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("$FreeBSD$"); 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 52 int max_linkhdr; 53 int max_protohdr; 54 int max_hdr; 55 int max_datalen; 56 #ifdef MBUF_STRESS_TEST 57 int m_defragpackets; 58 int m_defragbytes; 59 int m_defraguseless; 60 int m_defragfailure; 61 int m_defragrandomfailures; 62 #endif 63 64 /* 65 * sysctl(8) exported objects 66 */ 67 SYSCTL_DECL(_kern_ipc); 68 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 69 &max_linkhdr, 0, ""); 70 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 71 &max_protohdr, 0, ""); 72 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, ""); 73 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 74 &max_datalen, 0, ""); 75 #ifdef MBUF_STRESS_TEST 76 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 77 &m_defragpackets, 0, ""); 78 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 79 &m_defragbytes, 0, ""); 80 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 81 &m_defraguseless, 0, ""); 82 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 83 &m_defragfailure, 0, ""); 84 SYSCTL_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 */ 91 static 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 */ 100 struct mbuf * 101 m_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; 152 failed: 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 */ 162 void 163 m_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 */ 191 void 192 m_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 */ 221 void 222 mb_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 */ 276 void 277 m_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 #ifdef MAC 288 /* 289 * XXXMAC: It could be this should also occur for non-MAC? 290 */ 291 if (to->m_flags & M_PKTHDR) 292 m_tag_delete_chain(to, NULL); 293 #endif 294 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 295 if ((to->m_flags & M_EXT) == 0) 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 */ 307 int 308 m_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 */ 341 struct mbuf * 342 m_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 */ 371 struct mbuf * 372 m_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); 435 nospace: 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 */ 450 struct mbuf * 451 m_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; 498 nospace: 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 */ 508 void 509 m_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 */ 538 struct mbuf * 539 m_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); 600 601 nospace: 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 */ 612 void 613 m_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 } 631 632 void 633 m_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 */ 714 struct mbuf * 715 m_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); 763 bad: 764 m_freem(n); 765 mbstat.m_mpfail++; /* XXX: No consistency. */ 766 return (NULL); 767 } 768 769 /* 770 * Like m_pullup(), except a new mbuf is always allocated, and we allow 771 * the amount of empty space before the data in the new mbuf to be specified 772 * (in the event that the caller expects to prepend later). 773 */ 774 int MSFail; 775 776 struct mbuf * 777 m_copyup(struct mbuf *n, int len, int dstoff) 778 { 779 struct mbuf *m; 780 int count, space; 781 782 if (len > (MHLEN - dstoff)) 783 goto bad; 784 MGET(m, M_DONTWAIT, n->m_type); 785 if (m == NULL) 786 goto bad; 787 m->m_len = 0; 788 if (n->m_flags & M_PKTHDR) 789 M_MOVE_PKTHDR(m, n); 790 m->m_data += dstoff; 791 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 792 do { 793 count = min(min(max(len, max_protohdr), space), n->m_len); 794 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), 795 (unsigned)count); 796 len -= count; 797 m->m_len += count; 798 n->m_len -= count; 799 space -= count; 800 if (n->m_len) 801 n->m_data += count; 802 else 803 n = m_free(n); 804 } while (len > 0 && n); 805 if (len > 0) { 806 (void) m_free(m); 807 goto bad; 808 } 809 m->m_next = n; 810 return (m); 811 bad: 812 m_freem(n); 813 MSFail++; 814 return (NULL); 815 } 816 817 /* 818 * Partition an mbuf chain in two pieces, returning the tail -- 819 * all but the first len0 bytes. In case of failure, it returns NULL and 820 * attempts to restore the chain to its original state. 