIL)) { 886 printf("sb %p: fnrdy %p, m %p is avail\n", 887 sb, sb->sb_fnrdy, m); 888 goto fail; 889 } 890 } else 891 acc += m->m_len; 892 ccc += m->m_len; 893 mbcnt += MSIZE; 894 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 895 mbcnt += m->m_ext.ext_size; 896 } 897 } 898 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) { 899 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n", 900 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt); 901 goto fail; 902 } 903 return; 904 fail: 905 panic("%s from %s:%u", __func__, file, line); 906 } 907 #endif 908 909 /* 910 * As above, except the mbuf chain begins a new record. 911 */ 912 void 913 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) 914 { 915 struct mbuf *m; 916 917 SOCKBUF_LOCK_ASSERT(sb); 918 919 if (m0 == NULL) 920 return; 921 m_clrprotoflags(m0); 922 /* 923 * Put the first mbuf on the queue. Note this permits zero length 924 * records. 925 */ 926 sballoc(sb, m0); 927 SBLASTRECORDCHK(sb); 928 SBLINKRECORD(sb, m0); 929 sb->sb_mbtail = m0; 930 m = m0->m_next; 931 m0->m_next = 0; 932 if (m && (m0->m_flags & M_EOR)) { 933 m0->m_flags &= ~M_EOR; 934 m->m_flags |= M_EOR; 935 } 936 /* always call sbcompress() so it can do SBLASTMBUFCHK() */ 937 sbcompress(sb, m, m0); 938 } 939 940 /* 941 * As above, except the mbuf chain begins a new record. 942 */ 943 void 944 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 945 { 946 947 SOCKBUF_LOCK(sb); 948 sbappendrecord_locked(sb, m0); 949 SOCKBUF_UNLOCK(sb); 950 } 951 952 /* Helper routine that appends data, control, and address to a sockbuf. */ 953 static int 954 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa, 955 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last) 956 { 957 struct mbuf *m, *n, *nlast; 958 #if MSIZE <= 256 959 if (asa->sa_len > MLEN) 960 return (0); 961 #endif 962 m = m_get(M_NOWAIT, MT_SONAME); 963 if (m == NULL) 964 return (0); 965 m->m_len = asa->sa_len; 966 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 967 if (m0) { 968 m_clrprotoflags(m0); 969 m_tag_delete_chain(m0, NULL); 970 /* 971 * Clear some persistent info from pkthdr. 972 * We don't use m_demote(), because some netgraph consumers 973 * expect M_PKTHDR presence. 974 */ 975 m0->m_pkthdr.rcvif = NULL; 976 m0->m_pkthdr.flowid = 0; 977 m0->m_pkthdr.csum_flags = 0; 978 m0->m_pkthdr.fibnum = 0; 979 m0->m_pkthdr.rsstype = 0; 980 } 981 if (ctrl_last) 982 ctrl_last->m_next = m0; /* concatenate data to control */ 983 else 984 control = m0; 985 m->m_next = control; 986 for (n = m; n->m_next != NULL; n = n->m_next) 987 sballoc(sb, n); 988 sballoc(sb, n); 989 nlast = n; 990 SBLINKRECORD(sb, m); 991 992 sb->sb_mbtail = nlast; 993 SBLASTMBUFCHK(sb); 994 995 SBLASTRECORDCHK(sb); 996 return (1); 997 } 998 999 /* 1000 * Append address and data, and optionally, control (ancillary) data to the 1001 * receive queue of a socket. If present, m0 must include a packet header 1002 * with total length. Returns 0 if no space in sockbuf or insufficient 1003 * mbufs. 