uipc_mbuf.c revision 149602
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 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: head/sys/kern/uipc_mbuf.c 149602 2005-08-29 20:15:33Z andre $"); 34 35#include "opt_mac.h" 36#include "opt_param.h" 37#include "opt_mbuf_stress_test.h" 38 39#include <sys/param.h> 40#include <sys/systm.h> 41#include <sys/kernel.h> 42#include <sys/limits.h> 43#include <sys/lock.h> 44#include <sys/mac.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/sysctl.h> 48#include <sys/domain.h> 49#include <sys/protosw.h> 50#include <sys/uio.h> 51 52int max_linkhdr; 53int max_protohdr; 54int max_hdr; 55int max_datalen; 56#ifdef MBUF_STRESS_TEST 57int m_defragpackets; 58int m_defragbytes; 59int m_defraguseless; 60int m_defragfailure; 61int m_defragrandomfailures; 62#endif 63 64/* 65 * sysctl(8) exported objects 66 */ 67SYSCTL_DECL(_kern_ipc); 68SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 69 &max_linkhdr, 0, ""); 70SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 71 &max_protohdr, 0, ""); 72SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, ""); 73SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 74 &max_datalen, 0, ""); 75#ifdef MBUF_STRESS_TEST 76SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 77 &m_defragpackets, 0, ""); 78SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 79 &m_defragbytes, 0, ""); 80SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 81 &m_defraguseless, 0, ""); 82SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 83 &m_defragfailure, 0, ""); 84SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 85 &m_defragrandomfailures, 0, ""); 86#endif 87 88/* 89 * Malloc-type for external ext_buf ref counts. 90 */ 91static MALLOC_DEFINE(M_MBUF, "mbextcnt", "mbuf external ref counts"); 92 93/* 94 * Allocate a given length worth of mbufs and/or clusters (whatever fits 95 * best) and return a pointer to the top of the allocated chain. If an 96 * existing mbuf chain is provided, then we will append the new chain 97 * to the existing one but still return the top of the newly allocated 98 * chain. 99 */ 100struct mbuf * 101m_getm(struct mbuf *m, int len, int how, short type) 102{ 103 struct mbuf *mb, *top, *cur, *mtail; 104 int num, rem; 105 int i; 106 107 KASSERT(len >= 0, ("m_getm(): len is < 0")); 108 109 /* If m != NULL, we will append to the end of that chain. */ 110 if (m != NULL) 111 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next); 112 else 113 mtail = NULL; 114 115 /* 116 * Calculate how many mbufs+clusters ("packets") we need and how much 117 * leftover there is after that and allocate the first mbuf+cluster 118 * if required. 119 */ 120 num = len / MCLBYTES; 121 rem = len % MCLBYTES; 122 top = cur = NULL; 123 if (num > 0) { 124 if ((top = cur = m_getcl(how, type, 0)) == NULL) 125 goto failed; 126 top->m_len = 0; 127 } 128 num--; 129 130 for (i = 0; i < num; i++) { 131 mb = m_getcl(how, type, 0); 132 if (mb == NULL) 133 goto failed; 134 mb->m_len = 0; 135 cur = (cur->m_next = mb); 136 } 137 if (rem > 0) { 138 mb = (rem > MINCLSIZE) ? 139 m_getcl(how, type, 0) : m_get(how, type); 140 if (mb == NULL) 141 goto failed; 142 mb->m_len = 0; 143 if (cur == NULL) 144 top = mb; 145 else 146 cur->m_next = mb; 147 } 148 149 if (mtail != NULL) 150 mtail->m_next = top; 151 return top; 152failed: 153 if (top != NULL) 154 m_freem(top); 155 return NULL; 156} 157 158/* 159 * Free an entire chain of mbufs and associated external buffers, if 160 * applicable. 161 */ 162void 163m_freem(struct mbuf *mb) 164{ 165 166 while (mb != NULL) 167 mb = m_free(mb); 168} 169 170/*- 171 * Configure a provided mbuf to refer to the provided external storage 172 * buffer and setup a reference count for said buffer. If the setting 173 * up of the reference count fails, the M_EXT bit will not be set. If 174 * successfull, the M_EXT bit is set in the mbuf's flags. 175 * 176 * Arguments: 177 * mb The existing mbuf to which to attach the provided buffer. 178 * buf The address of the provided external storage buffer. 179 * size The size of the provided buffer. 180 * freef A pointer to a routine that is responsible for freeing the 181 * provided external storage buffer. 182 * args A pointer to an argument structure (of any type) to be passed 183 * to the provided freef routine (may be NULL). 184 * flags Any other flags to be passed to the provided mbuf. 185 * type The type that the external storage buffer should be 186 * labeled with. 187 * 188 * Returns: 189 * Nothing. 190 */ 191void 192m_extadd(struct mbuf *mb, caddr_t buf, u_int size, 193 void (*freef)(void *, void *), void *args, int flags, int type) 194{ 195 u_int *ref_cnt = NULL; 196 197 /* XXX Shouldn't be adding EXT_CLUSTER with this API */ 198 if (type == EXT_CLUSTER) 199 ref_cnt = (u_int *)uma_find_refcnt(zone_clust, 200 mb->m_ext.ext_buf); 201 else if (type == EXT_EXTREF) 202 ref_cnt = __DEVOLATILE(u_int *, mb->m_ext.ref_cnt); 203 mb->m_ext.ref_cnt = (ref_cnt == NULL) ? 204 malloc(sizeof(u_int), M_MBUF, M_NOWAIT) : (u_int *)ref_cnt; 205 if (mb->m_ext.ref_cnt != NULL) { 206 *(mb->m_ext.ref_cnt) = 1; 207 mb->m_flags |= (M_EXT | flags); 208 mb->m_ext.ext_buf = buf; 209 mb->m_data = mb->m_ext.ext_buf; 210 mb->m_ext.ext_size = size; 211 mb->m_ext.ext_free = freef; 212 mb->m_ext.ext_args = args; 213 mb->m_ext.ext_type = type; 214 } 215} 216 217/* 218 * Non-directly-exported function to clean up after mbufs with M_EXT 219 * storage attached to them if the reference count hits 0. 