uipc_mbuf.c revision 116182
1275970Scy/* 2275970Scy * Copyright (c) 1982, 1986, 1988, 1991, 1993 3275970Scy * The Regents of the University of California. All rights reserved. 4275970Scy * 5275970Scy * Redistribution and use in source and binary forms, with or without 6275970Scy * modification, are permitted provided that the following conditions 7275970Scy * are met: 8275970Scy * 1. Redistributions of source code must retain the above copyright 9275970Scy * notice, this list of conditions and the following disclaimer. 10275970Scy * 2. Redistributions in binary form must reproduce the above copyright 11275970Scy * notice, this list of conditions and the following disclaimer in the 12275970Scy * documentation and/or other materials provided with the distribution. 13275970Scy * 3. All advertising materials mentioning features or use of this software 14275970Scy * must display the following acknowledgement: 15275970Scy * This product includes software developed by the University of 16275970Scy * California, Berkeley and its contributors. 17275970Scy * 4. Neither the name of the University nor the names of its contributors 18275970Scy * may be used to endorse or promote products derived from this software 19275970Scy * without specific prior written permission. 20275970Scy * 21275970Scy * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22275970Scy * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23275970Scy * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24275970Scy * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25275970Scy * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26275970Scy * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27275970Scy * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28275970Scy * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29275970Scy * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30275970Scy * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31275970Scy * SUCH DAMAGE. 32275970Scy * 33275970Scy * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 34275970Scy */ 35275970Scy 36275970Scy#include <sys/cdefs.h> 37275970Scy__FBSDID("$FreeBSD: head/sys/kern/uipc_mbuf.c 116182 2003-06-11 00:56:59Z obrien $"); 38275970Scy 39275970Scy#include "opt_mac.h" 40275970Scy#include "opt_param.h" 41275970Scy#include "opt_mbuf_stress_test.h" 42275970Scy 43275970Scy#include <sys/param.h> 44275970Scy#include <sys/systm.h> 45275970Scy#include <sys/kernel.h> 46275970Scy#include <sys/lock.h> 47275970Scy#include <sys/mac.h> 48275970Scy#include <sys/malloc.h> 49275970Scy#include <sys/mbuf.h> 50275970Scy#include <sys/sysctl.h> 51275970Scy#include <sys/domain.h> 52275970Scy#include <sys/protosw.h> 53275970Scy 54275970Scyint max_linkhdr; 55275970Scyint max_protohdr; 56275970Scyint max_hdr; 57275970Scyint max_datalen; 58275970Scyint m_defragpackets; 59275970Scyint m_defragbytes; 60275970Scyint m_defraguseless; 61275970Scyint m_defragfailure; 62275970Scy#ifdef MBUF_STRESS_TEST 63275970Scyint m_defragrandomfailures; 64275970Scy#endif 65275970Scy 66275970Scy/* 67275970Scy * sysctl(8) exported objects 68275970Scy */ 69275970ScySYSCTL_DECL(_kern_ipc); 70275970ScySYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 71275970Scy &max_linkhdr, 0, ""); 72275970ScySYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 73275970Scy &max_protohdr, 0, ""); 74275970ScySYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, ""); 75275970ScySYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 76275970Scy &max_datalen, 0, ""); 77275970ScySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 78275970Scy &m_defragpackets, 0, ""); 79275970ScySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 80275970Scy &m_defragbytes, 0, ""); 81275970ScySYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 82275970Scy &m_defraguseless, 0, ""); 83275970ScySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 84275970Scy &m_defragfailure, 0, ""); 85275970Scy#ifdef MBUF_STRESS_TEST 86275970ScySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 87275970Scy &m_defragrandomfailures, 0, ""); 88275970Scy#endif 89275970Scy 90275970Scy/* 91275970Scy * "Move" mbuf pkthdr from "from" to "to". 