uipc_mbuf.c revision 63203
1/* 2 * Copyright (c) 1982, 1986, 1988, 1991, 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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 34 * $FreeBSD: head/sys/kern/uipc_mbuf.c 63203 2000-07-15 06:02:48Z alfred $ 35 */ 36 37#include "opt_param.h" 38#include <sys/param.h> 39#include <sys/systm.h> 40#include <sys/malloc.h> 41#include <sys/mbuf.h> 42#include <sys/kernel.h> 43#include <sys/sysctl.h> 44#include <sys/domain.h> 45#include <sys/protosw.h> 46 47#include <vm/vm.h> 48#include <vm/vm_kern.h> 49#include <vm/vm_extern.h> 50 51#ifdef INVARIANTS 52#include <machine/cpu.h> 53#endif 54 55static void mbinit __P((void *)); 56SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL) 57 58struct mbuf *mbutl; 59char *mclrefcnt; 60struct mbstat mbstat; 61u_long mbtypes[MT_NTYPES]; 62struct mbuf *mmbfree; 63union mcluster *mclfree; 64int max_linkhdr; 65int max_protohdr; 66int max_hdr; 67int max_datalen; 68int nmbclusters; 69int nmbufs; 70u_int m_mballoc_wid = 0; 71u_int m_clalloc_wid = 0; 72 73SYSCTL_DECL(_kern_ipc); 74SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 75 &max_linkhdr, 0, ""); 76SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 77 &max_protohdr, 0, ""); 78SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, ""); 79SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 80 &max_datalen, 0, ""); 81SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW, 82 &mbuf_wait, 0, ""); 83SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, ""); 84SYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes, 85 sizeof(mbtypes), "LU", ""); 86SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD, 87 &nmbclusters, 0, "Maximum number of mbuf clusters available"); 88SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0, 89 "Maximum number of mbufs available"); 90#ifndef NMBCLUSTERS 91#define NMBCLUSTERS (512 + MAXUSERS * 16) 92#endif 93TUNABLE_INT_DECL("kern.ipc.nmbclusters", NMBCLUSTERS, nmbclusters); 94TUNABLE_INT_DECL("kern.ipc.nmbufs", NMBCLUSTERS * 4, nmbufs); 95 96static void m_reclaim __P((void)); 97 98/* "number of clusters of pages" */ 99#define NCL_INIT 1 100 101#define NMB_INIT 16 102 103/* ARGSUSED*/ 104static void 105mbinit(dummy) 106 void *dummy; 107{ 108 int s; 109 110 mmbfree = NULL; mclfree = NULL; 111 mbstat.m_msize = MSIZE; 112 mbstat.m_mclbytes = MCLBYTES; 113 mbstat.m_minclsize = MINCLSIZE; 114 mbstat.m_mlen = MLEN; 115 mbstat.m_mhlen = MHLEN; 116 117 s = splimp(); 118 if (m_mballoc(NMB_INIT, M_DONTWAIT) == 0) 119 goto bad; 120#if MCLBYTES <= PAGE_SIZE 121 if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0) 122 goto bad; 123#else 124 /* It's OK to call contigmalloc in this context. */ 125 if (m_clalloc(16, M_WAIT) == 0) 126 goto bad; 127#endif 128 splx(s); 129 return; 130bad: 131 panic("mbinit"); 132} 133 134/* 135 * Allocate at least nmb mbufs and place on mbuf free list. 136 * Must be called at splimp. 137 */ 138/* ARGSUSED */ 139int 140m_mballoc(nmb, how) 141 register int nmb; 142 int how; 143{ 144 register caddr_t p; 145 register int i; 146 int nbytes; 147 148 /* 149 * If we've hit the mbuf limit, stop allocating from mb_map, 150 * (or trying to) in order to avoid dipping into the section of 151 * mb_map which we've "reserved" for clusters. 152 */ 153 if ((nmb + mbstat.m_mbufs) > nmbufs) 154 return (0); 155 156 /* 157 * Once we run out of map space, it will be impossible to get 158 * any more (nothing is ever freed back to the map) 159 * -- however you are not dead as m_reclaim might 160 * still be able to free a substantial amount of space. 