220 */ 221void 222mb_free_ext(struct mbuf *m) 223{ 224 u_int cnt; 225 int dofree; 226 227 /* Account for lazy ref count assign. */ 228 if (m->m_ext.ref_cnt == NULL) 229 dofree = 1; 230 else 231 dofree = 0; 232 233 /* 234 * This is tricky. We need to make sure to decrement the 235 * refcount in a safe way but to also clean up if we're the 236 * last reference. This method seems to do it without race. 237 */ 238 while (dofree == 0) { 239 cnt = *(m->m_ext.ref_cnt); 240 if (atomic_cmpset_int(m->m_ext.ref_cnt, cnt, cnt - 1)) { 241 if (cnt == 1) 242 dofree = 1; 243 break; 244 } 245 } 246 247 if (dofree) { 248 /* 249 * Do the free, should be safe. 250 */ 251 if (m->m_ext.ext_type == EXT_PACKET) { 252 uma_zfree(zone_pack, m); 253 return; 254 } else if (m->m_ext.ext_type == EXT_CLUSTER) { 255 uma_zfree(zone_clust, m->m_ext.ext_buf); 256 m->m_ext.ext_buf = NULL; 257 } else { 258 (*(m->m_ext.ext_free))(m->m_ext.ext_buf, 259 m->m_ext.ext_args); 260 if (m->m_ext.ext_type != EXT_EXTREF) { 261 if (m->m_ext.ref_cnt != NULL) 262 free(__DEVOLATILE(u_int *, 263 m->m_ext.ref_cnt), M_MBUF); 264 m->m_ext.ref_cnt = NULL; 265 } 266 m->m_ext.ext_buf = NULL; 267 } 268 } 269 uma_zfree(zone_mbuf, m); 270} 271 272/* 273 * Clean up mbuf (chain) from any tags and packet headers. 274 * If "all" is set then the first mbuf in the chain will be 275 * cleaned too. 276 */ 277void 278m_demote(struct mbuf *m0, int all) 279{ 280 struct mbuf *m; 281 282 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { 283 if (m->m_flags & M_PKTHDR) { 284 m_tag_delete_chain(m, NULL); 285 m->m_flags &= ~M_PKTHDR; 286 bzero(&m->m_pkthdr, sizeof(struct pkthdr)); 287 } 288 if (m->m_type & MT_HEADER) 289 m->m_type = MT_DATA; 290 if (m != m0 && m->m_nextpkt) 291 m->m_nextpkt = NULL; 292 m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST); 293 } 294} 295 296/* 297 * Sanity checks on mbuf (chain). 298 * Returns 0 bad, 1 good, panic worse. 299 * sanitize, 0 run M_SANITY_ACTION, 1 garble things so they blow up later. 300 */ 301int 302m_sanity(struct mbuf *m0, int sanitize) 303{ 304 struct mbuf *m; 305 caddr_t a, b; 306 int pktlen = 0; 307 308#define M_SANITY_ACTION(s) return (0) 309/* #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) */ 310 311 m = m0; 312 while (m) { 313 /* 314 * Basic pointer checks. If any of these fails then some 315 * unrelated kernel memory before or after us is trashed. 316 * No way to recover from that. 317 */ 318 a = (m->m_flags & M_EXT ? m->m_ext.ext_buf : 319 (m->m_flags & M_PKTHDR ? (caddr_t)(&m->m_pktdat) : 320 (caddr_t)(&m->m_dat)) ); 321 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size : 322 (m->m_flags & M_PKTHDR ? MHLEN : MLEN))); 323 if ((caddr_t)m->m_data < a) 324 M_SANITY_ACTION("m_data outside mbuf data range left"); 325 if ((caddr_t)m->m_data > b) 326 M_SANITY_ACTION("m_data outside mbuf data range right"); 327 if ((caddr_t)m->m_data + m->m_len > b) 328 M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); 329 if (m->m_flags & M_PKTHDR && m->m_pkthdr.header) { 330 if ((caddr_t)m->m_pkthdr.header < a || 331 (caddr_t)m->m_pkthdr.header > b) 332 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range"); 333 } 334 335 /* m->m_nextpkt may only be set on first mbuf in chain. */ 336 if (m != m0 && m->m_nextpkt) { 337 if (sanitize) { 338 m_freem(m->m_nextpkt); 339 m->m_nextpkt = (struct mbuf *)0xDEADC0DE; 340 } else 341 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); 342 } 343 344 /* correct type correlations. */ 345 if (m->m_type == MT_HEADER && !(m->m_flags & M_PKTHDR)) { 346 if (sanitize) 347 m->m_type = MT_DATA; 348 else 349 M_SANITY_ACTION("MT_HEADER set but not M_PKTHDR"); 350 } 351 352 /* packet length (not mbuf length!) calculation */ 353 if (m0->m_flags & M_PKTHDR) 354 pktlen += m->m_len; 355 356 /* m_tags may only be attached to first mbuf in chain. */ 357 if (m != m0 && m->m_flags & M_PKTHDR && 358 !SLIST_EMPTY(&m->m_pkthdr.tags)) { 359 if (sanitize) { 360 m_tag_delete_chain(m, NULL); 361 /* put in 0xDEADC0DE perhaps? */ 362 } 363 else 364 M_SANITY_ACTION("m_tags on in-chain mbuf"); 365 } 366 367 /* M_PKTHDR may only be set on first mbuf in chain */ 368 if (m != m0 && m->m_flags & M_PKTHDR) { 369 if (sanitize) { 370 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); 371 m->m_flags &= ~M_PKTHDR; 372 /* put in 0xDEADCODE and leave hdr flag in */ 373 } else 374 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); 375 } 376 377 m = m->m_next; 378 } 379 if (pktlen && pktlen != m0->m_pkthdr.len) { 380 if (sanitize) 381 m0->m_pkthdr.len = 0; 382 else 383 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); 384 } 385#undef M_SANITY_ACTION 386 387 return 1; 388} 389 390 391/* 392 * "Move" mbuf pkthdr from "from" to "to". 393 * "from" must have M_PKTHDR set, and "to" must be empty. 394 */ 395void 396m_move_pkthdr(struct mbuf *to, struct mbuf *from) 397{ 398 399#if 0 400 /* see below for why these are not enabled */ 401 M_ASSERTPKTHDR(to); 402 /* Note: with MAC, this may not be a good assertion. */ 403 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 404 ("m_move_pkthdr: to has tags")); 405#endif 406#ifdef MAC 407 /* 408 * XXXMAC: It could be this should also occur for non-MAC? 409 */ 410 if (to->m_flags & M_PKTHDR) 411 m_tag_delete_chain(to, NULL); 412#endif 413 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 414 if ((to->m_flags & M_EXT) == 0) 415 to->m_data = to->m_pktdat; 416 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 417 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 418 from->m_flags &= ~M_PKTHDR; 419} 420 421/* 422 * Duplicate "from"'s mbuf pkthdr in "to". 