92275970Scy * "from" must have M_PKTHDR set, and "to" must be empty. 93275970Scy */ 94275970Scyvoid 95275970Scym_move_pkthdr(struct mbuf *to, struct mbuf *from) 96275970Scy{ 97275970Scy 98275970Scy#if 0 99275970Scy /* see below for why these are not enabled */ 100275970Scy M_ASSERTPKTHDR(to); 101275970Scy /* Note: with MAC, this may not be a good assertion. */ 102275970Scy KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 103275970Scy ("m_move_pkthdr: to has tags")); 104275970Scy#endif 105275970Scy KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster")); 106275970Scy#ifdef MAC 107275970Scy /* 108275970Scy * XXXMAC: It could be this should also occur for non-MAC? 109275970Scy */ 110275970Scy if (to->m_flags & M_PKTHDR) 111275970Scy m_tag_delete_chain(to, NULL); 112275970Scy#endif 113275970Scy to->m_flags = from->m_flags & M_COPYFLAGS; 114275970Scy to->m_data = to->m_pktdat; 115275970Scy to->m_pkthdr = from->m_pkthdr; /* especially tags */ 116275970Scy SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 117275970Scy from->m_flags &= ~M_PKTHDR; 118275970Scy} 119275970Scy 120275970Scy/* 121275970Scy * Duplicate "from"'s mbuf pkthdr in "to". 122275970Scy * "from" must have M_PKTHDR set, and "to" must be empty. 123275970Scy * In particular, this does a deep copy of the packet tags. 124275970Scy */ 125275970Scyint 126275970Scym_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 127275970Scy{ 128275970Scy 129275970Scy#if 0 130275970Scy /* 131275970Scy * The mbuf allocator only initializes the pkthdr 132275970Scy * when the mbuf is allocated with MGETHDR. Many users 133275970Scy * (e.g. m_copy*, m_prepend) use MGET and then 134275970Scy * smash the pkthdr as needed causing these 135275970Scy * assertions to trip. For now just disable them. 136275970Scy */ 137275970Scy M_ASSERTPKTHDR(to); 138275970Scy /* Note: with MAC, this may not be a good assertion. */ 139275970Scy KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 140275970Scy#endif 141275970Scy#ifdef MAC 142275970Scy if (to->m_flags & M_PKTHDR) 143275970Scy m_tag_delete_chain(to, NULL); 144275970Scy#endif 145275970Scy to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 146275970Scy if ((to->m_flags & M_EXT) == 0) 147275970Scy to->m_data = to->m_pktdat; 148275970Scy to->m_pkthdr = from->m_pkthdr; 149275970Scy SLIST_INIT(&to->m_pkthdr.tags); 150275970Scy return (m_tag_copy_chain(to, from, MBTOM(how))); 151275970Scy} 152275970Scy 153275970Scy/* 154275970Scy * Lesser-used path for M_PREPEND: 155275970Scy * allocate new mbuf to prepend to chain, 156275970Scy * copy junk along. 157275970Scy */ 158275970Scystruct mbuf * 159275970Scym_prepend(struct mbuf *m, int len, int how) 160275970Scy{ 161275970Scy struct mbuf *mn; 162275970Scy 163275970Scy MGET(mn, how, m->m_type); 164275970Scy if (mn == NULL) { 165275970Scy m_freem(m); 166275970Scy return (NULL); 167275970Scy } 168275970Scy if (m->m_flags & M_PKTHDR) 169275970Scy M_MOVE_PKTHDR(mn, m); 170275970Scy mn->m_next = m; 171275970Scy m = mn; 172275970Scy if (len < MHLEN) 173275970Scy MH_ALIGN(m, len); 174275970Scy m->m_len = len; 175275970Scy return (m); 176275970Scy} 177275970Scy 178275970Scy/* 179275970Scy * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 180275970Scy * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 181275970Scy * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller. 182275970Scy * Note that the copy is read-only, because clusters are not copied, 183275970Scy * only their reference counts are incremented. 