uipc_mbuf.c revision 12662
144743Smarkm/* 244743Smarkm * Copyright (c) 1982, 1986, 1988, 1991, 1993 344743Smarkm * The Regents of the University of California. All rights reserved. 444743Smarkm * 544743Smarkm * Redistribution and use in source and binary forms, with or without 656977Sshin * modification, are permitted provided that the following conditions 756977Sshin * are met: 844743Smarkm * 1. Redistributions of source code must retain the above copyright 944743Smarkm * notice, this list of conditions and the following disclaimer. 1044743Smarkm * 2. Redistributions in binary form must reproduce the above copyright 1144743Smarkm * notice, this list of conditions and the following disclaimer in the 1244743Smarkm * documentation and/or other materials provided with the distribution. 1344743Smarkm * 3. All advertising materials mentioning features or use of this software 1444743Smarkm * must display the following acknowledgement: 1544743Smarkm * This product includes software developed by the University of 1656977Sshin * California, Berkeley and its contributors. 1756977Sshin * 4. Neither the name of the University nor the names of its contributors 1856977Sshin * may be used to endorse or promote products derived from this software 1944743Smarkm * without specific prior written permission. 2044743Smarkm * 2144743Smarkm * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2244743Smarkm * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2344743Smarkm * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2444743Smarkm * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2544743Smarkm * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2644743Smarkm * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2744743Smarkm * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2844743Smarkm * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2944743Smarkm * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3044743Smarkm * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3144743Smarkm * SUCH DAMAGE. 3244743Smarkm * 3344743Smarkm * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 3444743Smarkm * $Id: uipc_mbuf.c,v 1.15 1995/12/02 18:58:42 bde Exp $ 3544743Smarkm */ 3644743Smarkm 37210386Srpaulo#include <sys/param.h> 3844743Smarkm#include <sys/systm.h> 3944743Smarkm#include <sys/proc.h> 4044743Smarkm#include <sys/malloc.h> 4144743Smarkm#define MBTYPES 4244743Smarkm#include <sys/mbuf.h> 4344743Smarkm#include <sys/kernel.h> 4444743Smarkm#include <sys/syslog.h> 4544743Smarkm#include <sys/domain.h> 4644743Smarkm#include <sys/protosw.h> 4744743Smarkm 4844743Smarkm#include <vm/vm.h> 4944743Smarkm#include <vm/vm_param.h> 5056977Sshin#include <vm/vm_kern.h> 5156977Sshin#include <vm/vm_extern.h> 5256977Sshin 5344743Smarkmstatic void mbinit __P((void *)); 5456977SshinSYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL) 5556977Sshin 5656977Sshinstruct mbuf *mbutl; 5756977Sshinchar *mclrefcnt; 5856977Sshinstruct mbstat mbstat; 5956977Sshinunion mcluster *mclfree; 6056977Sshinint max_linkhdr; 6156977Sshinint max_protohdr; 6256977Sshinint max_hdr; 6356977Sshinint max_datalen; 6456977Sshin 6556977Sshin/* ARGSUSED*/ 6656977Sshinstatic void 6744743Smarkmmbinit(dummy) 6844743Smarkm void *dummy; 6944743Smarkm{ 7044743Smarkm int s; 7144743Smarkm 7244743Smarkm#if CLBYTES < 4096 7344743Smarkm#define NCL_INIT (4096/CLBYTES) 7444743Smarkm#else 7544743Smarkm#define NCL_INIT 1 7644743Smarkm#endif 7744743Smarkm s = splimp(); 7844743Smarkm if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0) 7944743Smarkm goto bad; 8044743Smarkm splx(s); 8144743Smarkm return; 8244743Smarkmbad: 8344743Smarkm panic("mbinit"); 8444743Smarkm} 8544743Smarkm 8644743Smarkm/* 8744743Smarkm * Allocate some number of mbuf clusters 8844743Smarkm * and place on cluster free list. 8944743Smarkm * Must be called at splimp. 