423 * "from" must have M_PKTHDR set, and "to" must be empty. 424 * In particular, this does a deep copy of the packet tags. 425 */ 426int 427m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 428{ 429 430#if 0 431 /* 432 * The mbuf allocator only initializes the pkthdr 433 * when the mbuf is allocated with MGETHDR. Many users 434 * (e.g. m_copy*, m_prepend) use MGET and then 435 * smash the pkthdr as needed causing these 436 * assertions to trip. For now just disable them. 437 */ 438 M_ASSERTPKTHDR(to); 439 /* Note: with MAC, this may not be a good assertion. */ 440 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 441#endif 442 MBUF_CHECKSLEEP(how); 443#ifdef MAC 444 if (to->m_flags & M_PKTHDR) 445 m_tag_delete_chain(to, NULL); 446#endif 447 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 448 if ((to->m_flags & M_EXT) == 0) 449 to->m_data = to->m_pktdat; 450 to->m_pkthdr = from->m_pkthdr; 451 SLIST_INIT(&to->m_pkthdr.tags); 452 return (m_tag_copy_chain(to, from, MBTOM(how))); 453} 454 455/* 456 * Lesser-used path for M_PREPEND: 457 * allocate new mbuf to prepend to chain, 458 * copy junk along. 459 */ 460struct mbuf * 461m_prepend(struct mbuf *m, int len, int how) 462{ 463 struct mbuf *mn; 464 465 if (m->m_flags & M_PKTHDR) 466 MGETHDR(mn, how, m->m_type); 467 else 468 MGET(mn, how, m->m_type); 469 if (mn == NULL) { 470 m_freem(m); 471 return (NULL); 472 } 473 if (m->m_flags & M_PKTHDR) 474 M_MOVE_PKTHDR(mn, m); 475 mn->m_next = m; 476 m = mn; 477 if (len < MHLEN) 478 MH_ALIGN(m, len); 479 m->m_len = len; 480 return (m); 481} 482 483/* 484 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 485 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 486 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller. 487 * Note that the copy is read-only, because clusters are not copied, 488 * only their reference counts are incremented. 489 */ 490struct mbuf * 491m_copym(struct mbuf *m, int off0, int len, int wait) 492{ 493 struct mbuf *n, **np; 494 int off = off0; 495 struct mbuf *top; 496 int copyhdr = 0; 497 498 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 499 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 500 MBUF_CHECKSLEEP(wait); 501 if (off == 0 && m->m_flags & M_PKTHDR) 502 copyhdr = 1; 503 while (off > 0) { 504 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 505 if (off < m->m_len) 506 break; 507 off -= m->m_len; 508 m = m->m_next; 509 } 510 np = ⊤ 511 top = 0; 512 while (len > 0) { 513 if (m == NULL) { 514 KASSERT(len == M_COPYALL, 515 ("m_copym, length > size of mbuf chain")); 516 break; 517 } 518 if (copyhdr) 519 MGETHDR(n, wait, m->m_type); 520 else 521 MGET(n, wait, m->m_type); 522 *np = n; 523 if (n == NULL) 524 goto nospace; 525 if (copyhdr) { 526 if (!m_dup_pkthdr(n, m, wait)) 527 goto nospace; 528 if (len == M_COPYALL) 529 n->m_pkthdr.len -= off0; 530 else 531 n->m_pkthdr.len = len; 532 copyhdr = 0; 533 } 534 n->m_len = min(len, m->m_len - off); 535 if (m->m_flags & M_EXT) { 536 n->m_data = m->m_data + off; 537 n->m_ext = m->m_ext; 538 n->m_flags |= M_EXT; 539 MEXT_ADD_REF(m); 540 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 541 } else 542 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 543 (u_int)n->m_len); 544 if (len != M_COPYALL) 545 len -= n->m_len; 546 off = 0; 547 m = m->m_next; 548 np = &n->m_next; 549 } 550 if (top == NULL) 551 mbstat.m_mcfail++; /* XXX: No consistency. */ 552 553 return (top); 554nospace: 555 m_freem(top); 556 mbstat.m_mcfail++; /* XXX: No consistency. */ 557 return (NULL); 558} 559 560/* 561 * Returns mbuf chain with new head for the prepending case. 562 * Copies from mbuf (chain) n from off for len to mbuf (chain) m 563 * either prepending or appending the data. 564 * The resulting mbuf (chain) m is fully writeable. 565 * m is destination (is made writeable) 566 * n is source, off is offset in source, len is len from offset 567 * dir, 0 append, 1 prepend 568 * how, wait or nowait 569 */ 570 571static int 572m_bcopyxxx(void *s, void *t, u_int len) 573{ 574 bcopy(s, t, (size_t)len); 575 return 0; 576} 577 578struct mbuf * 579m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len, 580 int prep, int how) 581{ 582 struct mbuf *mm, *x, *z; 583 caddr_t p; 584 int i, mlen, nlen = 0; 585 caddr_t buf[MLEN]; 586 587 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source")); 588 KASSERT(off >= 0, ("m_copymdata, negative off %d", off)); 589 KASSERT(len >= 0, ("m_copymdata, negative len %d", len)); 590 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep)); 591 592 /* Make sure environment is sane. */ 593 for (z = m; z != NULL; z = z->m_next) { 594 mlen += z->m_len; 595 if (!M_WRITABLE(z)) { 596 /* Make clusters writeable. */ 597 if (z->m_flags & M_RDONLY) 598 return NULL; /* Can't handle ext ref. */ 599 x = m_getcl(how, MT_DATA, 0); 600 if (!x) 601 return NULL; 602 bcopy(z->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size); 603 p = x->m_ext.ext_buf + (z->m_data - z->m_ext.ext_buf); 604 MEXT_REM_REF(z); /* XXX */ 605 z->m_data = p; 606 x->m_flags &= ~M_EXT; 607 (void)m_free(x); 608 } 609 } 610 mm = prep ? m : z; 611 for (z = n; z != NULL; z = z->m_next) 612 nlen += z->m_len; 613 if (len == M_COPYALL) 614 len = nlen - off; 615 if (off + len > nlen || len < 1) 616 return NULL; 617 618 /* 619 * Append/prepend the data. Allocating mbufs as necessary. 