uipc_mbuf.c revision 143188
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 143188 2005-03-06 20:09:00Z alc $"); 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 * "Move" mbuf pkthdr from "from" to "to". 274 * "from" must have M_PKTHDR set, and "to" must be empty. 275 */ 276void 277m_move_pkthdr(struct mbuf *to, struct mbuf *from) 278{ 279 280#if 0 281 /* see below for why these are not enabled */ 282 M_ASSERTPKTHDR(to); 283 /* Note: with MAC, this may not be a good assertion. */ 284 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 285 ("m_move_pkthdr: to has tags")); 286#endif 287 KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster")); 288#ifdef MAC 289 /* 290 * XXXMAC: It could be this should also occur for non-MAC? 291 */ 292 if (to->m_flags & M_PKTHDR) 293 m_tag_delete_chain(to, NULL); 294#endif 295 to->m_flags = from->m_flags & M_COPYFLAGS; 296 to->m_data = to->m_pktdat; 297 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 298 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 299 from->m_flags &= ~M_PKTHDR; 300} 301 302/* 303 * Duplicate "from"'s mbuf pkthdr in "to". 304 * "from" must have M_PKTHDR set, and "to" must be empty. 305 * In particular, this does a deep copy of the packet tags. 306 */ 307int 308m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 309{ 310 311#if 0 312 /* 313 * The mbuf allocator only initializes the pkthdr 314 * when the mbuf is allocated with MGETHDR. Many users 315 * (e.g. m_copy*, m_prepend) use MGET and then 316 * smash the pkthdr as needed causing these 317 * assertions to trip. For now just disable them. 318 */ 319 M_ASSERTPKTHDR(to); 320 /* Note: with MAC, this may not be a good assertion. */ 321 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 322#endif 323 MBUF_CHECKSLEEP(how); 324#ifdef MAC 325 if (to->m_flags & M_PKTHDR) 326 m_tag_delete_chain(to, NULL); 327#endif 328 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 329 if ((to->m_flags & M_EXT) == 0) 330 to->m_data = to->m_pktdat; 331 to->m_pkthdr = from->m_pkthdr; 332 SLIST_INIT(&to->m_pkthdr.tags); 333 return (m_tag_copy_chain(to, from, MBTOM(how))); 334} 335 336/* 337 * Lesser-used path for M_PREPEND: 338 * allocate new mbuf to prepend to chain, 339 * copy junk along. 340 */ 341struct mbuf * 342m_prepend(struct mbuf *m, int len, int how) 343{ 344 struct mbuf *mn; 345 346 if (m->m_flags & M_PKTHDR) 347 MGETHDR(mn, how, m->m_type); 348 else 349 MGET(mn, how, m->m_type); 350 if (mn == NULL) { 351 m_freem(m); 352 return (NULL); 353 } 354 if (m->m_flags & M_PKTHDR) 355 M_MOVE_PKTHDR(mn, m); 356 mn->m_next = m; 357 m = mn; 358 if (len < MHLEN) 359 MH_ALIGN(m, len); 360 m->m_len = len; 361 return (m); 362} 363 364/* 365 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 366 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 367 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller. 368 * Note that the copy is read-only, because clusters are not copied, 369 * only their reference counts are incremented. 