184275970Scy */ 185275970Scystruct mbuf * 186275970Scym_copym(struct mbuf *m, int off0, int len, int wait) 187275970Scy{ 188275970Scy struct mbuf *n, **np; 189275970Scy int off = off0; 190275970Scy struct mbuf *top; 191275970Scy int copyhdr = 0; 192275970Scy 193275970Scy KASSERT(off >= 0, ("m_copym, negative off %d", off)); 194275970Scy KASSERT(len >= 0, ("m_copym, negative len %d", len)); 195275970Scy if (off == 0 && m->m_flags & M_PKTHDR) 196275970Scy copyhdr = 1; 197275970Scy while (off > 0) { 198275970Scy KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 199275970Scy if (off < m->m_len) 200275970Scy break; 201275970Scy off -= m->m_len; 202275970Scy m = m->m_next; 203275970Scy } 204275970Scy np = ⊤ 205275970Scy top = 0; 206275970Scy while (len > 0) { 207275970Scy if (m == NULL) { 208275970Scy KASSERT(len == M_COPYALL, 209275970Scy ("m_copym, length > size of mbuf chain")); 210275970Scy break; 211275970Scy } 212275970Scy MGET(n, wait, m->m_type); 213275970Scy *np = n; 214275970Scy if (n == NULL) 215275970Scy goto nospace; 216275970Scy if (copyhdr) { 217275970Scy if (!m_dup_pkthdr(n, m, wait)) 218275970Scy goto nospace; 219275970Scy if (len == M_COPYALL) 220275970Scy n->m_pkthdr.len -= off0; 221275970Scy else 222275970Scy n->m_pkthdr.len = len; 223275970Scy copyhdr = 0; 224275970Scy } 225275970Scy n->m_len = min(len, m->m_len - off); 226275970Scy if (m->m_flags & M_EXT) { 227275970Scy n->m_data = m->m_data + off; 228275970Scy n->m_ext = m->m_ext; 229275970Scy n->m_flags |= M_EXT; 230275970Scy MEXT_ADD_REF(m); 231275970Scy } else 232275970Scy bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 233275970Scy (u_int)n->m_len); 234275970Scy if (len != M_COPYALL) 235275970Scy len -= n->m_len; 236275970Scy off = 0; 237275970Scy m = m->m_next; 238275970Scy np = &n->m_next; 239275970Scy } 240275970Scy if (top == NULL) 241275970Scy mbstat.m_mcfail++; /* XXX: No consistency. */ 242275970Scy 243275970Scy return (top); 244275970Scynospace: 245275970Scy m_freem(top); 246275970Scy mbstat.m_mcfail++; /* XXX: No consistency. */ 247275970Scy return (NULL); 248275970Scy} 249275970Scy 250275970Scy/* 251275970Scy * Copy an entire packet, including header (which must be present). 252275970Scy * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 253275970Scy * Note that the copy is read-only, because clusters are not copied, 254275970Scy * only their reference counts are incremented. 255275970Scy * Preserve alignment of the first mbuf so if the creator has left 256275970Scy * some room at the beginning (e.g. for inserting protocol headers) 257275970Scy * the copies still have the room available. 258275970Scy */ 259275970Scystruct mbuf * 260275970Scym_copypacket(struct mbuf *m, int how) 261275970Scy{ 262275970Scy struct mbuf *top, *n, *o; 263275970Scy 264275970Scy MGET(n, how, m->m_type); 265275970Scy top = n; 266275970Scy if (n == NULL) 267275970Scy goto nospace; 268275970Scy 269275970Scy if (!m_dup_pkthdr(n, m, how)) 270275970Scy goto nospace; 271275970Scy n->m_len = m->m_len; 272275970Scy if (m->m_flags & M_EXT) { 273275970Scy n->m_data = m->m_data; 274275970Scy n->m_ext = m->m_ext; 275275970Scy n->m_flags |= M_EXT; 276275970Scy MEXT_ADD_REF(m); 277275970Scy } else { 278275970Scy n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 279275970Scy bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 280275970Scy } 281275970Scy 282275970Scy m = m->m_next; 283275970Scy while (m) { 284275970Scy MGET(o, how, m->m_type); 285275970Scy if (o == NULL) 286275970Scy goto nospace; 287275970Scy 288275970Scy n->m_next = o; 289275970Scy n = n->m_next; 290275970Scy 291275970Scy n->m_len = m->m_len; 292275970Scy if (m->m_flags & M_EXT) { 293275970Scy n->m_data = m->m_data; 294275970Scy n->m_ext = m->m_ext; 295275970Scy n->m_flags |= M_EXT; 296275970Scy MEXT_ADD_REF(m); 297275970Scy } else { 298275970Scy bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 299275970Scy } 300275970Scy 301275970Scy m = m->m_next; 302275970Scy } 303275970Scy return top; 304275970Scynospace: 305275970Scy m_freem(top); 306275970Scy mbstat.m_mcfail++; /* XXX: No consistency. */ 307275970Scy return (NULL); 308275970Scy} 309275970Scy 310275970Scy/* 311275970Scy * Copy data from an mbuf chain starting "off" bytes from the beginning, 312275970Scy * continuing for "len" bytes, into the indicated buffer. 