821 * 822 * Note that the resulting mbufs might be read-only, because the new 823 * mbuf can end up sharing an mbuf cluster with the original mbuf if 824 * the "breaking point" happens to lie within a cluster mbuf. Use the 825 * M_WRITABLE() macro to check for this case. 826 */ 827 struct mbuf * 828 m_split(struct mbuf *m0, int len0, int wait) 829 { 830 struct mbuf *m, *n; 831 u_int len = len0, remain; 832 833 MBUF_CHECKSLEEP(wait); 834 for (m = m0; m && len > m->m_len; m = m->m_next) 835 len -= m->m_len; 836 if (m == NULL) 837 return (NULL); 838 remain = m->m_len - len; 839 if (m0->m_flags & M_PKTHDR) { 840 MGETHDR(n, wait, m0->m_type); 841 if (n == NULL) 842 return (NULL); 843 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 844 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 845 m0->m_pkthdr.len = len0; 846 if (m->m_flags & M_EXT) 847 goto extpacket; 848 if (remain > MHLEN) { 849 /* m can't be the lead packet */ 850 MH_ALIGN(n, 0); 851 n->m_next = m_split(m, len, wait); 852 if (n->m_next == NULL) { 853 (void) m_free(n); 854 return (NULL); 855 } else { 856 n->m_len = 0; 857 return (n); 858 } 859 } else 860 MH_ALIGN(n, remain); 861 } else if (remain == 0) { 862 n = m->m_next; 863 m->m_next = NULL; 864 return (n); 865 } else { 866 MGET(n, wait, m->m_type); 867 if (n == NULL) 868 return (NULL); 869 M_ALIGN(n, remain); 870 } 871 extpacket: 872 if (m->m_flags & M_EXT) { 873 n->m_flags |= M_EXT; 874 n->m_ext = m->m_ext; 875 MEXT_ADD_REF(m); 876 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 877 n->m_data = m->m_data + len; 878 } else { 879 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 880 } 881 n->m_len = remain; 882 m->m_len = len; 883 n->m_next = m->m_next; 884 m->m_next = NULL; 885 return (n); 886 } 887 /* 888 * Routine to copy from device local memory into mbufs. 889 * Note that `off' argument is offset into first mbuf of target chain from 890 * which to begin copying the data to. 891 */ 892 struct mbuf * 893 m_devget(char *buf, int totlen, int off, struct ifnet *ifp, 894 void (*copy)(char *from, caddr_t to, u_int len)) 895 { 896 struct mbuf *m; 897 struct mbuf *top = NULL, **mp = ⊤ 898 int len; 899 900 if (off < 0 || off > MHLEN) 901 return (NULL); 902 903 while (totlen > 0) { 904 if (top == NULL) { /* First one, must be PKTHDR */ 905 if (totlen + off >= MINCLSIZE) { 906 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 907 len = MCLBYTES; 908 } else { 909 m = m_gethdr(M_DONTWAIT, MT_DATA); 910 len = MHLEN; 911 912 /* Place initial small packet/header at end of mbuf */ 913 if (m && totlen + off + max_linkhdr <= MLEN) { 914 m->m_data += max_linkhdr; 915 len -= max_linkhdr; 916 } 917 } 918 if (m == NULL) 919 return NULL; 920 m->m_pkthdr.rcvif = ifp; 921 m->m_pkthdr.len = totlen; 922 } else { 923 if (totlen + off >= MINCLSIZE) { 924 m = m_getcl(M_DONTWAIT, MT_DATA, 0); 925 len = MCLBYTES; 926 } else { 927 m = m_get(M_DONTWAIT, MT_DATA); 928 len = MLEN; 929 } 930 if (m == NULL) { 931 m_freem(top); 932 return NULL; 933 } 934 } 935 if (off) { 936 m->m_data += off; 937 len -= off; 938 off = 0; 939 } 940 m->m_len = len = min(totlen, len); 941 if (copy) 942 copy(buf, mtod(m, caddr_t), (u_int)len); 943 else 944 bcopy(buf, mtod(m, caddr_t), (u_int)len); 945 buf += len; 946 *mp = m; 947 mp = &m->m_next; 948 totlen -= len; 949 } 950 return (top); 951 } 952 953 /* 954 * Copy data from a buffer back into the indicated mbuf chain, 955 * starting "off" bytes from the beginning, extending the mbuf 956 * chain if necessary. 957 */ 958 void 959 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 960 { 961 int mlen; 962 struct mbuf *m = m0, *n; 963 int totlen = 0; 964 965 if (m0 == NULL) 966 return; 967 while (off > (mlen = m->m_len)) { 968 off -= mlen; 969 totlen += mlen; 970 if (m->m_next == NULL) { 971 n = m_get(M_DONTWAIT, m->m_type); 972 if (n == NULL) 973 goto out; 974 bzero(mtod(n, caddr_t), MLEN); 975 n->m_len = min(MLEN, len + off); 976 m->m_next = n; 977 } 978 m = m->m_next; 979 } 980 while (len > 0) { 981 mlen = min (m->m_len - off, len); 982 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 983 cp += mlen; 984 len -= mlen; 985 mlen += off; 986 off = 0; 987 totlen += mlen; 988 if (len == 0) 989 break; 990 if (m->m_next == NULL) { 991 n = m_get(M_DONTWAIT, m->m_type); 992 if (n == NULL) 993 break; 994 n->m_len = min(MLEN, len); 995 m->m_next = n; 996 } 997 m = m->m_next; 998 } 999 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1000 m->m_pkthdr.len = totlen; 1001 } 1002 1003 /* 1004 * Append the specified data to the indicated mbuf chain, 1005 * Extend the mbuf chain if the new data does not fit in 1006 * existing space. 