370 */ 371struct mbuf * 372m_copym(struct mbuf *m, int off0, int len, int wait) 373{ 374 struct mbuf *n, **np; 375 int off = off0; 376 struct mbuf *top; 377 int copyhdr = 0; 378 379 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 380 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 381 MBUF_CHECKSLEEP(wait); 382 if (off == 0 && m->m_flags & M_PKTHDR) 383 copyhdr = 1; 384 while (off > 0) { 385 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 386 if (off < m->m_len) 387 break; 388 off -= m->m_len; 389 m = m->m_next; 390 } 391 np = ⊤ 392 top = 0; 393 while (len > 0) { 394 if (m == NULL) { 395 KASSERT(len == M_COPYALL, 396 ("m_copym, length > size of mbuf chain")); 397 break; 398 } 399 if (copyhdr) 400 MGETHDR(n, wait, m->m_type); 401 else 402 MGET(n, wait, m->m_type); 403 *np = n; 404 if (n == NULL) 405 goto nospace; 406 if (copyhdr) { 407 if (!m_dup_pkthdr(n, m, wait)) 408 goto nospace; 409 if (len == M_COPYALL) 410 n->m_pkthdr.len -= off0; 411 else 412 n->m_pkthdr.len = len; 413 copyhdr = 0; 414 } 415 n->m_len = min(len, m->m_len - off); 416 if (m->m_flags & M_EXT) { 417 n->m_data = m->m_data + off; 418 n->m_ext = m->m_ext; 419 n->m_flags |= M_EXT; 420 MEXT_ADD_REF(m); 421 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 422 } else 423 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 424 (u_int)n->m_len); 425 if (len != M_COPYALL) 426 len -= n->m_len; 427 off = 0; 428 m = m->m_next; 429 np = &n->m_next; 430 } 431 if (top == NULL) 432 mbstat.m_mcfail++; /* XXX: No consistency. */ 433 434 return (top); 435nospace: 436 m_freem(top); 437 mbstat.m_mcfail++; /* XXX: No consistency. */ 438 return (NULL); 439} 440 441/* 442 * Copy an entire packet, including header (which must be present). 443 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 444 * Note that the copy is read-only, because clusters are not copied, 445 * only their reference counts are incremented. 446 * Preserve alignment of the first mbuf so if the creator has left 447 * some room at the beginning (e.g. for inserting protocol headers) 448 * the copies still have the room available. 449 */ 450struct mbuf * 451m_copypacket(struct mbuf *m, int how) 452{ 453 struct mbuf *top, *n, *o; 454 455 MBUF_CHECKSLEEP(how); 456 MGET(n, how, m->m_type); 457 top = n; 458 if (n == NULL) 459 goto nospace; 460 461 if (!m_dup_pkthdr(n, m, how)) 462 goto nospace; 463 n->m_len = m->m_len; 464 if (m->m_flags & M_EXT) { 465 n->m_data = m->m_data; 466 n->m_ext = m->m_ext; 467 n->m_flags |= M_EXT; 468 MEXT_ADD_REF(m); 469 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 470 } else { 471 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 472 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 473 } 474 475 m = m->m_next; 476 while (m) { 477 MGET(o, how, m->m_type); 478 if (o == NULL) 479 goto nospace; 480 481 n->m_next = o; 482 n = n->m_next; 483 484 n->m_len = m->m_len; 485 if (m->m_flags & M_EXT) { 486 n->m_data = m->m_data; 487 n->m_ext = m->m_ext; 488 n->m_flags |= M_EXT; 489 MEXT_ADD_REF(m); 490 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 491 } else { 492 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 493 } 494 495 m = m->m_next; 496 } 497 return top; 498nospace: 499 m_freem(top); 500 mbstat.m_mcfail++; /* XXX: No consistency. */ 501 return (NULL); 502} 503 504/* 505 * Copy data from an mbuf chain starting "off" bytes from the beginning, 506 * continuing for "len" bytes, into the indicated buffer. 