1004 */ 1005 int 1006 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, 1007 struct mbuf *m0, struct mbuf *control) 1008 { 1009 struct mbuf *ctrl_last; 1010 int space = asa->sa_len; 1011 1012 SOCKBUF_LOCK_ASSERT(sb); 1013 1014 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1015 panic("sbappendaddr_locked"); 1016 if (m0) 1017 space += m0->m_pkthdr.len; 1018 space += m_length(control, &ctrl_last); 1019 1020 if (space > sbspace(sb)) 1021 return (0); 1022 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1023 } 1024 1025 /* 1026 * Append address and data, and optionally, control (ancillary) data to the 1027 * receive queue of a socket. If present, m0 must include a packet header 1028 * with total length. Returns 0 if insufficient mbufs. Does not validate space 1029 * on the receiving sockbuf. 1030 */ 1031 int 1032 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa, 1033 struct mbuf *m0, struct mbuf *control) 1034 { 1035 struct mbuf *ctrl_last; 1036 1037 SOCKBUF_LOCK_ASSERT(sb); 1038 1039 ctrl_last = (control == NULL) ? NULL : m_last(control); 1040 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1041 } 1042 1043 /* 1044 * Append address and data, and optionally, control (ancillary) data to the 1045 * receive queue of a socket. If present, m0 must include a packet header 1046 * with total length. Returns 0 if no space in sockbuf or insufficient 1047 * mbufs. 1048 */ 1049 int 1050 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, 1051 struct mbuf *m0, struct mbuf *control) 1052 { 1053 int retval; 1054 1055 SOCKBUF_LOCK(sb); 1056 retval = sbappendaddr_locked(sb, asa, m0, control); 1057 SOCKBUF_UNLOCK(sb); 1058 return (retval); 1059 } 1060 1061 void 1062 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, 1063 struct mbuf *control) 1064 { 1065 struct mbuf *m, *mlast; 1066 1067 m_clrprotoflags(m0); 1068 m_last(control)->m_next = m0; 1069 1070 SBLASTRECORDCHK(sb); 1071 1072 for (m = control; m->m_next; m = m->m_next) 1073 sballoc(sb, m); 1074 sballoc(sb, m); 1075 mlast = m; 1076 SBLINKRECORD(sb, control); 1077 1078 sb->sb_mbtail = mlast; 1079 SBLASTMBUFCHK(sb); 1080 1081 SBLASTRECORDCHK(sb); 1082 } 1083 1084 void 1085 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 1086 { 1087 1088 SOCKBUF_LOCK(sb); 1089 sbappendcontrol_locked(sb, m0, control); 1090 SOCKBUF_UNLOCK(sb); 1091 } 1092 1093 /* 1094 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf 1095 * (n). If (n) is NULL, the buffer is presumed empty. 1096 * 1097 * When the data is compressed, mbufs in the chain may be handled in one of 1098 * three ways: 1099 * 1100 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no 1101 * record boundary, and no change in data type). 1102 * 1103 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into 1104 * an mbuf already in the socket buffer. This can occur if an 1105 * appropriate mbuf exists, there is room, both mbufs are not marked as 1106 * not ready, and no merging of data types will occur. 1107 * 1108 * (3) The mbuf may be appended to the end of the existing mbuf chain. 1109 * 1110 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as 1111 * end-of-record. 1112 */ 1113 void 1114 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1115 { 1116 int eor = 0; 1117 struct mbuf *o; 1118 1119 SOCKBUF_LOCK_ASSERT(sb); 1120 1121 while (m) { 1122 eor |= m->m_flags & M_EOR; 1123 if (m->m_len == 0 && 1124 (eor == 0 || 1125 (((o = m->m_next) || (o = n)) && 1126 o->m_type == m->m_type))) { 1127 if (sb->sb_lastrecord == m) 1128 sb->sb_lastrecord = m->m_next; 1129 m = m_free(m); 1130 continue; 1131 } 1132 if (n && (n->m_flags & M_EOR) == 0 && 1133 M_WRITABLE(n) && 1134 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1135 !