161 * 162 * XXX Furthermore, we can also work with "recycled" mbufs (when 163 * we're calling with M_WAIT the sleep procedure will be woken 164 * up when an mbuf is freed. See m_mballoc_wait()). 165 */ 166 if (mb_map_full) 167 return (0); 168 169 nbytes = round_page(nmb * MSIZE); 170 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT); 171 if (p == 0 && how == M_WAIT) { 172 mbstat.m_wait++; 173 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_WAITOK); 174 } 175 176 /* 177 * Either the map is now full, or `how' is M_NOWAIT and there 178 * are no pages left. 179 */ 180 if (p == NULL) 181 return (0); 182 183 nmb = nbytes / MSIZE; 184 for (i = 0; i < nmb; i++) { 185 ((struct mbuf *)p)->m_next = mmbfree; 186 mmbfree = (struct mbuf *)p; 187 p += MSIZE; 188 } 189 mbstat.m_mbufs += nmb; 190 mbtypes[MT_FREE] += nmb; 191 return (1); 192} 193 194/* 195 * Once the mb_map has been exhausted and if the call to the allocation macros 196 * (or, in some cases, functions) is with M_WAIT, then it is necessary to rely 197 * solely on reclaimed mbufs. Here we wait for an mbuf to be freed for a 198 * designated (mbuf_wait) time. 199 */ 200struct mbuf * 201m_mballoc_wait(int caller, int type) 202{ 203 struct mbuf *p; 204 int s; 205 206 m_mballoc_wid++; 207 if ((tsleep(&m_mballoc_wid, PVM, "mballc", mbuf_wait)) == EWOULDBLOCK) 208 m_mballoc_wid--; 209 210 /* 211 * Now that we (think) that we've got something, we will redo an 212 * MGET, but avoid getting into another instance of m_mballoc_wait() 213 * XXX: We retry to fetch _even_ if the sleep timed out. This is left 214 * this way, purposely, in the [unlikely] case that an mbuf was 215 * freed but the sleep was not awakened in time. 216 */ 217 p = NULL; 218 switch (caller) { 219 case MGET_C: 220 MGET(p, M_DONTWAIT, type); 221 break; 222 case MGETHDR_C: 223 MGETHDR(p, M_DONTWAIT, type); 224 break; 225 default: 226 panic("m_mballoc_wait: invalid caller (%d)", caller); 227 } 228 229 s = splimp(); 230 if (p != NULL) { /* We waited and got something... */ 231 mbstat.m_wait++; 232 /* Wake up another if we have more free. */ 233 if (mmbfree != NULL) 234 MMBWAKEUP(); 235 } 236 splx(s); 237 return (p); 238} 239 240#if MCLBYTES > PAGE_SIZE 241static int i_want_my_mcl; 242 243static void 244kproc_mclalloc(void) 245{ 246 int status; 247 248 while (1) { 249 tsleep(&i_want_my_mcl, PVM, "mclalloc", 0); 250 251 for (; i_want_my_mcl; i_want_my_mcl--) { 252 if (m_clalloc(1, M_WAIT) == 0) 253 printf("m_clalloc failed even in process context!\n"); 254 } 255 } 256} 257 258static struct proc *mclallocproc; 259static struct kproc_desc mclalloc_kp = { 260 "mclalloc", 261 kproc_mclalloc, 262 &mclallocproc 263}; 264SYSINIT(mclallocproc, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start, 265 &mclalloc_kp); 266#endif 267 268/* 269 * Allocate some number of mbuf clusters 270 * and place on cluster free list. 271 * Must be called at splimp. 272 */ 273/* ARGSUSED */ 274int 275m_clalloc(ncl, how) 276 register int ncl; 277 int how; 278{ 279 register caddr_t p; 280 register int i; 281 int npg; 282 283 /* 284 * If we've hit the mcluster number limit, stop allocating from 285 * mb_map, (or trying to) in order to avoid dipping into the section 286 * of mb_map which we've "reserved" for mbufs. 