313275970Scy */ 314275970Scyvoid 315275970Scym_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 316275970Scy{ 317275970Scy u_int count; 318275970Scy 319275970Scy KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 320275970Scy KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 321275970Scy while (off > 0) { 322275970Scy KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 323275970Scy if (off < m->m_len) 324275970Scy break; 325275970Scy off -= m->m_len; 326275970Scy m = m->m_next; 327275970Scy } 328275970Scy while (len > 0) { 329275970Scy KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 330275970Scy count = min(m->m_len - off, len); 331275970Scy bcopy(mtod(m, caddr_t) + off, cp, count); 332275970Scy len -= count; 333275970Scy cp += count; 334275970Scy off = 0; 335275970Scy m = m->m_next; 336275970Scy } 337275970Scy} 338275970Scy 339275970Scy/* 340275970Scy * Copy a packet header mbuf chain into a completely new chain, including 341275970Scy * copying any mbuf clusters. Use this instead of m_copypacket() when 342275970Scy * you need a writable copy of an mbuf chain. 343275970Scy */ 344275970Scystruct mbuf * 345275970Scym_dup(struct mbuf *m, int how) 346275970Scy{ 347275970Scy struct mbuf **p, *top = NULL; 348275970Scy int remain, moff, nsize; 349275970Scy 350275970Scy /* Sanity check */ 351275970Scy if (m == NULL) 352275970Scy return (NULL); 353275970Scy M_ASSERTPKTHDR(m); 354275970Scy 355275970Scy /* While there's more data, get a new mbuf, tack it on, and fill it */ 356275970Scy remain = m->m_pkthdr.len; 357275970Scy moff = 0; 358275970Scy p = ⊤ 359275970Scy while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 360275970Scy struct mbuf *n; 361275970Scy 362275970Scy /* Get the next new mbuf */ 363275970Scy MGET(n, how, m->m_type); 364275970Scy if (n == NULL) 365275970Scy goto nospace; 366275970Scy if (top == NULL) { /* first one, must be PKTHDR */ 367275970Scy if (!m_dup_pkthdr(n, m, how)) 368275970Scy goto nospace; 369275970Scy nsize = MHLEN; 370275970Scy } else /* not the first one */ 371275970Scy nsize = MLEN; 372275970Scy if (remain >= MINCLSIZE) { 373275970Scy MCLGET(n, how); 374275970Scy if ((n->m_flags & M_EXT) == 0) { 375275970Scy (void)m_free(n); 376275970Scy goto nospace; 377275970Scy } 378275970Scy nsize = MCLBYTES; 379275970Scy } 380275970Scy n->m_len = 0; 381275970Scy 382275970Scy /* Link it into the new chain */ 383275970Scy *p = n; 384275970Scy p = &n->m_next; 385275970Scy 386275970Scy /* Copy data from original mbuf(s) into new mbuf */ 387275970Scy while (n->m_len < nsize && m != NULL) { 388275970Scy int chunk = min(nsize - n->m_len, m->m_len - moff); 389275970Scy 390275970Scy bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 391275970Scy moff += chunk; 392275970Scy n->m_len += chunk; 393275970Scy remain -= chunk; 394275970Scy if (moff == m->m_len) { 395282408Scy m = m->m_next; 396275970Scy moff = 0; 397275970Scy } 398275970Scy } 399275970Scy 400275970Scy /* Check correct total mbuf length */ 401275970Scy KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 402275970Scy ("%s: bogus m_pkthdr.len", __func__)); 403275970Scy } 404275970Scy return (top); 405275970Scy 406275970Scynospace: 407275970Scy m_freem(top); 408275970Scy mbstat.m_mcfail++; /* XXX: No consistency. */ 409275970Scy return (NULL); 410275970Scy} 411275970Scy 412275970Scy/* 413275970Scy * Concatenate mbuf chain n to m. 414275970Scy * Both chains must be of the same type (e.g. MT_DATA). 