620 */ 621 /* Shortcut if enough free space in first/last mbuf. */ 622 if (!prep && M_TRAILINGSPACE(mm) >= len) { 623 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) + 624 mm->m_len); 625 mm->m_len += len; 626 mm->m_pkthdr.len += len; 627 return m; 628 } 629 if (prep && M_LEADINGSPACE(mm) >= len) { 630 mm->m_data = mtod(mm, caddr_t) - len; 631 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t)); 632 mm->m_len += len; 633 mm->m_pkthdr.len += len; 634 return mm; 635 } 636 637 /* Expand first/last mbuf to cluster if possible. */ 638 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) { 639 bcopy(mm->m_data, &buf, mm->m_len); 640 m_clget(mm, how); 641 if (!(mm->m_flags & M_EXT)) 642 return NULL; 643 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len); 644 mm->m_data = mm->m_ext.ext_buf; 645 mm->m_pkthdr.header = NULL; 646 } 647 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) { 648 bcopy(mm->m_data, &buf, mm->m_len); 649 m_clget(mm, how); 650 if (!(mm->m_flags & M_EXT)) 651 return NULL; 652 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf + 653 mm->m_ext.ext_size - mm->m_len, mm->m_len); 654 mm->m_data = (caddr_t)mm->m_ext.ext_buf + 655 mm->m_ext.ext_size - mm->m_len; 656 mm->m_pkthdr.header = NULL; 657 } 658 659 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */ 660 if (!prep && len > M_TRAILINGSPACE(mm)) { 661 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA)) 662 return NULL; 663 } 664 if (prep && len > M_LEADINGSPACE(mm)) { 665 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA))) 666 return NULL; 667 i = 0; 668 for (x = z; x != NULL; x = x->m_next) { 669 i += x->m_flags & M_EXT ? x->m_ext.ext_size : 670 (x->m_flags & M_PKTHDR ? MHLEN : MLEN); 671 if (!x->m_next) 672 break; 673 } 674 z->m_data += i - len; 675 m_move_pkthdr(mm, z); 676 x->m_next = mm; 677 mm = z; 678 } 679 680 /* Seek to start position in source mbuf. Optimization for long chains. */ 681 while (off > 0) { 682 if (off < n->m_len) 683 break; 684 off -= n->m_len; 685 n = n->m_next; 686 } 687 688 /* Copy data into target mbuf. */ 689 z = mm; 690 while (len > 0) { 691 KASSERT(z != NULL, ("m_copymdata, falling off target edge")); 692 i = M_TRAILINGSPACE(z); 693 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len); 694 z->m_len += i; 695 /* fixup pkthdr.len if necessary */ 696 if ((prep ? mm : m)->m_flags & M_PKTHDR) 697 (prep ? mm : m)->m_pkthdr.len += i; 698 off += i; 699 len -= i; 700 z = z->m_next; 701 } 702 return (prep ? mm : m); 703} 704 705/* 706 * Copy an entire packet, including header (which must be present). 707 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 708 * Note that the copy is read-only, because clusters are not copied, 709 * only their reference counts are incremented. 710 * Preserve alignment of the first mbuf so if the creator has left 711 * some room at the beginning (e.g. for inserting protocol headers) 712 * the copies still have the room available. 713 */ 714struct mbuf * 715m_copypacket(struct mbuf *m, int how) 716{ 717 struct mbuf *top, *n, *o; 718 719 MBUF_CHECKSLEEP(how); 720 MGET(n, how, m->m_type); 721 top = n; 722 if (n == NULL) 723 goto nospace; 724 725 if (!m_dup_pkthdr(n, m, how)) 726 goto nospace; 727 n->m_len = m->m_len; 728 if (m->m_flags & M_EXT) { 729 n->m_data = m->m_data; 730 n->m_ext = m->m_ext; 731 n->m_flags |= M_EXT; 732 MEXT_ADD_REF(m); 733 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 734 } else { 735 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 736 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 737 } 738 739 m = m->m_next; 740 while (m) { 741 MGET(o, how, m->m_type); 742 if (o == NULL) 743 goto nospace; 744 745 n->m_next = o; 746 n = n->m_next; 747 748 n->m_len = m->m_len; 749 if (m->m_flags & M_EXT) { 750 n->m_data = m->m_data; 751 n->m_ext = m->m_ext; 752 n->m_flags |= M_EXT; 753 MEXT_ADD_REF(m); 754 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 755 } else { 756 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 757 } 758 759 m = m->m_next; 760 } 761 return top; 762nospace: 763 m_freem(top); 764 mbstat.m_mcfail++; /* XXX: No consistency. */ 765 return (NULL); 766} 767 768/* 769 * Copy data from an mbuf chain starting "off" bytes from the beginning, 770 * continuing for "len" bytes, into the indicated buffer. 771 */ 772void 773m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 774{ 775 u_int count; 776 777 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 778 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 779 while (off > 0) { 780 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 781 if (off < m->m_len) 782 break; 783 off -= m->m_len; 784 m = m->m_next; 785 } 786 while (len > 0) { 787 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 788 count = min(m->m_len - off, len); 789 bcopy(mtod(m, caddr_t) + off, cp, count); 790 len -= count; 791 cp += count; 792 off = 0; 793 m = m->m_next; 794 } 795} 796 797/* 798 * Copy a packet header mbuf chain into a completely new chain, including 799 * copying any mbuf clusters. Use this instead of m_copypacket() when 800 * you need a writable copy of an mbuf chain. 