287 */ 288 if ((ncl + mbstat.m_clusters) > nmbclusters) { 289 mbstat.m_drops++; 290 return (0); 291 } 292 293 /* 294 * Once we run out of map space, it will be impossible 295 * to get any more (nothing is ever freed back to the 296 * map). From this point on, we solely rely on freed 297 * mclusters. 298 */ 299 if (mb_map_full) { 300 mbstat.m_drops++; 301 return (0); 302 } 303 304#if MCLBYTES > PAGE_SIZE 305 if (how != M_WAIT) { 306 i_want_my_mcl += ncl; 307 wakeup(&i_want_my_mcl); 308 mbstat.m_wait++; 309 p = 0; 310 } else { 311 p = contigmalloc1(MCLBYTES * ncl, M_DEVBUF, M_WAITOK, 0ul, 312 ~0ul, PAGE_SIZE, 0, mb_map); 313 } 314#else 315 npg = ncl; 316 p = (caddr_t)kmem_malloc(mb_map, ctob(npg), 317 how != M_WAIT ? M_NOWAIT : M_WAITOK); 318 ncl = ncl * PAGE_SIZE / MCLBYTES; 319#endif 320 /* 321 * Either the map is now full, or `how' is M_NOWAIT and there 322 * are no pages left. 323 */ 324 if (p == NULL) { 325 mbstat.m_drops++; 326 return (0); 327 } 328 329 for (i = 0; i < ncl; i++) { 330 ((union mcluster *)p)->mcl_next = mclfree; 331 mclfree = (union mcluster *)p; 332 p += MCLBYTES; 333 mbstat.m_clfree++; 334 } 335 mbstat.m_clusters += ncl; 336 return (1); 337} 338 339/* 340 * Once the mb_map submap has been exhausted and the allocation is called with 341 * M_WAIT, we rely on the mclfree union pointers. If nothing is free, we will 342 * sleep for a designated amount of time (mbuf_wait) or until we're woken up 343 * due to sudden mcluster availability. 344 */ 345caddr_t 346m_clalloc_wait(void) 347{ 348 caddr_t p; 349 int s; 350 351#ifdef __i386__ 352 /* If in interrupt context, and INVARIANTS, maintain sanity and die. */ 353 KASSERT(intr_nesting_level == 0, ("CLALLOC: CANNOT WAIT IN INTERRUPT")); 354#endif 355 356 /* Sleep until something's available or until we expire. */ 357 m_clalloc_wid++; 358 if ((tsleep(&m_clalloc_wid, PVM, "mclalc", mbuf_wait)) == EWOULDBLOCK) 359 m_clalloc_wid--; 360 361 /* 362 * Now that we (think) that we've got something, we will redo and 363 * MGET, but avoid getting into another instance of m_clalloc_wait() 364 */ 365 p = NULL; 366 MCLALLOC(p, M_DONTWAIT); 367 368 s = splimp(); 369 if (p != NULL) { /* We waited and got something... */ 370 mbstat.m_wait++; 371 /* Wake up another if we have more free. */ 372 if (mclfree != NULL) 373 MCLWAKEUP(); 374 } 375 376 splx(s); 377 return (p); 378} 379 380/* 381 * When MGET fails, ask protocols to free space when short of memory, 382 * then re-attempt to allocate an mbuf. 383 */ 384struct mbuf * 385m_retry(i, t) 386 int i, t; 387{ 388 register struct mbuf *m; 389 390 /* 391 * Must only do the reclaim if not in an interrupt context. 392 */ 393 if (i == M_WAIT) { 394#ifdef __i386__ 395 KASSERT(intr_nesting_level == 0, 396 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 397#endif 398 m_reclaim(); 399 } 400 401 /* 402 * Both m_mballoc_wait and m_retry must be nulled because 403 * when the MGET macro is run from here, we deffinately do _not_ 404 * want to enter an instance of m_mballoc_wait() or m_retry() (again!) 405 */ 406#define m_mballoc_wait(caller,type) (struct mbuf *)0 407#define m_retry(i, t) (struct mbuf *)0 408 MGET(m, i, t); 409#undef m_retry 410#undef m_mballoc_wait 411 412 if (m != NULL) 413 mbstat.m_wait++; 414 else 415 mbstat.m_drops++; 416 417 return (m); 418} 419 420/* 421 * As above; retry an MGETHDR. 422 */ 423struct mbuf * 424m_retryhdr(i, t) 425 int i, t; 426{ 427 register struct mbuf *m; 428 429 /* 430 * Must only do the reclaim if not in an interrupt context. 