415275970Scy * Any m_pkthdr is not updated. 416275970Scy */ 417275970Scyvoid 418275970Scym_cat(struct mbuf *m, struct mbuf *n) 419275970Scy{ 420275970Scy while (m->m_next) 421275970Scy m = m->m_next; 422275970Scy while (n) { 423275970Scy if (m->m_flags & M_EXT || 424275970Scy m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 425275970Scy /* just join the two chains */ 426275970Scy m->m_next = n; 427275970Scy return; 428275970Scy } 429275970Scy /* splat the data from one into the other */ 430275970Scy bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 431275970Scy (u_int)n->m_len); 432275970Scy m->m_len += n->m_len; 433275970Scy n = m_free(n); 434275970Scy } 435275970Scy} 436275970Scy 437275970Scyvoid 438275970Scym_adj(struct mbuf *mp, int req_len) 439275970Scy{ 440275970Scy int len = req_len; 441275970Scy struct mbuf *m; 442275970Scy int count; 443275970Scy 444275970Scy if ((m = mp) == NULL) 445275970Scy return; 446275970Scy if (len >= 0) { 447275970Scy /* 448275970Scy * Trim from head. 449275970Scy */ 450275970Scy while (m != NULL && len > 0) { 451275970Scy if (m->m_len <= len) { 452275970Scy len -= m->m_len; 453275970Scy m->m_len = 0; 454275970Scy m = m->m_next; 455275970Scy } else { 456275970Scy m->m_len -= len; 457275970Scy m->m_data += len; 458275970Scy len = 0; 459275970Scy } 460275970Scy } 461275970Scy m = mp; 462275970Scy if (mp->m_flags & M_PKTHDR) 463275970Scy m->m_pkthdr.len -= (req_len - len); 464275970Scy } else { 465275970Scy /* 466275970Scy * Trim from tail. Scan the mbuf chain, 467275970Scy * calculating its length and finding the last mbuf. 468275970Scy * If the adjustment only affects this mbuf, then just 469275970Scy * adjust and return. Otherwise, rescan and truncate 470275970Scy * after the remaining size. 471275970Scy */ 472275970Scy len = -len; 473275970Scy count = 0; 474275970Scy for (;;) { 475275970Scy count += m->m_len; 476275970Scy if (m->m_next == (struct mbuf *)0) 477275970Scy break; 478275970Scy m = m->m_next; 479275970Scy } 480 if (m->m_len >= len) { 481 m->m_len -= len; 482 if (mp->m_flags & M_PKTHDR) 483 mp->m_pkthdr.len -= len; 484 return; 485 } 486 count -= len; 487 if (count < 0) 488 count = 0; 489 /* 490 * Correct length for chain is "count". 491 * Find the mbuf with last data, adjust its length, 492 * and toss data from remaining mbufs on chain. 493 */ 494 m = mp; 495 if (m->m_flags & M_PKTHDR) 496 m->m_pkthdr.len = count; 497 for (; m; m = m->m_next) { 498 if (m->m_len >= count) { 499 m->m_len = count; 500 break; 501 } 502 count -= m->m_len; 503 } 504 while (m->m_next) 505 (m = m->m_next) ->m_len = 0; 506 } 507} 508 509/* 510 * Rearange an mbuf chain so that len bytes are contiguous 511 * and in the data area of an mbuf (so that mtod and dtom 512 * will work for a structure of size len). Returns the resulting 513 * mbuf chain on success, frees it and returns null on failure. 514 * If there is room, it will add up to max_protohdr-len extra bytes to the 515 * contiguous region in an attempt to avoid being called next time. 516 */ 517struct mbuf * 518m_pullup(struct mbuf *n, int len) 519{ 520 struct mbuf *m; 521 int count; 522 int space; 523 524 /* 525 * If first mbuf has no cluster, and has room for len bytes 526 * without shifting current data, pullup into it, 527 * otherwise allocate a new mbuf to prepend to the chain. 