801 */ 802struct mbuf * 803m_dup(struct mbuf *m, int how) 804{ 805 struct mbuf **p, *top = NULL; 806 int remain, moff, nsize; 807 808 MBUF_CHECKSLEEP(how); 809 /* Sanity check */ 810 if (m == NULL) 811 return (NULL); 812 M_ASSERTPKTHDR(m); 813 814 /* While there's more data, get a new mbuf, tack it on, and fill it */ 815 remain = m->m_pkthdr.len; 816 moff = 0; 817 p = ⊤ 818 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 819 struct mbuf *n; 820 821 /* Get the next new mbuf */ 822 if (remain >= MINCLSIZE) { 823 n = m_getcl(how, m->m_type, 0); 824 nsize = MCLBYTES; 825 } else { 826 n = m_get(how, m->m_type); 827 nsize = MLEN; 828 } 829 if (n == NULL) 830 goto nospace; 831 832 if (top == NULL) { /* First one, must be PKTHDR */ 833 if (!m_dup_pkthdr(n, m, how)) { 834 m_free(n); 835 goto nospace; 836 } 837 nsize = MHLEN; 838 } 839 n->m_len = 0; 840 841 /* Link it into the new chain */ 842 *p = n; 843 p = &n->m_next; 844 845 /* Copy data from original mbuf(s) into new mbuf */ 846 while (n->m_len < nsize && m != NULL) { 847 int chunk = min(nsize - n->m_len, m->m_len - moff); 848 849 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 850 moff += chunk; 851 n->m_len += chunk; 852 remain -= chunk; 853 if (moff == m->m_len) { 854 m = m->m_next; 855 moff = 0; 856 } 857 } 858 859 /* Check correct total mbuf length */ 860 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 861 ("%s: bogus m_pkthdr.len", __func__)); 862 } 863 return (top); 864 865nospace: 866 m_freem(top); 867 mbstat.m_mcfail++; /* XXX: No consistency. */ 868 return (NULL); 869} 870 871/* 872 * Concatenate mbuf chain n to m. 873 * Both chains must be of the same type (e.g. MT_DATA). 874 * Any m_pkthdr is not updated. 875 */ 876void 877m_cat(struct mbuf *m, struct mbuf *n) 878{ 879 while (m->m_next) 880 m = m->m_next; 881 while (n) { 882 if (m->m_flags & M_EXT || 883 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 884 /* just join the two chains */ 885 m->m_next = n; 886 return; 887 } 888 /* splat the data from one into the other */ 889 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 890 (u_int)n->m_len); 891 m->m_len += n->m_len; 892 n = m_free(n); 893 } 894} 895 896void 897m_adj(struct mbuf *mp, int req_len) 898{ 899 int len = req_len; 900 struct mbuf *m; 901 int count; 902 903 if ((m = mp) == NULL) 904 return; 905 if (len >= 0) { 906 /* 907 * Trim from head. 908 */ 909 while (m != NULL && len > 0) { 910 if (m->m_len <= len) { 911 len -= m->m_len; 912 m->m_len = 0; 913 m = m->m_next; 914 } else { 915 m->m_len -= len; 916 m->m_data += len; 917 len = 0; 918 } 919 } 920 m = mp; 921 if (mp->m_flags & M_PKTHDR) 922 m->m_pkthdr.len -= (req_len - len); 923 } else { 924 /* 925 * Trim from tail. Scan the mbuf chain, 926 * calculating its length and finding the last mbuf. 927 * If the adjustment only affects this mbuf, then just 928 * adjust and return. Otherwise, rescan and truncate 929 * after the remaining size. 930 */ 931 len = -len; 932 count = 0; 933 for (;;) { 934 count += m->m_len; 935 if (m->m_next == (struct mbuf *)0) 936 break; 937 m = m->m_next; 938 } 939 if (m->m_len >= len) { 940 m->m_len -= len; 941 if (mp->m_flags & M_PKTHDR) 942 mp->m_pkthdr.len -= len; 943 return; 944 } 945 count -= len; 946 if (count < 0) 947 count = 0; 948 /* 949 * Correct length for chain is "count". 950 * Find the mbuf with last data, adjust its length, 951 * and toss data from remaining mbufs on chain. 952 */ 953 m = mp; 954 if (m->m_flags & M_PKTHDR) 955 m->m_pkthdr.len = count; 956 for (; m; m = m->m_next) { 957 if (m->m_len >= count) { 958 m->m_len = count; 959 if (m->m_next != NULL) { 960 m_freem(m->m_next); 961 m->m_next = NULL; 962 } 963 break; 964 } 965 count -= m->m_len; 966 } 967 } 968} 969 970/* 971 * Rearange an mbuf chain so that len bytes are contiguous 972 * and in the data area of an mbuf (so that mtod and dtom 973 * will work for a structure of size len). Returns the resulting 974 * mbuf chain on success, frees it and returns null on failure. 975 * If there is room, it will add up to max_protohdr-len extra bytes to the 976 * contiguous region in an attempt to avoid being called next time. 977 */ 978struct mbuf * 979m_pullup(struct mbuf *n, int len) 980{ 981 struct mbuf *m; 982 int count; 983 int space; 984 985 /* 986 * If first mbuf has no cluster, and has room for len bytes 987 * without shifting current data, pullup into it, 988 * otherwise allocate a new mbuf to prepend to the chain. 989 */ 990 if ((n->m_flags & M_EXT) == 0 && 991 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 992 if (n->m_len >= len) 993 return (n); 994 m = n; 995 n = n->m_next; 996 len -= m->m_len; 997 } else { 998 if (len > MHLEN) 999 goto bad; 1000 MGET(m, M_DONTWAIT, n->m_type); 1001 if (m == NULL) 1002 goto bad; 1003 m->m_len = 0; 1004 if (n->m_flags & M_PKTHDR) 1005 M_MOVE_PKTHDR(m, n); 1006 } 1007 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1008 do { 1009 count = min(min(max(len, max_protohdr), space), n->m_len); 1010 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1011 (u_int)count); 1012 len -= count; 1013 m->m_len += count; 1014 n->m_len -= count; 1015 space -= count; 1016 if (n->m_len) 1017 n->m_data += count; 1018 else 1019 n = m_free(n); 1020 } while (len > 0 && n); 1021 if (len > 0) { 1022 (void) m_free(m); 1023 goto bad; 1024 } 1025 m->m_next = n; 1026 return (m); 1027bad: 1028 m_freem(n); 1029 mbstat.m_mpfail++; /* XXX: No consistency. */ 1030 return (NULL); 1031} 1032 1033/* 1034 * Like m_pullup(), except a new mbuf is always allocated, and we allow 1035 * the amount of empty space before the data in the new mbuf to be specified 1036 * (in the event that the caller expects to prepend later). 