9044743Smarkm */ 9144743Smarkm/* ARGSUSED */ 9244743Smarkmint 9344743Smarkmm_clalloc(ncl, nowait) 9444743Smarkm register int ncl; 9544743Smarkm int nowait; 9644743Smarkm{ 9744743Smarkm register caddr_t p; 9844743Smarkm register int i; 9944743Smarkm int npg; 10044743Smarkm 10144743Smarkm /* 10244743Smarkm * Once we run out of map space, it will be impossible 10344743Smarkm * to get any more (nothing is ever freed back to the 10444743Smarkm * map). 10544743Smarkm */ 10644743Smarkm if (mb_map_full) 10744743Smarkm return (0); 10844743Smarkm 10944743Smarkm npg = ncl * CLSIZE; 11044743Smarkm p = (caddr_t)kmem_malloc(mb_map, ctob(npg), 11144743Smarkm nowait ? M_NOWAIT : M_WAITOK); 11244743Smarkm /* 11344743Smarkm * Either the map is now full, or this is nowait and there 11444743Smarkm * are no pages left. 11544743Smarkm */ 11644743Smarkm if (p == NULL) 11744743Smarkm return (0); 11844743Smarkm 11944743Smarkm ncl = ncl * CLBYTES / MCLBYTES; 12044743Smarkm for (i = 0; i < ncl; i++) { 12144743Smarkm ((union mcluster *)p)->mcl_next = mclfree; 12244743Smarkm mclfree = (union mcluster *)p; 12344743Smarkm p += MCLBYTES; 12444743Smarkm mbstat.m_clfree++; 12544743Smarkm } 12644743Smarkm mbstat.m_clusters += ncl; 12744743Smarkm return (1); 12844743Smarkm} 12944743Smarkm 13044743Smarkm/* 13144743Smarkm * When MGET failes, ask protocols to free space when short of memory, 132 * then re-attempt to allocate an mbuf. 133 */ 134struct mbuf * 135m_retry(i, t) 136 int i, t; 137{ 138 register struct mbuf *m; 139 140 m_reclaim(); 141#define m_retry(i, t) (struct mbuf *)0 142 MGET(m, i, t); 143#undef m_retry 144 if (m != NULL) 145 mbstat.m_wait++; 146 else 147 mbstat.m_drops++; 148 return (m); 149} 150 151/* 152 * As above; retry an MGETHDR. 153 */ 154struct mbuf * 155m_retryhdr(i, t) 156 int i, t; 157{ 158 register struct mbuf *m; 159 160 m_reclaim(); 161#define m_retryhdr(i, t) (struct mbuf *)0 162 MGETHDR(m, i, t); 163#undef m_retryhdr 164 if (m != NULL) 165 mbstat.m_wait++; 166 else 167 mbstat.m_drops++; 168 return (m); 169} 170 171void 172m_reclaim() 173{ 174 register struct domain *dp; 175 register struct protosw *pr; 176 int s = splimp(); 177 178 for (dp = domains; dp; dp = dp->dom_next) 179 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 180 if (pr->pr_drain) 181 (*pr->pr_drain)(); 182 splx(s); 183 mbstat.m_drain++; 184} 185 186/* 187 * Space allocation routines. 188 * These are also available as macros 189 * for critical paths. 190 */ 191struct mbuf * 192m_get(nowait, type) 193 int nowait, type; 194{ 195 register struct mbuf *m; 196 197 MGET(m, nowait, type); 198 return (m); 199} 200 201struct mbuf * 202m_gethdr(nowait, type) 203 int nowait, type; 204{ 205 register struct mbuf *m; 206 207 MGETHDR(m, nowait, type); 208 return (m); 209} 210 211struct mbuf * 212m_getclr(nowait, type) 213 int nowait, type; 214{ 215 register struct mbuf *m; 216 217 MGET(m, nowait, type); 218 if (m == 0) 219 return (0); 220 bzero(mtod(m, caddr_t), MLEN); 221 return (m); 222} 223 224struct mbuf * 225m_free(m) 226 struct mbuf *m; 227{ 228 register struct mbuf *n; 229 230 MFREE(m, n); 231 return (n); 232} 233 234void 235m_freem(m) 236 register struct mbuf *m; 237{ 238 register struct mbuf *n; 239 240 if (m == NULL) 241 return; 242 do { 243 MFREE(m, n); 244 m = n; 245 } while (m); 246} 247 248/* 249 * Mbuffer utility routines. 250 */ 251 252/* 253 * Lesser-used path for M_PREPEND: 254 * allocate new mbuf to prepend to chain, 255 * copy junk along. 