528 */ 529 if ((n->m_flags & M_EXT) == 0 && 530 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 531 if (n->m_len >= len) 532 return (n); 533 m = n; 534 n = n->m_next; 535 len -= m->m_len; 536 } else { 537 if (len > MHLEN) 538 goto bad; 539 MGET(m, M_DONTWAIT, n->m_type); 540 if (m == NULL) 541 goto bad; 542 m->m_len = 0; 543 if (n->m_flags & M_PKTHDR) 544 M_MOVE_PKTHDR(m, n); 545 } 546 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 547 do { 548 count = min(min(max(len, max_protohdr), space), n->m_len); 549 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 550 (u_int)count); 551 len -= count; 552 m->m_len += count; 553 n->m_len -= count; 554 space -= count; 555 if (n->m_len) 556 n->m_data += count; 557 else 558 n = m_free(n); 559 } while (len > 0 && n); 560 if (len > 0) { 561 (void) m_free(m); 562 goto bad; 563 } 564 m->m_next = n; 565 return (m); 566bad: 567 m_freem(n); 568 mbstat.m_mpfail++; /* XXX: No consistency. */ 569 return (NULL); 570} 571 572/* 573 * Partition an mbuf chain in two pieces, returning the tail -- 574 * all but the first len0 bytes. In case of failure, it returns NULL and 575 * attempts to restore the chain to its original state. 576 * 577 * Note that the resulting mbufs might be read-only, because the new 578 * mbuf can end up sharing an mbuf cluster with the original mbuf if 579 * the "breaking point" happens to lie within a cluster mbuf. Use the 580 * M_WRITABLE() macro to check for this case. 581 */ 582struct mbuf * 583m_split(struct mbuf *m0, int len0, int wait) 584{ 585 struct mbuf *m, *n; 586 u_int len = len0, remain; 587 588 for (m = m0; m && len > m->m_len; m = m->m_next) 589 len -= m->m_len; 590 if (m == NULL) 591 return (NULL); 592 remain = m->m_len - len; 593 if (m0->m_flags & M_PKTHDR) { 594 MGETHDR(n, wait, m0->m_type); 595 if (n == NULL) 596 return (NULL); 597 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 598 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 599 m0->m_pkthdr.len = len0; 600 if (m->m_flags & M_EXT) 601 goto extpacket; 602 if (remain > MHLEN) { 603 /* m can't be the lead packet */ 604 MH_ALIGN(n, 0); 605 n->m_next = m_split(m, len, wait); 606 if (n->m_next == NULL) { 607 (void) m_free(n); 608 return (NULL); 609 } else { 610 n->m_len = 0; 611 return (n); 612 } 613 } else 614 MH_ALIGN(n, remain); 615 } else if (remain == 0) { 616 n = m->m_next; 617 m->m_next = NULL; 618 return (n); 619 } else { 620 MGET(n, wait, m->m_type); 621 if (n == NULL) 622 return (NULL); 623 M_ALIGN(n, remain); 624 } 625extpacket: 626 if (m->m_flags & M_EXT) { 627 n->m_flags |= M_EXT; 628 n->m_ext = m->m_ext; 629 MEXT_ADD_REF(m); 630 n->m_data = m->m_data + len; 631 } else { 632 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 633 } 634 n->m_len = remain; 635 m->m_len = len; 636 n->m_next = m->m_next; 637 m->m_next = NULL; 638 return (n); 639} 640/* 641 * Routine to copy from device local memory into mbufs. 642 * Note that `off' argument is offset into first mbuf of target chain from 643 * which to begin copying the data to. 644 */ 645struct mbuf * 646m_devget(char *buf, int totlen, int off, struct ifnet *ifp, 647 void (*copy)(char *from, caddr_t to, u_int len)) 648{ 649 struct mbuf *m; 650 struct mbuf *top = 0, **mp = ⊤ 651 int len; 652 653 if (off < 0 || off > MHLEN) 654 return (NULL); 655 656 MGETHDR(m, M_DONTWAIT, MT_DATA); 657 if (m == NULL) 658 return (NULL); 659 m->m_pkthdr.rcvif = ifp; 660 m->m_pkthdr.len = totlen; 661 len = MHLEN; 662 663 while (totlen > 0) { 664 if (top) { 665 MGET(m, M_DONTWAIT, MT_DATA); 666 if (m == NULL) { 667 m_freem(top); 668 return (NULL); 669 } 670 len = MLEN; 671 } 672 if (totlen + off >= MINCLSIZE) { 673 MCLGET(m, M_DONTWAIT); 674 if (m->m_flags & M_EXT) 675 len = MCLBYTES; 676 } else { 677 /* 678 * Place initial small packet/header at end of mbuf. 679 */ 680 if (top == NULL && totlen + off + max_linkhdr <= len) { 681 m->m_data += max_linkhdr; 682 len -= max_linkhdr; 683 } 684 } 685 if (off) { 686 m->m_data += off; 687 len -= off; 688 off = 0; 689 } 690 m->m_len = len = min(totlen, len); 691 if (copy) 692 copy(buf, mtod(m, caddr_t), (u_int)len); 693 else 694 bcopy(buf, mtod(m, caddr_t), (u_int)len); 695 buf += len; 696 *mp = m; 697 mp = &m->m_next; 698 totlen -= len; 699 } 700 return (top); 701} 702 703/* 704 * Copy data from a buffer back into the indicated mbuf chain, 705 * starting "off" bytes from the beginning, extending the mbuf 706 * chain if necessary. 