1037 */ 1038int MSFail; 1039 1040struct mbuf * 1041m_copyup(struct mbuf *n, int len, int dstoff) 1042{ 1043 struct mbuf *m; 1044 int count, space; 1045 1046 if (len > (MHLEN - dstoff)) 1047 goto bad; 1048 MGET(m, M_DONTWAIT, n->m_type); 1049 if (m == NULL) 1050 goto bad; 1051 m->m_len = 0; 1052 if (n->m_flags & M_PKTHDR) 1053 M_MOVE_PKTHDR(m, n); 1054 m->m_data += dstoff; 1055 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1056 do { 1057 count = min(min(max(len, max_protohdr), space), n->m_len); 1058 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), 1059 (unsigned)count); 1060 len -= count; 1061 m->m_len += count; 1062 n->m_len -= count; 1063 space -= count; 1064 if (n->m_len) 1065 n->m_data += count; 1066 else 1067 n = m_free(n); 1068 } while (len > 0 && n); 1069 if (len > 0) { 1070 (void) m_free(m); 1071 goto bad; 1072 } 1073 m->m_next = n; 1074 return (m); 1075 bad: 1076 m_freem(n); 1077 MSFail++; 1078 return (NULL); 1079} 1080 1081/* 1082 * Partition an mbuf chain in two pieces, returning the tail -- 1083 * all but the first len0 bytes. In case of failure, it returns NULL and 1084 * attempts to restore the chain to its original state. 1085 * 1086 * Note that the resulting mbufs might be read-only, because the new 1087 * mbuf can end up sharing an mbuf cluster with the original mbuf if 1088 * the "breaking point" happens to lie within a cluster mbuf. Use the 1089 * M_WRITABLE() macro to check for this case. 1090 */ 1091struct mbuf * 1092m_split(struct mbuf *m0, int len0, int wait) 1093{ 1094 struct mbuf *m, *n; 1095 u_int len = len0, remain; 1096 1097 MBUF_CHECKSLEEP(wait); 1098 for (m = m0; m && len > m->m_len; m = m->m_next) 1099 len -= m->m_len; 1100 if (m == NULL) 1101 return (NULL); 1102 remain = m->m_len - len; 1103 if (m0->m_flags & M_PKTHDR) { 1104 MGETHDR(n, wait, m0->m_type); 1105 if (n == NULL) 1106 return (NULL); 1107 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1108 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1109 m0->m_pkthdr.len = len0; 1110 if (m->m_flags & M_EXT) 1111 goto extpacket; 1112 if (remain > MHLEN) { 1113 /* m can't be the lead packet */ 1114 MH_ALIGN(n, 0); 1115 n->m_next = m_split(m, len, wait); 1116 if (n->m_next == NULL) { 1117 (void) m_free(n); 1118 return (NULL); 1119 } else { 1120 n->m_len = 0; 1121 return (n); 1122 } 1123 } else 1124 MH_ALIGN(n, remain); 1125 } else if (remain == 0) { 1126 n = m->m_next; 1127 m->m_next = NULL; 1128 return (n); 1129 } else { 1130 MGET(n, wait, m->m_type); 1131 if (n == NULL) 1132 return (NULL); 1133 M_ALIGN(n, remain); 1134 } 1135extpacket: 1136 if (m->m_flags & M_EXT) { 1137 n->m_flags |= M_EXT; 1138 n->m_ext = m->m_ext; 1139 MEXT_ADD_REF(m); 1140 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 1141 n->m_data = m->m_data + len; 1142 } else { 1143 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1144 } 1145 n->m_len = remain; 1146 m->m_len = len; 1147 n->m_next = m->m_next; 1148 m->m_next = NULL; 1149 return (n); 1150} 1151/* 1152 * Routine to copy from device local memory into mbufs. 1153 * Note that `off' argument is offset into first mbuf of target chain from 1154 * which to begin copying the data to. 1155 */ 1156struct mbuf * 1157m_devget(char *buf, int totlen, int off, struct ifnet *ifp, 1158 void (*copy)(char *from, caddr_t to, u_int len)) 1159{ 1160 struct mbuf *m; 1161 struct mbuf *top = NULL, **mp = ⊤ 1162 int len; 1163 1164 if (off < 0 || off > MHLEN) 1165 return (NULL); 1166 1167 while (totlen > 0) { 1168 if (top == NULL) { /* First one, must be PKTHDR */ 1169 if (totlen + off >= MINCLSIZE) { 1170 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 1171 len = MCLBYTES; 1172 } else { 1173 m = m_gethdr(M_DONTWAIT, MT_DATA); 1174 len = MHLEN; 1175 1176 /* Place initial small packet/header at end of mbuf */ 1177 if (m && totlen + off + max_linkhdr <= MLEN) { 1178 m->m_data += max_linkhdr; 1179 len -= max_linkhdr; 1180 } 1181 } 1182 if (m == NULL) 1183 return NULL; 1184 m->m_pkthdr.rcvif = ifp; 1185 m->m_pkthdr.len = totlen; 1186 } else { 1187 if (totlen + off >= MINCLSIZE) { 1188 m = m_getcl(M_DONTWAIT, MT_DATA, 0); 1189 len = MCLBYTES; 1190 } else { 1191 m = m_get(M_DONTWAIT, MT_DATA); 1192 len = MLEN; 1193 } 1194 if (m == NULL) { 1195 m_freem(top); 1196 return NULL; 1197 } 1198 } 1199 if (off) { 1200 m->m_data += off; 1201 len -= off; 1202 off = 0; 1203 } 1204 m->m_len = len = min(totlen, len); 1205 if (copy) 1206 copy(buf, mtod(m, caddr_t), (u_int)len); 1207 else 1208 bcopy(buf, mtod(m, caddr_t), (u_int)len); 1209 buf += len; 1210 *mp = m; 1211 mp = &m->m_next; 1212 totlen -= len; 1213 } 1214 return (top); 1215} 1216 1217/* 1218 * Copy data from a buffer back into the indicated mbuf chain, 1219 * starting "off" bytes from the beginning, extending the mbuf 1220 * chain if necessary. 