(m->m_flags & M_NOTREADY) && 1136 !(n->m_flags & (M_NOTREADY | M_NOMAP)) && 1137 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1138 m->m_len <= M_TRAILINGSPACE(n) && 1139 n->m_type == m->m_type) { 1140 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1141 n->m_len += m->m_len; 1142 sb->sb_ccc += m->m_len; 1143 if (sb->sb_fnrdy == NULL) 1144 sb->sb_acc += m->m_len; 1145 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1146 /* XXX: Probably don't need.*/ 1147 sb->sb_ctl += m->m_len; 1148 m = m_free(m); 1149 continue; 1150 } 1151 if (m->m_len <= MLEN && (m->m_flags & M_NOMAP) && 1152 (m->m_flags & M_NOTREADY) == 0) 1153 (void)mb_unmapped_compress(m); 1154 if (n) 1155 n->m_next = m; 1156 else 1157 sb->sb_mb = m; 1158 sb->sb_mbtail = m; 1159 sballoc(sb, m); 1160 n = m; 1161 m->m_flags &= ~M_EOR; 1162 m = m->m_next; 1163 n->m_next = 0; 1164 } 1165 if (eor) { 1166 KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); 1167 n->m_flags |= eor; 1168 } 1169 SBLASTMBUFCHK(sb); 1170 } 1171 1172 /* 1173 * Free all mbufs in a sockbuf. Check that all resources are reclaimed. 1174 */ 1175 static void 1176 sbflush_internal(struct sockbuf *sb) 1177 { 1178 1179 while (sb->sb_mbcnt) { 1180 /* 1181 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty: 1182 * we would loop forever. Panic instead. 1183 */ 1184 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 1185 break; 1186 m_freem(sbcut_internal(sb, (int)sb->sb_ccc)); 1187 } 1188 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 1189 ("%s: ccc %u mb %p mbcnt %u", __func__, 1190 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 1191 } 1192 1193 void 1194 sbflush_locked(struct sockbuf *sb) 1195 { 1196 1197 SOCKBUF_LOCK_ASSERT(sb); 1198 sbflush_internal(sb); 1199 } 1200 1201 void 1202 sbflush(struct sockbuf *sb) 1203 { 1204 1205 SOCKBUF_LOCK(sb); 1206 sbflush_locked(sb); 1207 SOCKBUF_UNLOCK(sb); 1208 } 1209 1210 /* 1211 * Cut data from (the front of) a sockbuf. 1212 */ 1213 static struct mbuf * 1214 sbcut_internal(struct sockbuf *sb, int len) 1215 { 1216 struct mbuf *m, *next, *mfree; 1217 1218 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0", 1219 __func__, len)); 1220 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u", 1221 __func__, len, sb->sb_ccc)); 1222 1223 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1224 mfree = NULL; 1225 1226 while (len > 0) { 1227 if (m == NULL) { 1228 KASSERT(next, ("%s: no next, len %d", __func__, len)); 1229 m = next; 1230 next = m->m_nextpkt; 1231 } 1232 if (m->m_len > len) { 1233 KASSERT(!(m->m_flags & M_NOTAVAIL), 1234 ("%s: m %p M_NOTAVAIL", __func__, m)); 1235 m->m_len -= len; 1236 m->m_data += len; 1237 sb->sb_ccc -= len; 1238 sb->sb_acc -= len; 1239 if (sb->sb_sndptroff != 0) 1240 sb->sb_sndptroff -= len; 1241 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1242 sb->sb_ctl -= len; 1243 break; 1244 } 1245 len -= m->m_len; 1246 sbfree(sb, m); 1247 /* 1248 * Do not put M_NOTREADY buffers to the free list, they 1249 * are referenced from outside. 