507 */ 508void 509m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 510{ 511 u_int count; 512 513 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 514 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 515 while (off > 0) { 516 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 517 if (off < m->m_len) 518 break; 519 off -= m->m_len; 520 m = m->m_next; 521 } 522 while (len > 0) { 523 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 524 count = min(m->m_len - off, len); 525 bcopy(mtod(m, caddr_t) + off, cp, count); 526 len -= count; 527 cp += count; 528 off = 0; 529 m = m->m_next; 530 } 531} 532 533/* 534 * Copy a packet header mbuf chain into a completely new chain, including 535 * copying any mbuf clusters. Use this instead of m_copypacket() when 536 * you need a writable copy of an mbuf chain. 537 */ 538struct mbuf * 539m_dup(struct mbuf *m, int how) 540{ 541 struct mbuf **p, *top = NULL; 542 int remain, moff, nsize; 543 544 MBUF_CHECKSLEEP(how); 545 /* Sanity check */ 546 if (m == NULL) 547 return (NULL); 548 M_ASSERTPKTHDR(m); 549 550 /* While there's more data, get a new mbuf, tack it on, and fill it */ 551 remain = m->m_pkthdr.len; 552 moff = 0; 553 p = ⊤ 554 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 555 struct mbuf *n; 556 557 /* Get the next new mbuf */ 558 if (remain >= MINCLSIZE) { 559 n = m_getcl(how, m->m_type, 0); 560 nsize = MCLBYTES; 561 } else { 562 n = m_get(how, m->m_type); 563 nsize = MLEN; 564 } 565 if (n == NULL) 566 goto nospace; 567 568 if (top == NULL) { /* First one, must be PKTHDR */ 569 if (!m_dup_pkthdr(n, m, how)) { 570 m_free(n); 571 goto nospace; 572 } 573 nsize = MHLEN; 574 } 575 n->m_len = 0; 576 577 /* Link it into the new chain */ 578 *p = n; 579 p = &n->m_next; 580 581 /* Copy data from original mbuf(s) into new mbuf */ 582 while (n->m_len < nsize && m != NULL) { 583 int chunk = min(nsize - n->m_len, m->m_len - moff); 584 585 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 586 moff += chunk; 587 n->m_len += chunk; 588 remain -= chunk; 589 if (moff == m->m_len) { 590 m = m->m_next; 591 moff = 0; 592 } 593 } 594 595 /* Check correct total mbuf length */ 596 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 597 ("%s: bogus m_pkthdr.len", __func__)); 598 } 599 return (top); 600 601nospace: 602 m_freem(top); 603 mbstat.m_mcfail++; /* XXX: No consistency. */ 604 return (NULL); 605} 606 607/* 608 * Concatenate mbuf chain n to m. 609 * Both chains must be of the same type (e.g. MT_DATA). 610 * Any m_pkthdr is not updated. 611 */ 612void 613m_cat(struct mbuf *m, struct mbuf *n) 614{ 615 while (m->m_next) 616 m = m->m_next; 617 while (n) { 618 if (m->m_flags & M_EXT || 619 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 620 /* just join the two chains */ 621 m->m_next = n; 622 return; 623 } 624 /* splat the data from one into the other */ 625 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 626 (u_int)n->m_len); 627 m->m_len += n->m_len; 628 n = m_free(n); 629 } 630} 631 632void 633m_adj(struct mbuf *mp, int req_len) 634{ 635 int len = req_len; 636 struct mbuf *m; 637 int count; 638 639 if ((m = mp) == NULL) 640 return; 641 if (len >= 0) { 642 /* 643 * Trim from head. 644 */ 645 while (m != NULL && len > 0) { 646 if (m->m_len <= len) { 647 len -= m->m_len; 648 m->m_len = 0; 649 m = m->m_next; 650 } else { 651 m->m_len -= len; 652 m->m_data += len; 653 len = 0; 654 } 655 } 656 m = mp; 657 if (mp->m_flags & M_PKTHDR) 658 m->m_pkthdr.len -= (req_len - len); 659 } else { 