431 */ 432 if (i == M_WAIT) { 433#ifdef __i386__ 434 KASSERT(intr_nesting_level == 0, 435 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 436#endif 437 m_reclaim(); 438 } 439 440#define m_mballoc_wait(caller,type) (struct mbuf *)0 441#define m_retryhdr(i, t) (struct mbuf *)0 442 MGETHDR(m, i, t); 443#undef m_retryhdr 444#undef m_mballoc_wait 445 446 if (m != NULL) 447 mbstat.m_wait++; 448 else 449 mbstat.m_drops++; 450 451 return (m); 452} 453 454static void 455m_reclaim() 456{ 457 register struct domain *dp; 458 register struct protosw *pr; 459 int s = splimp(); 460 461 for (dp = domains; dp; dp = dp->dom_next) 462 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 463 if (pr->pr_drain) 464 (*pr->pr_drain)(); 465 splx(s); 466 mbstat.m_drain++; 467} 468 469/* 470 * Space allocation routines. 471 * These are also available as macros 472 * for critical paths. 473 */ 474struct mbuf * 475m_get(how, type) 476 int how, type; 477{ 478 register struct mbuf *m; 479 480 MGET(m, how, type); 481 return (m); 482} 483 484struct mbuf * 485m_gethdr(how, type) 486 int how, type; 487{ 488 register struct mbuf *m; 489 490 MGETHDR(m, how, type); 491 return (m); 492} 493 494struct mbuf * 495m_getclr(how, type) 496 int how, type; 497{ 498 register struct mbuf *m; 499 500 MGET(m, how, type); 501 if (m == 0) 502 return (0); 503 bzero(mtod(m, caddr_t), MLEN); 504 return (m); 505} 506 507struct mbuf * 508m_free(m) 509 struct mbuf *m; 510{ 511 register struct mbuf *n; 512 513 MFREE(m, n); 514 return (n); 515} 516 517void 518m_freem(m) 519 register struct mbuf *m; 520{ 521 register struct mbuf *n; 522 523 if (m == NULL) 524 return; 525 do { 526 /* 527 * we do need to check non-first mbuf, since some of existing 528 * code does not call M_PREPEND properly. 529 * (example: call to bpf_mtap from drivers) 530 */ 531 if ((m->m_flags & M_PKTHDR) != 0 && m->m_pkthdr.aux) { 532 m_freem(m->m_pkthdr.aux); 533 m->m_pkthdr.aux = NULL; 534 } 535 MFREE(m, n); 536 m = n; 537 } while (m); 538} 539 540/* 541 * Mbuffer utility routines. 542 */ 543 544/* 545 * Lesser-used path for M_PREPEND: 546 * allocate new mbuf to prepend to chain, 547 * copy junk along. 548 */ 549struct mbuf * 550m_prepend(m, len, how) 551 register struct mbuf *m; 552 int len, how; 553{ 554 struct mbuf *mn; 555 556 MGET(mn, how, m->m_type); 557 if (mn == (struct mbuf *)NULL) { 558 m_freem(m); 559 return ((struct mbuf *)NULL); 560 } 561 if (m->m_flags & M_PKTHDR) { 562 M_COPY_PKTHDR(mn, m); 563 m->m_flags &= ~M_PKTHDR; 564 } 565 mn->m_next = m; 566 m = mn; 567 if (len < MHLEN) 568 MH_ALIGN(m, len); 569 m->m_len = len; 570 return (m); 571} 572 573/* 574 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 575 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 576 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller. 577 * Note that the copy is read-only, because clusters are not copied, 578 * only their reference counts are incremented. 579 */ 580#define MCFail (mbstat.m_mcfail) 581 582struct mbuf * 583m_copym(m, off0, len, wait) 584 register struct mbuf *m; 585 int off0, wait; 586 register int len; 587{ 588 register struct mbuf *n, **np; 589 register int off = off0; 590 struct mbuf *top; 591 int copyhdr = 0; 592 593 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 594 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 595 if (off == 0 && m->m_flags & M_PKTHDR) 596 copyhdr = 1; 597 while (off > 0) { 598 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 599 if (off < m->m_len) 600 break; 601 off -= m->m_len; 602 m = m->m_next; 603 } 604 np = ⊤ 605 top = 0; 606 while (len > 0) { 607 if (m == 0) { 608 