1221 */ 1222void 1223m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 1224{ 1225 int mlen; 1226 struct mbuf *m = m0, *n; 1227 int totlen = 0; 1228 1229 if (m0 == NULL) 1230 return; 1231 while (off > (mlen = m->m_len)) { 1232 off -= mlen; 1233 totlen += mlen; 1234 if (m->m_next == NULL) { 1235 n = m_get(M_DONTWAIT, m->m_type); 1236 if (n == NULL) 1237 goto out; 1238 bzero(mtod(n, caddr_t), MLEN); 1239 n->m_len = min(MLEN, len + off); 1240 m->m_next = n; 1241 } 1242 m = m->m_next; 1243 } 1244 while (len > 0) { 1245 mlen = min (m->m_len - off, len); 1246 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 1247 cp += mlen; 1248 len -= mlen; 1249 mlen += off; 1250 off = 0; 1251 totlen += mlen; 1252 if (len == 0) 1253 break; 1254 if (m->m_next == NULL) { 1255 n = m_get(M_DONTWAIT, m->m_type); 1256 if (n == NULL) 1257 break; 1258 n->m_len = min(MLEN, len); 1259 m->m_next = n; 1260 } 1261 m = m->m_next; 1262 } 1263out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1264 m->m_pkthdr.len = totlen; 1265} 1266 1267/* 1268 * Append the specified data to the indicated mbuf chain, 1269 * Extend the mbuf chain if the new data does not fit in 1270 * existing space. 1271 * 1272 * Return 1 if able to complete the job; otherwise 0. 1273 */ 1274int 1275m_append(struct mbuf *m0, int len, c_caddr_t cp) 1276{ 1277 struct mbuf *m, *n; 1278 int remainder, space; 1279 1280 for (m = m0; m->m_next != NULL; m = m->m_next) 1281 ; 1282 remainder = len; 1283 space = M_TRAILINGSPACE(m); 1284 if (space > 0) { 1285 /* 1286 * Copy into available space. 1287 */ 1288 if (space > remainder) 1289 space = remainder; 1290 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 1291 m->m_len += space; 1292 cp += space, remainder -= space; 1293 } 1294 while (remainder > 0) { 1295 /* 1296 * Allocate a new mbuf; could check space 1297 * and allocate a cluster instead. 1298 */ 1299 n = m_get(M_DONTWAIT, m->m_type); 1300 if (n == NULL) 1301 break; 1302 n->m_len = min(MLEN, remainder); 1303 bcopy(cp, mtod(n, caddr_t), n->m_len); 1304 cp += n->m_len, remainder -= n->m_len; 1305 m->m_next = n; 1306 m = n; 1307 } 1308 if (m0->m_flags & M_PKTHDR) 1309 m0->m_pkthdr.len += len - remainder; 1310 return (remainder == 0); 1311} 1312 1313/* 1314 * Apply function f to the data in an mbuf chain starting "off" bytes from 1315 * the beginning, continuing for "len" bytes. 1316 */ 1317int 1318m_apply(struct mbuf *m, int off, int len, 1319 int (*f)(void *, void *, u_int), void *arg) 1320{ 1321 u_int count; 1322 int rval; 1323 1324 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1325 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1326 while (off > 0) { 1327 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1328 if (off < m->m_len) 1329 break; 1330 off -= m->m_len; 1331 m = m->m_next; 1332 } 1333 while (len > 0) { 1334 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1335 count = min(m->m_len - off, len); 1336 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1337 if (rval) 1338 return (rval); 1339 len -= count; 1340 off = 0; 1341 m = m->m_next; 1342 } 1343 return (0); 1344} 1345 1346/* 1347 * Return a pointer to mbuf/offset of location in mbuf chain. 1348 */ 1349struct mbuf * 1350m_getptr(struct mbuf *m, int loc, int *off) 1351{ 1352 1353 while (loc >= 0) { 1354 /* Normal end of search. */ 1355 if (m->m_len > loc) { 1356 *off = loc; 1357 return (m); 1358 } else { 1359 loc -= m->m_len; 1360 if (m->m_next == NULL) { 1361 if (loc == 0) { 1362 /* Point at the end of valid data. */ 1363 *off = m->m_len; 1364 return (m); 1365 } 1366 return (NULL); 1367 } 1368 m = m->m_next; 1369 } 1370 } 1371 return (NULL); 1372} 1373 1374void 1375m_print(const struct mbuf *m, int maxlen) 1376{ 1377 int len; 1378 int pdata; 1379 const struct mbuf *m2; 1380 1381 if (m->m_flags & M_PKTHDR) 1382 len = m->m_pkthdr.len; 1383 else 1384 len = -1; 1385 m2 = m; 1386 while (m2 != NULL && (len == -1 || len)) { 1387 pdata = m2->m_len; 1388 if (maxlen != -1 && pdata > maxlen) 1389 pdata = maxlen; 1390 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1391 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1392 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1393 "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1394 if (pdata) 1395 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-"); 1396 if (len != -1) 1397 len -= m2->m_len; 1398 m2 = m2->m_next; 1399 } 1400 if (len > 0) 1401 printf("%d bytes unaccounted for.\n", len); 1402 return; 1403} 1404 1405u_int 1406m_fixhdr(struct mbuf *m0) 1407{ 1408 u_int len; 1409 1410 len = m_length(m0, NULL); 1411 m0->m_pkthdr.len = len; 1412 return (len); 1413} 1414 1415u_int 1416m_length(struct mbuf *m0, struct mbuf **last) 1417{ 1418 struct mbuf *m; 1419 u_int len; 1420 1421 len = 0; 1422 for (m = m0; m != NULL; m = m->m_next) { 1423 len += m->m_len; 1424 if (m->m_next == NULL) 1425 break; 1426 } 1427 if (last != NULL) 1428 *last = m; 1429 return (len); 1430} 1431 1432/* 1433 * Defragment a mbuf chain, returning the shortest possible 1434 * chain of mbufs and clusters. If allocation fails and 1435 * this cannot be completed, NULL will be returned, but 1436 * the passed in chain will be unchanged. Upon success, 1437 * the original chain will be freed, and the new chain 1438 * will be returned. 1439 * 1440 * If a non-packet header is passed in, the original 1441 * mbuf (chain?) will be returned unharmed. 