660 /* 661 * Trim from tail. Scan the mbuf chain, 662 * calculating its length and finding the last mbuf. 663 * If the adjustment only affects this mbuf, then just 664 * adjust and return. Otherwise, rescan and truncate 665 * after the remaining size. 666 */ 667 len = -len; 668 count = 0; 669 for (;;) { 670 count += m->m_len; 671 if (m->m_next == (struct mbuf *)0) 672 break; 673 m = m->m_next; 674 } 675 if (m->m_len >= len) { 676 m->m_len -= len; 677 if (mp->m_flags & M_PKTHDR) 678 mp->m_pkthdr.len -= len; 679 return; 680 } 681 count -= len; 682 if (count < 0) 683 count = 0; 684 /* 685 * Correct length for chain is "count". 686 * Find the mbuf with last data, adjust its length, 687 * and toss data from remaining mbufs on chain. 688 */ 689 m = mp; 690 if (m->m_flags & M_PKTHDR) 691 m->m_pkthdr.len = count; 692 for (; m; m = m->m_next) { 693 if (m->m_len >= count) { 694 m->m_len = count; 695 if (m->m_next != NULL) { 696 m_freem(m->m_next); 697 m->m_next = NULL; 698 } 699 break; 700 } 701 count -= m->m_len; 702 } 703 } 704} 705 706/* 707 * Rearange an mbuf chain so that len bytes are contiguous 708 * and in the data area of an mbuf (so that mtod and dtom 709 * will work for a structure of size len). Returns the resulting 710 * mbuf chain on success, frees it and returns null on failure. 711 * If there is room, it will add up to max_protohdr-len extra bytes to the 712 * contiguous region in an attempt to avoid being called next time. 713 */ 714struct mbuf * 715m_pullup(struct mbuf *n, int len) 716{ 717 struct mbuf *m; 718 int count; 719 int space; 720 721 /* 722 * If first mbuf has no cluster, and has room for len bytes 723 * without shifting current data, pullup into it, 724 * otherwise allocate a new mbuf to prepend to the chain. 725 */ 726 if ((n->m_flags & M_EXT) == 0 && 727 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 728 if (n->m_len >= len) 729 return (n); 730 m = n; 731 n = n->m_next; 732 len -= m->m_len; 733 } else { 734 if (len > MHLEN) 735 goto bad; 736 MGET(m, M_DONTWAIT, n->m_type); 737 if (m == NULL) 738 goto bad; 739 m->m_len = 0; 740 if (n->m_flags & M_PKTHDR) 741 M_MOVE_PKTHDR(m, n); 742 } 743 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 744 do { 745 count = min(min(max(len, max_protohdr), space), n->m_len); 746 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 747 (u_int)count); 748 len -= count; 749 m->m_len += count; 750 n->m_len -= count; 751 space -= count; 752 if (n->m_len) 753 n->m_data += count; 754 else 755 n = m_free(n); 756 } while (len > 0 && n); 757 if (len > 0) { 758 (void) m_free(m); 759 goto bad; 760 } 761 m->m_next = n; 762 return (m); 763bad: 764 m_freem(n); 765 mbstat.m_mpfail++; /* XXX: No consistency. */ 766 return (NULL); 767} 768 769/* 770 * Partition an mbuf chain in two pieces, returning the tail -- 771 * all but the first len0 bytes. In case of failure, it returns NULL and 772 * attempts to restore the chain to its original state. 