KASSERT(len == M_COPYALL, 609 ("m_copym, length > size of mbuf chain")); 610 break; 611 } 612 MGET(n, wait, m->m_type); 613 *np = n; 614 if (n == 0) 615 goto nospace; 616 if (copyhdr) { 617 M_COPY_PKTHDR(n, m); 618 if (len == M_COPYALL) 619 n->m_pkthdr.len -= off0; 620 else 621 n->m_pkthdr.len = len; 622 copyhdr = 0; 623 } 624 n->m_len = min(len, m->m_len - off); 625 if (m->m_flags & M_EXT) { 626 n->m_data = m->m_data + off; 627 if(!m->m_ext.ext_ref) 628 mclrefcnt[mtocl(m->m_ext.ext_buf)]++; 629 else 630 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf, 631 m->m_ext.ext_size); 632 n->m_ext = m->m_ext; 633 n->m_flags |= M_EXT; 634 } else 635 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 636 (unsigned)n->m_len); 637 if (len != M_COPYALL) 638 len -= n->m_len; 639 off = 0; 640 m = m->m_next; 641 np = &n->m_next; 642 } 643 if (top == 0) 644 MCFail++; 645 return (top); 646nospace: 647 m_freem(top); 648 MCFail++; 649 return (0); 650} 651 652/* 653 * Copy an entire packet, including header (which must be present). 654 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 655 * Note that the copy is read-only, because clusters are not copied, 656 * only their reference counts are incremented. 657 */ 658struct mbuf * 659m_copypacket(m, how) 660 struct mbuf *m; 661 int how; 662{ 663 struct mbuf *top, *n, *o; 664 665 MGET(n, how, m->m_type); 666 top = n; 667 if (!n) 668 goto nospace; 669 670 M_COPY_PKTHDR(n, m); 671 n->m_len = m->m_len; 672 if (m->m_flags & M_EXT) { 673 n->m_data = m->m_data; 674 if(!m->m_ext.ext_ref) 675 mclrefcnt[mtocl(m->m_ext.ext_buf)]++; 676 else 677 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf, 678 m->m_ext.ext_size); 679 n->m_ext = m->m_ext; 680 n->m_flags |= M_EXT; 681 } else { 682 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 683 } 684 685 m = m->m_next; 686 while (m) { 687 MGET(o, how, m->m_type); 688 if (!o) 689 goto nospace; 690 691 n->m_next = o; 692 n = n->m_next; 693 694 n->m_len = m->m_len; 695 if (m->m_flags & M_EXT) { 696 n->m_data = m->m_data; 697 if(!m->m_ext.ext_ref) 698 mclrefcnt[mtocl(m->m_ext.ext_buf)]++; 699 else 700 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf, 701 m->m_ext.ext_size); 702 n->m_ext = m->m_ext; 703 n->m_flags |= M_EXT; 704 } else { 705 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 706 } 707 708 m = m->m_next; 709 } 710 return top; 711nospace: 712 m_freem(top); 713 MCFail++; 714 return 0; 715} 716 717/* 718 * Copy data from an mbuf chain starting "off" bytes from the beginning, 719 * continuing for "len" bytes, into the indicated buffer. 720 */ 721void 722m_copydata(m, off, len, cp) 723 register struct mbuf *m; 724 register int off; 725 register int len; 726 caddr_t cp; 727{ 728 register unsigned count; 729 730 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 731 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 732 while (off > 0) { 733 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 734 if (off < m->m_len) 735 break; 736 off -= m->m_len; 737 m = m->m_next; 738 } 739 while (len > 0) { 740 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 741 count = min(m->m_len - off, len); 742 bcopy(mtod(m, caddr_t) + off, cp, count); 743 len -= count; 744 cp += count; 745 off = 0; 746 m = m->m_next; 747 } 748} 749 750/* 751 * Copy a packet header mbuf chain into a completely new chain, including 752 * copying any mbuf clusters. Use this instead of m_copypacket() when 753 * you need a writable copy of an mbuf chain. 