256 */ 257struct mbuf * 258m_prepend(m, len, how) 259 register struct mbuf *m; 260 int len, how; 261{ 262 struct mbuf *mn; 263 264 MGET(mn, how, m->m_type); 265 if (mn == (struct mbuf *)NULL) { 266 m_freem(m); 267 return ((struct mbuf *)NULL); 268 } 269 if (m->m_flags & M_PKTHDR) { 270 M_COPY_PKTHDR(mn, m); 271 m->m_flags &= ~M_PKTHDR; 272 } 273 mn->m_next = m; 274 m = mn; 275 if (len < MHLEN) 276 MH_ALIGN(m, len); 277 m->m_len = len; 278 return (m); 279} 280 281/* 282 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 283 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 284 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller. 285 */ 286int MCFail; 287 288struct mbuf * 289m_copym(m, off0, len, wait) 290 register struct mbuf *m; 291 int off0, wait; 292 register int len; 293{ 294 register struct mbuf *n, **np; 295 register int off = off0; 296 struct mbuf *top; 297 int copyhdr = 0; 298 299 if (off < 0 || len < 0) 300 panic("m_copym"); 301 if (off == 0 && m->m_flags & M_PKTHDR) 302 copyhdr = 1; 303 while (off > 0) { 304 if (m == 0) 305 panic("m_copym"); 306 if (off < m->m_len) 307 break; 308 off -= m->m_len; 309 m = m->m_next; 310 } 311 np = ⊤ 312 top = 0; 313 while (len > 0) { 314 if (m == 0) { 315 if (len != M_COPYALL) 316 panic("m_copym"); 317 break; 318 } 319 MGET(n, wait, m->m_type); 320 *np = n; 321 if (n == 0) 322 goto nospace; 323 if (copyhdr) { 324 M_COPY_PKTHDR(n, m); 325 if (len == M_COPYALL) 326 n->m_pkthdr.len -= off0; 327 else 328 n->m_pkthdr.len = len; 329 copyhdr = 0; 330 } 331 n->m_len = min(len, m->m_len - off); 332 if (m->m_flags & M_EXT) { 333 n->m_data = m->m_data + off; 334 mclrefcnt[mtocl(m->m_ext.ext_buf)]++; 335 n->m_ext = m->m_ext; 336 n->m_flags |= M_EXT; 337 } else 338 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 339 (unsigned)n->m_len); 340 if (len != M_COPYALL) 341 len -= n->m_len; 342 off = 0; 343 m = m->m_next; 344 np = &n->m_next; 345 } 346 if (top == 0) 347 MCFail++; 348 return (top); 349nospace: 350 m_freem(top); 351 MCFail++; 352 return (0); 353} 354 355/* 356 * Copy data from an mbuf chain starting "off" bytes from the beginning, 357 * continuing for "len" bytes, into the indicated buffer. 358 */ 359void 360m_copydata(m, off, len, cp) 361 register struct mbuf *m; 362 register int off; 363 register int len; 364 caddr_t cp; 365{ 366 register unsigned count; 367 368 if (off < 0 || len < 0) 369 panic("m_copydata"); 370 while (off > 0) { 371 if (m == 0) 372 panic("m_copydata"); 373 if (off < m->m_len) 374 break; 375 off -= m->m_len; 376 m = m->m_next; 377 } 378 while (len > 0) { 379 if (m == 0) 380 panic("m_copydata"); 381 count = min(m->m_len - off, len); 382 bcopy(mtod(m, caddr_t) + off, cp, count); 383 len -= count; 384 cp += count; 385 off = 0; 386 m = m->m_next; 387 } 388} 389 390/* 391 * Concatenate mbuf chain n to m. 392 * Both chains must be of the same type (e.g. MT_DATA). 393 * Any m_pkthdr is not updated. 394 */ 395void 396m_cat(m, n) 397 register struct mbuf *m, *n; 398{ 399 while (m->m_next) 400 m = m->m_next; 401 while (n) { 402 if (m->m_flags & M_EXT || 403 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 404 /* just join the two chains */ 405 m->m_next = n; 406 return; 407 } 408 /* splat the data from one into the other */ 409 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 410 (u_int)n->m_len); 411 m->m_len += n->m_len; 412 n = m_free(n); 413 } 414} 415 416void 417m_adj(mp, req_len) 418 struct mbuf *mp; 419 int req_len; 420{ 421 register int len = req_len; 422 register struct mbuf *m; 423 register count; 424 425 if ((m = mp) == NULL) 426 return; 427 if (len >= 0) { 428 /* 429 * Trim from head. 430 */ 431 while (m != NULL && len > 0) { 432 if (m->m_len <= len) { 433 len -= m->m_len; 434 m->m_len = 0; 435 m = m->m_next; 436 } else { 437 m->m_len -= len; 438 m->m_data += len; 439 len = 0; 440 } 441 } 442 m = mp; 443 if (mp->m_flags & M_PKTHDR) 444 m->m_pkthdr.len -= (req_len - len); 445 } else { 446 /* 447 * Trim from tail. Scan the mbuf chain, 448 * calculating its length and finding the last mbuf. 449 * If the adjustment only affects this mbuf, then just 450 * adjust and return. Otherwise, rescan and truncate 451 * after the remaining size. 452 */ 453 len = -len; 454 count = 0; 455 for (;;) { 456 count += m->m_len; 457 if (m->m_next == (struct mbuf *)0) 458 break; 459 m = m->m_next; 460 } 461 if (m->m_len >= len) { 462 m->m_len -= len; 463 if (mp->m_flags & M_PKTHDR) 464 mp->m_pkthdr.len -= len; 465 return; 466 } 467 count -= len; 468 if (count < 0) 469 count = 0; 470 /* 471 * Correct length for chain is "count". 472 * Find the mbuf with last data, adjust its length, 473 * and toss data from remaining mbufs on chain. 474 */ 475 m = mp; 476 if (m->m_flags & M_PKTHDR) 477 m->m_pkthdr.len = count; 478 for (; m; m = m->m_next) { 479 if (m->m_len >= count) { 480 m->m_len = count; 481 break; 482 } 483 count -= m->m_len; 484 } 485 while (m->m_next) 486 (m = m->m_next) ->m_len = 0; 487 } 488} 489 490/* 491 * Rearange an mbuf chain so that len bytes are contiguous 492 * and in the data area of an mbuf (so that mtod and dtom 493 * will work for a structure of size len). Returns the resulting 494 * mbuf chain on success, frees it and returns null on failure. 495 * If there is room, it will add up to max_protohdr-len extra bytes to the 496 * contiguous region in an attempt to avoid being called next time. 497 */ 498int MPFail; 499 500struct mbuf * 501m_pullup(n, len) 502 register struct mbuf *n; 503 int len; 504{ 505 register struct mbuf *m; 506 register int count; 507 int space; 508 509 /* 510 * If first mbuf has no cluster, and has room for len bytes 511 * without shifting current data, pullup into it, 512 * otherwise allocate a new mbuf to prepend to the chain. 513 */ 514 if ((n->m_flags & M_EXT) == 0 && 515 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 516 if (n->m_len >= len) 517 return (n); 518 m = n; 519 n = n->m_next; 520 len -= m->m_len; 521 } else { 522 if (len > MHLEN) 523 goto bad; 524 MGET(m, M_DONTWAIT, n->m_type); 525 if (m == 0) 526 goto bad; 527 m->m_len = 0; 528 if (n->m_flags & M_PKTHDR) { 529 M_COPY_PKTHDR(m, n); 530 n->m_flags &= ~M_PKTHDR; 531 } 532 } 533 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 534 do { 535 count = min(min(max(len, max_protohdr), space), n->m_len); 536 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 537 (unsigned)count); 538 len -= count; 539 m->m_len += count; 540 n->m_len -= count; 541 space -= count; 542 if (n->m_len) 543 n->m_data += count; 544 else 545 n = m_free(n); 546 } while (len > 0 && n); 547 if (len > 0) { 548 (void) m_free(m); 549 goto bad; 550 } 551 m->m_next = n; 552 return (m); 553bad: 554 m_freem(n); 555 MPFail++; 556 return (0); 557} 558 559/* 560 * Partition an mbuf chain in two pieces, returning the tail -- 561 * all but the first len0 bytes. In case of failure, it returns NULL and 562 * attempts to restore the chain to its original state. 