773 * 774 * Note that the resulting mbufs might be read-only, because the new 775 * mbuf can end up sharing an mbuf cluster with the original mbuf if 776 * the "breaking point" happens to lie within a cluster mbuf. Use the 777 * M_WRITABLE() macro to check for this case. 778 */ 779struct mbuf * 780m_split(struct mbuf *m0, int len0, int wait) 781{ 782 struct mbuf *m, *n; 783 u_int len = len0, remain; 784 785 MBUF_CHECKSLEEP(wait); 786 for (m = m0; m && len > m->m_len; m = m->m_next) 787 len -= m->m_len; 788 if (m == NULL) 789 return (NULL); 790 remain = m->m_len - len; 791 if (m0->m_flags & M_PKTHDR) { 792 MGETHDR(n, wait, m0->m_type); 793 if (n == NULL) 794 return (NULL); 795 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 796 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 797 m0->m_pkthdr.len = len0; 798 if (m->m_flags & M_EXT) 799 goto extpacket; 800 if (remain > MHLEN) { 801 /* m can't be the lead packet */ 802 MH_ALIGN(n, 0); 803 n->m_next = m_split(m, len, wait); 804 if (n->m_next == NULL) { 805 (void) m_free(n); 806 return (NULL); 807 } else { 808 n->m_len = 0; 809 return (n); 810 } 811 } else 812 MH_ALIGN(n, remain); 813 } else if (remain == 0) { 814 n = m->m_next; 815 m->m_next = NULL; 816 return (n); 817 } else { 818 MGET(n, wait, m->m_type); 819 if (n == NULL) 820 return (NULL); 821 M_ALIGN(n, remain); 822 } 823extpacket: 824 if (m->m_flags & M_EXT) { 825 n->m_flags |= M_EXT; 826 n->m_ext = m->m_ext; 827 MEXT_ADD_REF(m); 828 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 829 n->m_data = m->m_data + len; 830 } else { 831 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 832 } 833 n->m_len = remain; 834 m->m_len = len; 835 n->m_next = m->m_next; 836 m->m_next = NULL; 837 return (n); 838} 839/* 840 * Routine to copy from device local memory into mbufs. 841 * Note that `off' argument is offset into first mbuf of target chain from 842 * which to begin copying the data to. 843 */ 844struct mbuf * 845m_devget(char *buf, int totlen, int off, struct ifnet *ifp, 846 void (*copy)(char *from, caddr_t to, u_int len)) 847{ 848 struct mbuf *m; 849 struct mbuf *top = NULL, **mp = ⊤ 850 int len; 851 852 if (off < 0 || off > MHLEN) 853 return (NULL); 854 855 while (totlen > 0) { 856 if (top == NULL) { /* First one, must be PKTHDR */ 857 if (totlen + off >= MINCLSIZE) { 858 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 859 len = MCLBYTES; 860 } else { 861 m = m_gethdr(M_DONTWAIT, MT_DATA); 862 len = MHLEN; 863 864 /* Place initial small packet/header at end of mbuf */ 865 if (m && totlen + off + max_linkhdr <= MLEN) { 866 m->m_data += max_linkhdr; 867 len -= max_linkhdr; 868 } 869 } 870 if (m == NULL) 871 return NULL; 872 m->m_pkthdr.rcvif = ifp; 873 m->m_pkthdr.len = totlen; 874 } else { 875 if (totlen + off >= MINCLSIZE) { 876 m = m_getcl(M_DONTWAIT, MT_DATA, 0); 877 len = MCLBYTES; 878 } else { 879 m = m_get(M_DONTWAIT, MT_DATA); 880 len = MLEN; 881 } 882 if (m == NULL) { 883 m_freem(top); 884 return NULL; 885 } 886 } 887 if (off) { 888 m->m_data += off; 889 len -= off; 890 off = 0; 891 } 892 m->m_len = len = min(totlen, len); 893 if (copy) 894 copy(buf, mtod(m, caddr_t), (u_int)len); 895 else 896 bcopy(buf, mtod(m, caddr_t), (u_int)len); 897 buf += len; 898 *mp = m; 899 mp = &m->m_next; 900 totlen -= len; 901 } 902 return (top); 903} 904 905/* 906 * Copy data from a buffer back into the indicated mbuf chain, 907 * starting "off" bytes from the beginning, extending the mbuf 908 * chain if necessary. 