1250 */ 1251 if (m->m_flags & M_NOTREADY) 1252 m = m->m_next; 1253 else { 1254 struct mbuf *n; 1255 1256 n = m->m_next; 1257 m->m_next = mfree; 1258 mfree = m; 1259 m = n; 1260 } 1261 } 1262 /* 1263 * Free any zero-length mbufs from the buffer. 1264 * For SOCK_DGRAM sockets such mbufs represent empty records. 1265 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer, 1266 * when sosend_generic() needs to send only control data. 1267 */ 1268 while (m && m->m_len == 0) { 1269 struct mbuf *n; 1270 1271 sbfree(sb, m); 1272 n = m->m_next; 1273 m->m_next = mfree; 1274 mfree = m; 1275 m = n; 1276 } 1277 if (m) { 1278 sb->sb_mb = m; 1279 m->m_nextpkt = next; 1280 } else 1281 sb->sb_mb = next; 1282 /* 1283 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure 1284 * sb_lastrecord is up-to-date if we dropped part of the last record. 1285 */ 1286 m = sb->sb_mb; 1287 if (m == NULL) { 1288 sb->sb_mbtail = NULL; 1289 sb->sb_lastrecord = NULL; 1290 } else if (m->m_nextpkt == NULL) { 1291 sb->sb_lastrecord = m; 1292 } 1293 1294 return (mfree); 1295 } 1296 1297 /* 1298 * Drop data from (the front of) a sockbuf. 1299 */ 1300 void 1301 sbdrop_locked(struct sockbuf *sb, int len) 1302 { 1303 1304 SOCKBUF_LOCK_ASSERT(sb); 1305 m_freem(sbcut_internal(sb, len)); 1306 } 1307 1308 /* 1309 * Drop data from (the front of) a sockbuf, 1310 * and return it to caller. 1311 */ 1312 struct mbuf * 1313 sbcut_locked(struct sockbuf *sb, int len) 1314 { 1315 1316 SOCKBUF_LOCK_ASSERT(sb); 1317 return (sbcut_internal(sb, len)); 1318 } 1319 1320 void 1321 sbdrop(struct sockbuf *sb, int len) 1322 { 1323 struct mbuf *mfree; 1324 1325 SOCKBUF_LOCK(sb); 1326 mfree = sbcut_internal(sb, len); 1327 SOCKBUF_UNLOCK(sb); 1328 1329 m_freem(mfree); 1330 } 1331 1332 struct mbuf * 1333 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff) 1334 { 1335 struct mbuf *m; 1336 1337 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1338 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1339 *moff = off; 1340 if (sb->sb_sndptr == NULL) { 1341 sb->sb_sndptr = sb->sb_mb; 1342 sb->sb_sndptroff = 0; 1343 } 1344 return (sb->sb_mb); 1345 } else { 1346 m = sb->sb_sndptr; 1347 off -= sb->sb_sndptroff; 1348 } 1349 *moff = off; 1350 return (m); 1351 } 1352 1353 void 1354 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len) 1355 { 1356 /* 1357 * A small copy was done, advance forward the sb_sbsndptr to cover 1358 * it. 1359 */ 1360 struct mbuf *m; 1361 1362 if (mb != sb->sb_sndptr) { 1363 /* Did not copyout at the same mbuf */ 1364 return; 1365 } 1366 m = mb; 1367 while (m && (len > 0)) { 1368 if (len >= m->m_len) { 1369 len -= m->m_len; 1370 if (m->m_next) { 1371 sb->sb_sndptroff += m->m_len; 1372 sb->sb_sndptr = m->m_next; 1373 } 1374 m = m->m_next; 1375 } else { 1376 len = 0; 1377 } 1378 } 1379 } 1380 1381 /* 1382 * Return the first mbuf and the mbuf data offset for the provided 1383 * send offset without changing the "sb_sndptroff" field. 