563 */ 564struct mbuf * 565m_split(m0, len0, wait) 566 register struct mbuf *m0; 567 int len0, wait; 568{ 569 register struct mbuf *m, *n; 570 unsigned len = len0, remain; 571 572 for (m = m0; m && len > m->m_len; m = m->m_next) 573 len -= m->m_len; 574 if (m == 0) 575 return (0); 576 remain = m->m_len - len; 577 if (m0->m_flags & M_PKTHDR) { 578 MGETHDR(n, wait, m0->m_type); 579 if (n == 0) 580 return (0); 581 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 582 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 583 m0->m_pkthdr.len = len0; 584 if (m->m_flags & M_EXT) 585 goto extpacket; 586 if (remain > MHLEN) { 587 /* m can't be the lead packet */ 588 MH_ALIGN(n, 0); 589 n->m_next = m_split(m, len, wait); 590 if (n->m_next == 0) { 591 (void) m_free(n); 592 return (0); 593 } else 594 return (n); 595 } else 596 MH_ALIGN(n, remain); 597 } else if (remain == 0) { 598 n = m->m_next; 599 m->m_next = 0; 600 return (n); 601 } else { 602 MGET(n, wait, m->m_type); 603 if (n == 0) 604 return (0); 605 M_ALIGN(n, remain); 606 } 607extpacket: 608 if (m->m_flags & M_EXT) { 609 n->m_flags |= M_EXT; 610 n->m_ext = m->m_ext; 611 mclrefcnt[mtocl(m->m_ext.ext_buf)]++; 612 m->m_ext.ext_size = 0; /* For Accounting XXXXXX danger */ 613 n->m_data = m->m_data + len; 614 } else { 615 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 616 } 617 n->m_len = remain; 618 m->m_len = len; 619 n->m_next = m->m_next; 620 m->m_next = 0; 621 return (n); 622} 623/* 624 * Routine to copy from device local memory into mbufs. 625 */ 626struct mbuf * 627m_devget(buf, totlen, off0, ifp, copy) 628 char *buf; 629 int totlen, off0; 630 struct ifnet *ifp; 631 void (*copy) __P((char *from, caddr_t to, u_int len)); 632{ 633 register struct mbuf *m; 634 struct mbuf *top = 0, **mp = ⊤ 635 register int off = off0, len; 636 register char *cp; 637 char *epkt; 638 639 cp = buf; 640 epkt = cp + totlen; 641 if (off) { 642 cp += off + 2 * sizeof(u_short); 643 totlen -= 2 * sizeof(u_short); 644 } 645 MGETHDR(m, M_DONTWAIT, MT_DATA); 646 if (m == 0) 647 return (0); 648 m->m_pkthdr.rcvif = ifp; 649 m->m_pkthdr.len = totlen; 650 m->m_len = MHLEN; 651 652 while (totlen > 0) { 653 if (top) { 654 MGET(m, M_DONTWAIT, MT_DATA); 655 if (m == 0) { 656 m_freem(top); 657 return (0); 658 } 659 m->m_len = MLEN; 660 } 661 len = min(totlen, epkt - cp); 662 if (len >= MINCLSIZE) { 663 MCLGET(m, M_DONTWAIT); 664 if (m->m_flags & M_EXT) 665 m->m_len = len = min(len, MCLBYTES); 666 else 667 len = m->m_len; 668 } else { 669 /* 670 * Place initial small packet/header at end of mbuf. 671 */ 672 if (len < m->m_len) { 673 if (top == 0 && len + max_linkhdr <= m->m_len) 674 m->m_data += max_linkhdr; 675 m->m_len = len; 676 } else 677 len = m->m_len; 678 } 679 if (copy) 680 copy(cp, mtod(m, caddr_t), (unsigned)len); 681 else 682 bcopy(cp, mtod(m, caddr_t), (unsigned)len); 683 cp += len; 684 *mp = m; 685 mp = &m->m_next; 686 totlen -= len; 687 if (cp == epkt) 688 cp = buf; 689 } 690 return (top); 691} 692 693/* 694 * Copy data from a buffer back into the indicated mbuf chain, 695 * starting "off" bytes from the beginning, extending the mbuf 696 * chain if necessary. 697 */ 698void 699m_copyback(m0, off, len, cp) 700 struct mbuf *m0; 701 register int off; 702 register int len; 703 caddr_t cp; 704{ 705 register int mlen; 706 register struct mbuf *m = m0, *n; 707 int totlen = 0; 708 709 if (m0 == 0) 710 return; 711 while (off > (mlen = m->m_len)) { 712 off -= mlen; 713 totlen += mlen; 714 if (m->m_next == 0) { 715 n = m_getclr(M_DONTWAIT, m->m_type); 716 if (n == 0) 717 goto out; 718 n->m_len = min(MLEN, len + off); 719 m->m_next = n; 720 } 721 m = m->m_next; 722 } 723 while (len > 0) { 724 mlen = min (m->m_len - off, len); 725 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 726 cp += mlen; 727 len -= mlen; 728 mlen += off; 729 off = 0; 730 totlen += mlen; 731 if (len == 0) 732 break; 733 if (m->m_next == 0) { 734 n = m_get(M_DONTWAIT, m->m_type); 735 if (n == 0) 736 break; 737 n->m_len = min(MLEN, len); 738 m->m_next = n; 739 } 740 m = m->m_next; 741 } 742out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 743 m->m_pkthdr.len = totlen; 744} 745