909 */ 910void 911m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 912{ 913 int mlen; 914 struct mbuf *m = m0, *n; 915 int totlen = 0; 916 917 if (m0 == NULL) 918 return; 919 while (off > (mlen = m->m_len)) { 920 off -= mlen; 921 totlen += mlen; 922 if (m->m_next == NULL) { 923 n = m_get(M_DONTWAIT, m->m_type); 924 if (n == NULL) 925 goto out; 926 bzero(mtod(n, caddr_t), MLEN); 927 n->m_len = min(MLEN, len + off); 928 m->m_next = n; 929 } 930 m = m->m_next; 931 } 932 while (len > 0) { 933 mlen = min (m->m_len - off, len); 934 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 935 cp += mlen; 936 len -= mlen; 937 mlen += off; 938 off = 0; 939 totlen += mlen; 940 if (len == 0) 941 break; 942 if (m->m_next == NULL) { 943 n = m_get(M_DONTWAIT, m->m_type); 944 if (n == NULL) 945 break; 946 n->m_len = min(MLEN, len); 947 m->m_next = n; 948 } 949 m = m->m_next; 950 } 951out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 952 m->m_pkthdr.len = totlen; 953} 954 955/* 956 * Append the specified data to the indicated mbuf chain, 957 * Extend the mbuf chain if the new data does not fit in 958 * existing space. 959 * 960 * Return 1 if able to complete the job; otherwise 0. 961 */ 962int 963m_append(struct mbuf *m0, int len, c_caddr_t cp) 964{ 965 struct mbuf *m, *n; 966 int remainder, space; 967 968 for (m = m0; m->m_next != NULL; m = m->m_next) 969 ; 970 remainder = len; 971 space = M_TRAILINGSPACE(m); 972 if (space > 0) { 973 /* 974 * Copy into available space. 975 */ 976 if (space > remainder) 977 space = remainder; 978 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 979 m->m_len += space; 980 cp += space, remainder -= space; 981 } 982 while (remainder > 0) { 983 /* 984 * Allocate a new mbuf; could check space 985 * and allocate a cluster instead. 986 */ 987 n = m_get(M_DONTWAIT, m->m_type); 988 if (n == NULL) 989 break; 990 n->m_len = min(MLEN, remainder); 991 bcopy(cp, mtod(n, caddr_t), n->m_len); 992 cp += n->m_len, remainder -= n->m_len; 993 m->m_next = n; 994 m = n; 995 } 996 if (m0->m_flags & M_PKTHDR) 997 m0->m_pkthdr.len += len - remainder; 998 return (remainder == 0); 999} 1000 1001/* 1002 * Apply function f to the data in an mbuf chain starting "off" bytes from 1003 * the beginning, continuing for "len" bytes. 1004 */ 1005int 1006m_apply(struct mbuf *m, int off, int len, 1007 int (*f)(void *, void *, u_int), void *arg) 1008{ 1009 u_int count; 1010 int rval; 1011 1012 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1013 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1014 while (off > 0) { 1015 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1016 if (off < m->m_len) 1017 break; 1018 off -= m->m_len; 1019 m = m->m_next; 1020 } 1021 while (len > 0) { 1022 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1023 count = min(m->m_len - off, len); 1024 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1025 if (rval) 1026 return (rval); 1027 len -= count; 1028 off = 0; 1029 m = m->m_next; 1030 } 1031 return (0); 1032} 1033 1034/* 1035 * Return a pointer to mbuf/offset of location in mbuf chain. 