1384 */ 1385 struct mbuf * 1386 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff) 1387 { 1388 struct mbuf *m; 1389 1390 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1391 1392 /* 1393 * If the "off" is below the stored offset, which happens on 1394 * retransmits, just use "sb_mb": 1395 */ 1396 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1397 m = sb->sb_mb; 1398 } else { 1399 m = sb->sb_sndptr; 1400 off -= sb->sb_sndptroff; 1401 } 1402 while (off > 0 && m != NULL) { 1403 if (off < m->m_len) 1404 break; 1405 off -= m->m_len; 1406 m = m->m_next; 1407 } 1408 *moff = off; 1409 return (m); 1410 } 1411 1412 /* 1413 * Drop a record off the front of a sockbuf and move the next record to the 1414 * front. 1415 */ 1416 void 1417 sbdroprecord_locked(struct sockbuf *sb) 1418 { 1419 struct mbuf *m; 1420 1421 SOCKBUF_LOCK_ASSERT(sb); 1422 1423 m = sb->sb_mb; 1424 if (m) { 1425 sb->sb_mb = m->m_nextpkt; 1426 do { 1427 sbfree(sb, m); 1428 m = m_free(m); 1429 } while (m); 1430 } 1431 SB_EMPTY_FIXUP(sb); 1432 } 1433 1434 /* 1435 * Drop a record off the front of a sockbuf and move the next record to the 1436 * front. 1437 */ 1438 void 1439 sbdroprecord(struct sockbuf *sb) 1440 { 1441 1442 SOCKBUF_LOCK(sb); 1443 sbdroprecord_locked(sb); 1444 SOCKBUF_UNLOCK(sb); 1445 } 1446 1447 /* 1448 * Create a "control" mbuf containing the specified data with the specified 1449 * type for presentation on a socket buffer. 1450 */ 1451 struct mbuf * 1452 sbcreatecontrol(caddr_t p, int size, int type, int level) 1453 { 1454 struct cmsghdr *cp; 1455 struct mbuf *m; 1456 1457 if (CMSG_SPACE((u_int)size) > MCLBYTES) 1458 return ((struct mbuf *) NULL); 1459 if (CMSG_SPACE((u_int)size) > MLEN) 1460 m = m_getcl(M_NOWAIT, MT_CONTROL, 0); 1461 else 1462 m = m_get(M_NOWAIT, MT_CONTROL); 1463 if (m == NULL) 1464 return ((struct mbuf *) NULL); 1465 cp = mtod(m, struct cmsghdr *); 1466 m->m_len = 0; 1467 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), 1468 ("sbcreatecontrol: short mbuf")); 1469 /* 1470 * Don't leave the padding between the msg header and the 1471 * cmsg data and the padding after the cmsg data un-initialized. 1472 */ 1473 bzero(cp, CMSG_SPACE((u_int)size)); 1474 if (p != NULL) 1475 (void)memcpy(CMSG_DATA(cp), p, size); 1476 m->m_len = CMSG_SPACE(size); 1477 cp->cmsg_len = CMSG_LEN(size); 1478 cp->cmsg_level = level; 1479 cp->cmsg_type = type; 1480 return (m); 1481 } 1482 1483 /* 1484 * This does the same for socket buffers that sotoxsocket does for sockets: 1485 * generate an user-format data structure describing the socket buffer. Note 1486 * that the xsockbuf structure, since it is always embedded in a socket, does 1487 * not include a self pointer nor a length. We make this entry point public 1488 * in case some other mechanism needs it. 1489 */ 1490 void 1491 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 1492 { 1493 1494 xsb->sb_cc = sb->sb_ccc; 1495 xsb->sb_hiwat = sb->sb_hiwat; 1496 xsb->sb_mbcnt = sb->sb_mbcnt; 1497 xsb->sb_mcnt = sb->sb_mcnt; 1498 xsb->sb_ccnt = sb->sb_ccnt; 1499 xsb->sb_mbmax = sb->sb_mbmax; 1500 xsb->sb_lowat = sb->sb_lowat; 1501 xsb->sb_flags = sb->sb_flags; 1502 xsb->sb_timeo = sb->sb_timeo; 1503 } 1504 1505 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 1506 static int dummy; 1507 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 1508 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW, 1509 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size"); 1510 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1511 &sb_efficiency, 0, "Socket buffer size waste factor"); 1512