754 */ 755struct mbuf * 756m_dup(m, how) 757 struct mbuf *m; 758 int how; 759{ 760 struct mbuf **p, *top = NULL; 761 int remain, moff, nsize; 762 763 /* Sanity check */ 764 if (m == NULL) 765 return (0); 766 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __FUNCTION__)); 767 768 /* While there's more data, get a new mbuf, tack it on, and fill it */ 769 remain = m->m_pkthdr.len; 770 moff = 0; 771 p = ⊤ 772 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 773 struct mbuf *n; 774 775 /* Get the next new mbuf */ 776 MGET(n, how, m->m_type); 777 if (n == NULL) 778 goto nospace; 779 if (top == NULL) { /* first one, must be PKTHDR */ 780 M_COPY_PKTHDR(n, m); 781 nsize = MHLEN; 782 } else /* not the first one */ 783 nsize = MLEN; 784 if (remain >= MINCLSIZE) { 785 MCLGET(n, how); 786 if ((n->m_flags & M_EXT) == 0) { 787 (void)m_free(n); 788 goto nospace; 789 } 790 nsize = MCLBYTES; 791 } 792 n->m_len = 0; 793 794 /* Link it into the new chain */ 795 *p = n; 796 p = &n->m_next; 797 798 /* Copy data from original mbuf(s) into new mbuf */ 799 while (n->m_len < nsize && m != NULL) { 800 int chunk = min(nsize - n->m_len, m->m_len - moff); 801 802 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 803 moff += chunk; 804 n->m_len += chunk; 805 remain -= chunk; 806 if (moff == m->m_len) { 807 m = m->m_next; 808 moff = 0; 809 } 810 } 811 812 /* Check correct total mbuf length */ 813 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 814 ("%s: bogus m_pkthdr.len", __FUNCTION__)); 815 } 816 return (top); 817 818nospace: 819 m_freem(top); 820 MCFail++; 821 return (0); 822} 823 824/* 825 * Concatenate mbuf chain n to m. 826 * Both chains must be of the same type (e.g. MT_DATA). 827 * Any m_pkthdr is not updated. 828 */ 829void 830m_cat(m, n) 831 register struct mbuf *m, *n; 832{ 833 while (m->m_next) 834 m = m->m_next; 835 while (n) { 836 if (m->m_flags & M_EXT || 837 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 838 /* just join the two chains */ 839 m->m_next = n; 840 return; 841 } 842 /* splat the data from one into the other */ 843 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 844 (u_int)n->m_len); 845 m->m_len += n->m_len; 846 n = m_free(n); 847 } 848} 849 850void 851m_adj(mp, req_len) 852 struct mbuf *mp; 853 int req_len; 854{ 855 register int len = req_len; 856 register struct mbuf *m; 857 register int count; 858 859 if ((m = mp) == NULL) 860 return; 861 if (len >= 0) { 862 /* 863 * Trim from head. 864 */ 865 while (m != NULL && len > 0) { 866 if (m->m_len <= len) { 867 len -= m->m_len; 868 m->m_len = 0; 869 m = m->m_next; 870 } else { 871 m->m_len -= len; 872 m->m_data += len; 873 len = 0; 874 } 875 } 876 m = mp; 877 if (mp->m_flags & M_PKTHDR) 878 m->m_pkthdr.len -= (req_len - len); 879 } else { 880 /* 881 * Trim from tail. Scan the mbuf chain, 882 * calculating its length and finding the last mbuf. 883 * If the adjustment only affects this mbuf, then just 884 * adjust and return. Otherwise, rescan and truncate 885 * after the remaining size. 886 */ 887 len = -len; 888 count = 0; 889 for (;;) { 890 count += m->m_len; 891 if (m->m_next == (struct mbuf *)0) 892 break; 893 m = m->m_next; 894 } 895 if (m->m_len >= len) { 896 m->m_len -= len; 897 if (mp->m_flags & M_PKTHDR) 898 mp->m_pkthdr.len -= len; 899 return; 900 } 901 count -= len; 902 if (count < 0) 903 count = 0; 904 /* 905 * Correct length for chain is "count". 906 * Find the mbuf with last data, adjust its length, 907 * and toss data from remaining mbufs on chain. 