1036 */ 1037struct mbuf * 1038m_getptr(struct mbuf *m, int loc, int *off) 1039{ 1040 1041 while (loc >= 0) { 1042 /* Normal end of search. */ 1043 if (m->m_len > loc) { 1044 *off = loc; 1045 return (m); 1046 } else { 1047 loc -= m->m_len; 1048 if (m->m_next == NULL) { 1049 if (loc == 0) { 1050 /* Point at the end of valid data. */ 1051 *off = m->m_len; 1052 return (m); 1053 } 1054 return (NULL); 1055 } 1056 m = m->m_next; 1057 } 1058 } 1059 return (NULL); 1060} 1061 1062void 1063m_print(const struct mbuf *m, int maxlen) 1064{ 1065 int len; 1066 int pdata; 1067 const struct mbuf *m2; 1068 1069 if (m->m_flags & M_PKTHDR) 1070 len = m->m_pkthdr.len; 1071 else 1072 len = -1; 1073 m2 = m; 1074 while (m2 != NULL && (len == -1 || len)) { 1075 pdata = m2->m_len; 1076 if (maxlen != -1 && pdata > maxlen) 1077 pdata = maxlen; 1078 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1079 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1080 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1081 "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1082 if (pdata) 1083 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-"); 1084 if (len != -1) 1085 len -= m2->m_len; 1086 m2 = m2->m_next; 1087 } 1088 if (len > 0) 1089 printf("%d bytes unaccounted for.\n", len); 1090 return; 1091} 1092 1093u_int 1094m_fixhdr(struct mbuf *m0) 1095{ 1096 u_int len; 1097 1098 len = m_length(m0, NULL); 1099 m0->m_pkthdr.len = len; 1100 return (len); 1101} 1102 1103u_int 1104m_length(struct mbuf *m0, struct mbuf **last) 1105{ 1106 struct mbuf *m; 1107 u_int len; 1108 1109 len = 0; 1110 for (m = m0; m != NULL; m = m->m_next) { 1111 len += m->m_len; 1112 if (m->m_next == NULL) 1113 break; 1114 } 1115 if (last != NULL) 1116 *last = m; 1117 return (len); 1118} 1119 1120/* 1121 * Defragment a mbuf chain, returning the shortest possible 1122 * chain of mbufs and clusters. If allocation fails and 1123 * this cannot be completed, NULL will be returned, but 1124 * the passed in chain will be unchanged. Upon success, 1125 * the original chain will be freed, and the new chain 1126 * will be returned. 1127 * 1128 * If a non-packet header is passed in, the original 1129 * mbuf (chain?) will be returned unharmed. 1130 */ 1131struct mbuf * 1132m_defrag(struct mbuf *m0, int how) 1133{ 1134 struct mbuf *m_new = NULL, *m_final = NULL; 1135 int progress = 0, length; 1136 1137 MBUF_CHECKSLEEP(how); 1138 if (!(m0->m_flags & M_PKTHDR)) 1139 return (m0); 1140 1141 m_fixhdr(m0); /* Needed sanity check */ 1142 1143#ifdef MBUF_STRESS_TEST 1144 if (m_defragrandomfailures) { 1145 int temp = arc4random() & 0xff; 1146 if (temp == 0xba) 1147 goto nospace; 1148 } 1149#endif 1150 1151 if (m0->m_pkthdr.len > MHLEN) 1152 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1153 else 1154 m_final = m_gethdr(how, MT_DATA); 1155 1156 if (m_final == NULL) 1157 goto nospace; 1158 1159 if (m_dup_pkthdr(m_final, m0, how) == 0) 1160 goto nospace; 1161 1162 m_new = m_final; 1163 1164 while (progress < m0->m_pkthdr.len) { 1165 length = m0->m_pkthdr.len - progress; 1166 if (length > MCLBYTES) 1167 length = MCLBYTES; 1168 1169 if (m_new == NULL) { 1170 if (length > MLEN) 1171 m_new = m_getcl(how, MT_DATA, 0); 1172 else 1173 m_new = m_get(how, MT_DATA); 1174 if (m_new == NULL) 1175 goto nospace; 1176 } 1177 1178 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1179 progress += length; 1180 m_new->m_len = length; 1181 if (m_new != m_final) 1182 m_cat(m_final, m_new); 1183 m_new = NULL; 1184 } 1185#ifdef MBUF_STRESS_TEST 1186 if (m0->m_next == NULL) 1187 m_defraguseless++; 1188#endif 1189 m_freem(m0); 1190 m0 = m_final; 1191#ifdef MBUF_STRESS_TEST 1192 m_defragpackets++; 1193 m_defragbytes += m0->m_pkthdr.len; 1194#endif 1195 return (m0); 1196nospace: 1197#ifdef MBUF_STRESS_TEST 1198 m_defragfailure++; 1199#endif 1200 if (m_final) 1201 m_freem(m_final); 1202 return (NULL); 1203} 1204 1205#ifdef MBUF_STRESS_TEST 1206 1207/* 1208 * Fragment an mbuf chain. There's no reason you'd ever want to do 1209 * this in normal usage, but it's great for stress testing various 1210 * mbuf consumers. 