707 */ 708void 709m_copyback(struct mbuf *m0, int off, int len, caddr_t cp) 710{ 711 int mlen; 712 struct mbuf *m = m0, *n; 713 int totlen = 0; 714 715 if (m0 == NULL) 716 return; 717 while (off > (mlen = m->m_len)) { 718 off -= mlen; 719 totlen += mlen; 720 if (m->m_next == NULL) { 721 n = m_get_clrd(M_DONTWAIT, m->m_type); 722 if (n == NULL) 723 goto out; 724 n->m_len = min(MLEN, len + off); 725 m->m_next = n; 726 } 727 m = m->m_next; 728 } 729 while (len > 0) { 730 mlen = min (m->m_len - off, len); 731 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 732 cp += mlen; 733 len -= mlen; 734 mlen += off; 735 off = 0; 736 totlen += mlen; 737 if (len == 0) 738 break; 739 if (m->m_next == NULL) { 740 n = m_get(M_DONTWAIT, m->m_type); 741 if (n == NULL) 742 break; 743 n->m_len = min(MLEN, len); 744 m->m_next = n; 745 } 746 m = m->m_next; 747 } 748out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 749 m->m_pkthdr.len = totlen; 750} 751 752void 753m_print(const struct mbuf *m) 754{ 755 int len; 756 const struct mbuf *m2; 757 758 len = m->m_pkthdr.len; 759 m2 = m; 760 while (len) { 761 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-"); 762 len -= m2->m_len; 763 m2 = m2->m_next; 764 } 765 return; 766} 767 768u_int 769m_fixhdr(struct mbuf *m0) 770{ 771 u_int len; 772 773 len = m_length(m0, NULL); 774 m0->m_pkthdr.len = len; 775 return (len); 776} 777 778u_int 779m_length(struct mbuf *m0, struct mbuf **last) 780{ 781 struct mbuf *m; 782 u_int len; 783 784 len = 0; 785 for (m = m0; m != NULL; m = m->m_next) { 786 len += m->m_len; 787 if (m->m_next == NULL) 788 break; 789 } 790 if (last != NULL) 791 *last = m; 792 return (len); 793} 794 795/* 796 * Defragment a mbuf chain, returning the shortest possible 797 * chain of mbufs and clusters. If allocation fails and 798 * this cannot be completed, NULL will be returned, but 799 * the passed in chain will be unchanged. Upon success, 800 * the original chain will be freed, and the new chain 801 * will be returned. 802 * 803 * If a non-packet header is passed in, the original 804 * mbuf (chain?) will be returned unharmed. 805 */ 806struct mbuf * 807m_defrag(struct mbuf *m0, int how) 808{ 809 struct mbuf *m_new = NULL, *m_final = NULL; 810 int progress = 0, length; 811 812 if (!(m0->m_flags & M_PKTHDR)) 813 return (m0); 814 815#ifdef MBUF_STRESS_TEST 816 if (m_defragrandomfailures) { 817 int temp = arc4random() & 0xff; 818 if (temp == 0xba) 819 goto nospace; 820 } 821#endif 822 823 if (m0->m_pkthdr.len > MHLEN) 824 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 825 else 826 m_final = m_gethdr(how, MT_DATA); 827 828 if (m_final == NULL) 829 goto nospace; 830 831 if (m_dup_pkthdr(m_final, m0, how) == NULL) 832 goto nospace; 833 834 m_new = m_final; 835 836 while (progress < m0->m_pkthdr.len) { 837 length = m0->m_pkthdr.len - progress; 838 if (length > MCLBYTES) 839 length = MCLBYTES; 840 841 if (m_new == NULL) { 842 if (length > MLEN) 843 m_new = m_getcl(how, MT_DATA, 0); 844 else 845 m_new = m_get(how, MT_DATA); 846 if (m_new == NULL) 847 goto nospace; 848 } 849 850 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 851 progress += length; 852 m_new->m_len = length; 853 if (m_new != m_final) 854 m_cat(m_final, m_new); 855 m_new = NULL; 856 } 857 if (m0->m_next == NULL) 858 m_defraguseless++; 859 m_freem(m0); 860 m0 = m_final; 861 m_defragpackets++; 862 m_defragbytes += m0->m_pkthdr.len; 863 return (m0); 864nospace: 865 m_defragfailure++; 866 if (m_new) 867 m_free(m_new); 868 if (m_final) 869 m_freem(m_final); 870 return (NULL); 871} 872