908 */ 909 m = mp; 910 if (m->m_flags & M_PKTHDR) 911 m->m_pkthdr.len = count; 912 for (; m; m = m->m_next) { 913 if (m->m_len >= count) { 914 m->m_len = count; 915 break; 916 } 917 count -= m->m_len; 918 } 919 while (m->m_next) 920 (m = m->m_next) ->m_len = 0; 921 } 922} 923 924/* 925 * Rearange an mbuf chain so that len bytes are contiguous 926 * and in the data area of an mbuf (so that mtod and dtom 927 * will work for a structure of size len). Returns the resulting 928 * mbuf chain on success, frees it and returns null on failure. 929 * If there is room, it will add up to max_protohdr-len extra bytes to the 930 * contiguous region in an attempt to avoid being called next time. 931 */ 932#define MPFail (mbstat.m_mpfail) 933 934struct mbuf * 935m_pullup(n, len) 936 register struct mbuf *n; 937 int len; 938{ 939 register struct mbuf *m; 940 register int count; 941 int space; 942 943 /* 944 * If first mbuf has no cluster, and has room for len bytes 945 * without shifting current data, pullup into it, 946 * otherwise allocate a new mbuf to prepend to the chain. 947 */ 948 if ((n->m_flags & M_EXT) == 0 && 949 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 950 if (n->m_len >= len) 951 return (n); 952 m = n; 953 n = n->m_next; 954 len -= m->m_len; 955 } else { 956 if (len > MHLEN) 957 goto bad; 958 MGET(m, M_DONTWAIT, n->m_type); 959 if (m == 0) 960 goto bad; 961 m->m_len = 0; 962 if (n->m_flags & M_PKTHDR) { 963 M_COPY_PKTHDR(m, n); 964 n->m_flags &= ~M_PKTHDR; 965 } 966 } 967 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 968 do { 969 count = min(min(max(len, max_protohdr), space), n->m_len); 970 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 971 (unsigned)count); 972 len -= count; 973 m->m_len += count; 974 n->m_len -= count; 975 space -= count; 976 if (n->m_len) 977 n->m_data += count; 978 else 979 n = m_free(n); 980 } while (len > 0 && n); 981 if (len > 0) { 982 (void) m_free(m); 983 goto bad; 984 } 985 m->m_next = n; 986 return (m); 987bad: 988 m_freem(n); 989 MPFail++; 990 return (0); 991} 992 993/* 994 * Partition an mbuf chain in two pieces, returning the tail -- 995 * all but the first len0 bytes. In case of failure, it returns NULL and 996 * attempts to restore the chain to its original state. 997 */ 998struct mbuf * 999m_split(m0, len0, wait) 1000 register struct mbuf *m0; 1001 int len0, wait; 1002{ 1003 register struct mbuf *m, *n; 1004 unsigned len = len0, remain; 1005 1006 for (m = m0; m && len > m->m_len; m = m->m_next) 1007 len -= m->m_len; 1008 if (m == 0) 1009 return (0); 1010 remain = m->m_len - len; 1011 if (m0->m_flags & M_PKTHDR) { 1012 MGETHDR(n, wait, m0->m_type); 1013 if (n == 0) 1014 return (0); 1015 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1016 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1017 m0->m_pkthdr.len = len0; 1018 if (m->m_flags & M_EXT) 1019 goto extpacket; 1020 if (remain > MHLEN) { 1021 /* m can't be the lead packet */ 1022 MH_ALIGN(n, 0); 1023 n->m_next = m_split(m, len, wait); 1024 if (n->m_next == 0) { 1025 (void) m_free(n); 1026 return (0); 1027 } else 1028 return (n); 1029 } else 1030 MH_ALIGN(n, remain); 1031 } else if (remain == 0) { 1032 n = m->m_next; 1033 m->m_next = 0; 1034 return (n); 1035 } else { 1036 MGET(n, wait, m->m_type); 1037 if (n == 0) 1038 return (0); 1039 M_ALIGN(n, remain); 1040 } 1041extpacket: 1042 if (m->m_flags & M_EXT) { 1043 n->m_flags |= M_EXT; 1044 n->m_ext = m->m_ext; 1045 if(!m->m_ext.ext_ref) 1046 mclrefcnt[mtocl(m->m_ext.ext_buf)]++; 1047 else 1048 (*(m->m_ext.ext_ref))(m->m_ext.ext_buf, 1049 m->m_ext.ext_size); 1050 m->m_ext.ext_size = 0; /* For Accounting XXXXXX danger */ 1051 n->m_data = m->m_data + len; 1052 } else { 1053 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1054 } 1055 n->m_len = remain; 1056 m->m_len = len; 1057 n->m_next = m->m_next; 1058 m->m_next = 0; 1059 return (n); 1060} 1061/* 1062 * Routine to copy from device local memory into mbufs. 