1211 * 1212 * If fragmentation is not possible, the original chain will be 1213 * returned. 1214 * 1215 * Possible length values: 1216 * 0 no fragmentation will occur 1217 * > 0 each fragment will be of the specified length 1218 * -1 each fragment will be the same random value in length 1219 * -2 each fragment's length will be entirely random 1220 * (Random values range from 1 to 256) 1221 */ 1222struct mbuf * 1223m_fragment(struct mbuf *m0, int how, int length) 1224{ 1225 struct mbuf *m_new = NULL, *m_final = NULL; 1226 int progress = 0; 1227 1228 if (!(m0->m_flags & M_PKTHDR)) 1229 return (m0); 1230 1231 if ((length == 0) || (length < -2)) 1232 return (m0); 1233 1234 m_fixhdr(m0); /* Needed sanity check */ 1235 1236 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1237 1238 if (m_final == NULL) 1239 goto nospace; 1240 1241 if (m_dup_pkthdr(m_final, m0, how) == 0) 1242 goto nospace; 1243 1244 m_new = m_final; 1245 1246 if (length == -1) 1247 length = 1 + (arc4random() & 255); 1248 1249 while (progress < m0->m_pkthdr.len) { 1250 int fraglen; 1251 1252 if (length > 0) 1253 fraglen = length; 1254 else 1255 fraglen = 1 + (arc4random() & 255); 1256 if (fraglen > m0->m_pkthdr.len - progress) 1257 fraglen = m0->m_pkthdr.len - progress; 1258 1259 if (fraglen > MCLBYTES) 1260 fraglen = MCLBYTES; 1261 1262 if (m_new == NULL) { 1263 m_new = m_getcl(how, MT_DATA, 0); 1264 if (m_new == NULL) 1265 goto nospace; 1266 } 1267 1268 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1269 progress += fraglen; 1270 m_new->m_len = fraglen; 1271 if (m_new != m_final) 1272 m_cat(m_final, m_new); 1273 m_new = NULL; 1274 } 1275 m_freem(m0); 1276 m0 = m_final; 1277 return (m0); 1278nospace: 1279 if (m_final) 1280 m_freem(m_final); 1281 /* Return the original chain on failure */ 1282 return (m0); 1283} 1284 1285#endif 1286 1287struct mbuf * 1288m_uiotombuf(struct uio *uio, int how, int len) 1289{ 1290 struct mbuf *m_new = NULL, *m_final = NULL; 1291 int progress = 0, error = 0, length, total; 1292 1293 if (len > 0) 1294 total = min(uio->uio_resid, len); 1295 else 1296 total = uio->uio_resid; 1297 if (total > MHLEN) 1298 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1299 else 1300 m_final = m_gethdr(how, MT_DATA); 1301 if (m_final == NULL) 1302 goto nospace; 1303 m_new = m_final; 1304 while (progress < total) { 1305 length = total - progress; 1306 if (length > MCLBYTES) 1307 length = MCLBYTES; 1308 if (m_new == NULL) { 1309 if (length > MLEN) 1310 m_new = m_getcl(how, MT_DATA, 0); 1311 else 1312 m_new = m_get(how, MT_DATA); 1313 if (m_new == NULL) 1314 goto nospace; 1315 } 1316 error = uiomove(mtod(m_new, void *), length, uio); 1317 if (error) 1318 goto nospace; 1319 progress += length; 1320 m_new->m_len = length; 1321 if (m_new != m_final) 1322 m_cat(m_final, m_new); 1323 m_new = NULL; 1324 } 1325 m_fixhdr(m_final); 1326 return (m_final); 1327nospace: 1328 if (m_new) 1329 m_free(m_new); 1330 if (m_final) 1331 m_freem(m_final); 1332 return (NULL); 1333} 1334