1442 */ 1443struct mbuf * 1444m_defrag(struct mbuf *m0, int how) 1445{ 1446 struct mbuf *m_new = NULL, *m_final = NULL; 1447 int progress = 0, length; 1448 1449 MBUF_CHECKSLEEP(how); 1450 if (!(m0->m_flags & M_PKTHDR)) 1451 return (m0); 1452 1453 m_fixhdr(m0); /* Needed sanity check */ 1454 1455#ifdef MBUF_STRESS_TEST 1456 if (m_defragrandomfailures) { 1457 int temp = arc4random() & 0xff; 1458 if (temp == 0xba) 1459 goto nospace; 1460 } 1461#endif 1462 1463 if (m0->m_pkthdr.len > MHLEN) 1464 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1465 else 1466 m_final = m_gethdr(how, MT_DATA); 1467 1468 if (m_final == NULL) 1469 goto nospace; 1470 1471 if (m_dup_pkthdr(m_final, m0, how) == 0) 1472 goto nospace; 1473 1474 m_new = m_final; 1475 1476 while (progress < m0->m_pkthdr.len) { 1477 length = m0->m_pkthdr.len - progress; 1478 if (length > MCLBYTES) 1479 length = MCLBYTES; 1480 1481 if (m_new == NULL) { 1482 if (length > MLEN) 1483 m_new = m_getcl(how, MT_DATA, 0); 1484 else 1485 m_new = m_get(how, MT_DATA); 1486 if (m_new == NULL) 1487 goto nospace; 1488 } 1489 1490 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1491 progress += length; 1492 m_new->m_len = length; 1493 if (m_new != m_final) 1494 m_cat(m_final, m_new); 1495 m_new = NULL; 1496 } 1497#ifdef MBUF_STRESS_TEST 1498 if (m0->m_next == NULL) 1499 m_defraguseless++; 1500#endif 1501 m_freem(m0); 1502 m0 = m_final; 1503#ifdef MBUF_STRESS_TEST 1504 m_defragpackets++; 1505 m_defragbytes += m0->m_pkthdr.len; 1506#endif 1507 return (m0); 1508nospace: 1509#ifdef MBUF_STRESS_TEST 1510 m_defragfailure++; 1511#endif 1512 if (m_final) 1513 m_freem(m_final); 1514 return (NULL); 1515} 1516 1517#ifdef MBUF_STRESS_TEST 1518 1519/* 1520 * Fragment an mbuf chain. There's no reason you'd ever want to do 1521 * this in normal usage, but it's great for stress testing various 1522 * mbuf consumers. 1523 * 1524 * If fragmentation is not possible, the original chain will be 1525 * returned. 1526 * 1527 * Possible length values: 1528 * 0 no fragmentation will occur 1529 * > 0 each fragment will be of the specified length 1530 * -1 each fragment will be the same random value in length 1531 * -2 each fragment's length will be entirely random 1532 * (Random values range from 1 to 256) 1533 */ 1534struct mbuf * 1535m_fragment(struct mbuf *m0, int how, int length) 1536{ 1537 struct mbuf *m_new = NULL, *m_final = NULL; 1538 int progress = 0; 1539 1540 if (!(m0->m_flags & M_PKTHDR)) 1541 return (m0); 1542 1543 if ((length == 0) || (length < -2)) 1544 return (m0); 1545 1546 m_fixhdr(m0); /* Needed sanity check */ 1547 1548 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1549 1550 if (m_final == NULL) 1551 goto nospace; 1552 1553 if (m_dup_pkthdr(m_final, m0, how) == 0) 1554 goto nospace; 1555 1556 m_new = m_final; 1557 1558 if (length == -1) 1559 length = 1 + (arc4random() & 255); 1560 1561 while (progress < m0->m_pkthdr.len) { 1562 int fraglen; 1563 1564 if (length > 0) 1565 fraglen = length; 1566 else 1567 fraglen = 1 + (arc4random() & 255); 1568 if (fraglen > m0->m_pkthdr.len - progress) 1569 fraglen = m0->m_pkthdr.len - progress; 1570 1571 if (fraglen > MCLBYTES) 1572 fraglen = MCLBYTES; 1573 1574 if (m_new == NULL) { 1575 m_new = m_getcl(how, MT_DATA, 0); 1576 if (m_new == NULL) 1577 goto nospace; 1578 } 1579 1580 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1581 progress += fraglen; 1582 m_new->m_len = fraglen; 1583 if (m_new != m_final) 1584 m_cat(m_final, m_new); 1585 m_new = NULL; 1586 } 1587 m_freem(m0); 1588 m0 = m_final; 1589 return (m0); 1590nospace: 1591 if (m_final) 1592 m_freem(m_final); 1593 /* Return the original chain on failure */ 1594 return (m0); 1595} 1596 1597#endif 1598 1599struct mbuf * 1600m_uiotombuf(struct uio *uio, int how, int len, int align) 1601{ 1602 struct mbuf *m_new = NULL, *m_final = NULL; 1603 int progress = 0, error = 0, length, total; 1604 1605 if (len > 0) 1606 total = min(uio->uio_resid, len); 1607 else 1608 total = uio->uio_resid; 1609 if (align >= MHLEN) 1610 goto nospace; 1611 if (total + align > MHLEN) 1612 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1613 else 1614 m_final = m_gethdr(how, MT_DATA); 1615 if (m_final == NULL) 1616 goto nospace; 1617 m_final->m_data += align; 1618 m_new = m_final; 1619 while (progress < total) { 1620 length = total - progress; 1621 if (length > MCLBYTES) 1622 length = MCLBYTES; 1623 if (m_new == NULL) { 1624 if (length > MLEN) 1625 m_new = m_getcl(how, MT_DATA, 0); 1626 else 1627 m_new = m_get(how, MT_DATA); 1628 if (m_new == NULL) 1629 goto nospace; 1630 } 1631 error = uiomove(mtod(m_new, void *), length, uio); 1632 if (error) 1633 goto nospace; 1634 progress += length; 1635 m_new->m_len = length; 1636 if (m_new != m_final) 1637 m_cat(m_final, m_new); 1638 m_new = NULL; 1639 } 1640 m_fixhdr(m_final); 1641 return (m_final); 1642nospace: 1643 if (m_new) 1644 m_free(m_new); 1645 if (m_final) 1646 m_freem(m_final); 1647 return (NULL); 1648} 1649 1650/* 1651 * Set the m_data pointer of a newly-allocated mbuf 1652 * to place an object of the specified size at the 1653 * end of the mbuf, longword aligned. 1654 */ 1655void 1656m_align(struct mbuf *m, int len) 1657{ 1658 int adjust; 1659 1660 if (m->m_flags & M_EXT) 1661 adjust = m->m_ext.ext_size - len; 1662 else if (m->m_flags & M_PKTHDR) 1663 adjust = MHLEN - len; 1664 else 1665 adjust = MLEN - len; 1666 m->m_data += adjust &~ (sizeof(long)-1); 1667} 1668