1063 */ 1064struct mbuf * 1065m_devget(buf, totlen, off0, ifp, copy) 1066 char *buf; 1067 int totlen, off0; 1068 struct ifnet *ifp; 1069 void (*copy) __P((char *from, caddr_t to, u_int len)); 1070{ 1071 register struct mbuf *m; 1072 struct mbuf *top = 0, **mp = ⊤ 1073 register int off = off0, len; 1074 register char *cp; 1075 char *epkt; 1076 1077 cp = buf; 1078 epkt = cp + totlen; 1079 if (off) { 1080 cp += off + 2 * sizeof(u_short); 1081 totlen -= 2 * sizeof(u_short); 1082 } 1083 MGETHDR(m, M_DONTWAIT, MT_DATA); 1084 if (m == 0) 1085 return (0); 1086 m->m_pkthdr.rcvif = ifp; 1087 m->m_pkthdr.len = totlen; 1088 m->m_len = MHLEN; 1089 1090 while (totlen > 0) { 1091 if (top) { 1092 MGET(m, M_DONTWAIT, MT_DATA); 1093 if (m == 0) { 1094 m_freem(top); 1095 return (0); 1096 } 1097 m->m_len = MLEN; 1098 } 1099 len = min(totlen, epkt - cp); 1100 if (len >= MINCLSIZE) { 1101 MCLGET(m, M_DONTWAIT); 1102 if (m->m_flags & M_EXT) 1103 m->m_len = len = min(len, MCLBYTES); 1104 else 1105 len = m->m_len; 1106 } else { 1107 /* 1108 * Place initial small packet/header at end of mbuf. 1109 */ 1110 if (len < m->m_len) { 1111 if (top == 0 && len + max_linkhdr <= m->m_len) 1112 m->m_data += max_linkhdr; 1113 m->m_len = len; 1114 } else 1115 len = m->m_len; 1116 } 1117 if (copy) 1118 copy(cp, mtod(m, caddr_t), (unsigned)len); 1119 else 1120 bcopy(cp, mtod(m, caddr_t), (unsigned)len); 1121 cp += len; 1122 *mp = m; 1123 mp = &m->m_next; 1124 totlen -= len; 1125 if (cp == epkt) 1126 cp = buf; 1127 } 1128 return (top); 1129} 1130 1131/* 1132 * Copy data from a buffer back into the indicated mbuf chain, 1133 * starting "off" bytes from the beginning, extending the mbuf 1134 * chain if necessary. 1135 */ 1136void 1137m_copyback(m0, off, len, cp) 1138 struct mbuf *m0; 1139 register int off; 1140 register int len; 1141 caddr_t cp; 1142{ 1143 register int mlen; 1144 register struct mbuf *m = m0, *n; 1145 int totlen = 0; 1146 1147 if (m0 == 0) 1148 return; 1149 while (off > (mlen = m->m_len)) { 1150 off -= mlen; 1151 totlen += mlen; 1152 if (m->m_next == 0) { 1153 n = m_getclr(M_DONTWAIT, m->m_type); 1154 if (n == 0) 1155 goto out; 1156 n->m_len = min(MLEN, len + off); 1157 m->m_next = n; 1158 } 1159 m = m->m_next; 1160 } 1161 while (len > 0) { 1162 mlen = min (m->m_len - off, len); 1163 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 1164 cp += mlen; 1165 len -= mlen; 1166 mlen += off; 1167 off = 0; 1168 totlen += mlen; 1169 if (len == 0) 1170 break; 1171 if (m->m_next == 0) { 1172 n = m_get(M_DONTWAIT, m->m_type); 1173 if (n == 0) 1174 break; 1175 n->m_len = min(MLEN, len); 1176 m->m_next = n; 1177 } 1178 m = m->m_next; 1179 } 1180out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1181 m->m_pkthdr.len = totlen; 1182} 1183 1184void 1185m_print(const struct mbuf *m) 1186{ 1187 int len; 1188 const struct mbuf *m2; 1189 1190 len = m->m_pkthdr.len; 1191 m2 = m; 1192 while (len) { 1193 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-"); 1194 len -= m2->m_len; 1195 m2 = m2->m_next; 1196 } 1197 return; 1198} 1199