1007 * 1008 * Return 1 if able to complete the job; otherwise 0. 1009 */ 1010 int 1011 m_append(struct mbuf *m0, int len, c_caddr_t cp) 1012 { 1013 struct mbuf *m, *n; 1014 int remainder, space; 1015 1016 for (m = m0; m->m_next != NULL; m = m->m_next) 1017 ; 1018 remainder = len; 1019 space = M_TRAILINGSPACE(m); 1020 if (space > 0) { 1021 /* 1022 * Copy into available space. 1023 */ 1024 if (space > remainder) 1025 space = remainder; 1026 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 1027 m->m_len += space; 1028 cp += space, remainder -= space; 1029 } 1030 while (remainder > 0) { 1031 /* 1032 * Allocate a new mbuf; could check space 1033 * and allocate a cluster instead. 1034 */ 1035 n = m_get(M_DONTWAIT, m->m_type); 1036 if (n == NULL) 1037 break; 1038 n->m_len = min(MLEN, remainder); 1039 bcopy(cp, mtod(n, caddr_t), n->m_len); 1040 cp += n->m_len, remainder -= n->m_len; 1041 m->m_next = n; 1042 m = n; 1043 } 1044 if (m0->m_flags & M_PKTHDR) 1045 m0->m_pkthdr.len += len - remainder; 1046 return (remainder == 0); 1047 } 1048 1049 /* 1050 * Apply function f to the data in an mbuf chain starting "off" bytes from 1051 * the beginning, continuing for "len" bytes. 1052 */ 1053 int 1054 m_apply(struct mbuf *m, int off, int len, 1055 int (*f)(void *, void *, u_int), void *arg) 1056 { 1057 u_int count; 1058 int rval; 1059 1060 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1061 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1062 while (off > 0) { 1063 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1064 if (off < m->m_len) 1065 break; 1066 off -= m->m_len; 1067 m = m->m_next; 1068 } 1069 while (len > 0) { 1070 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1071 count = min(m->m_len - off, len); 1072 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1073 if (rval) 1074 return (rval); 1075 len -= count; 1076 off = 0; 1077 m = m->m_next; 1078 } 1079 return (0); 1080 } 1081 1082 /* 1083 * Return a pointer to mbuf/offset of location in mbuf chain. 1084 */ 1085 struct mbuf * 1086 m_getptr(struct mbuf *m, int loc, int *off) 1087 { 1088 1089 while (loc >= 0) { 1090 /* Normal end of search. */ 1091 if (m->m_len > loc) { 1092 *off = loc; 1093 return (m); 1094 } else { 1095 loc -= m->m_len; 1096 if (m->m_next == NULL) { 1097 if (loc == 0) { 1098 /* Point at the end of valid data. */ 1099 *off = m->m_len; 1100 return (m); 1101 } 1102 return (NULL); 1103 } 1104 m = m->m_next; 1105 } 1106 } 1107 return (NULL); 1108 } 1109 1110 void 1111 m_print(const struct mbuf *m, int maxlen) 1112 { 1113 int len; 1114 int pdata; 1115 const struct mbuf *m2; 1116 1117 if (m->m_flags & M_PKTHDR) 1118 len = m->m_pkthdr.len; 1119 else 1120 len = -1; 1121 m2 = m; 1122 while (m2 != NULL && (len == -1 || len)) { 1123 pdata = m2->m_len; 1124 if (maxlen != -1 && pdata > maxlen) 1125 pdata = maxlen; 1126 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1127 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1128 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1129 "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1130 if (pdata) 1131 printf(", %*D\n", m2->m_len, (u_char *)m2->m_data, "-"); 1132 if (len != -1) 1133 len -= m2->m_len; 1134 m2 = m2->m_next; 1135 } 1136 if (len > 0) 1137 printf("%d bytes unaccounted for.\n", len); 1138 return; 1139 } 1140 1141 u_int 1142 m_fixhdr(struct mbuf *m0) 1143 { 1144 u_int len; 1145 1146 len = m_length(m0, NULL); 1147 m0->m_pkthdr.len = len; 1148 return (len); 1149 } 1150 1151 u_int 1152 m_length(struct mbuf *m0, struct mbuf **last) 1153 { 1154 struct mbuf *m; 1155 u_int len; 1156 1157 len = 0; 1158 for (m = m0; m != NULL; m = m->m_next) { 1159 len += m->m_len; 1160 if (m->m_next == NULL) 1161 break; 1162 } 1163 if (last != NULL) 1164 *last = m; 1165 return (len); 1166 } 1167 1168 /* 1169 * Defragment a mbuf chain, returning the shortest possible 1170 * chain of mbufs and clusters. If allocation fails and 1171 * this cannot be completed, NULL will be returned, but 1172 * the passed in chain will be unchanged. Upon success, 1173 * the original chain will be freed, and the new chain 1174 * will be returned. 1175 * 1176 * If a non-packet header is passed in, the original 1177 * mbuf (chain?) will be returned unharmed. 1178 */ 1179 struct mbuf * 1180 m_defrag(struct mbuf *m0, int how) 1181 { 1182 struct mbuf *m_new = NULL, *m_final = NULL; 1183 int progress = 0, length; 1184 1185 MBUF_CHECKSLEEP(how); 1186 if (!(m0->m_flags & M_PKTHDR)) 1187 return (m0); 1188 1189 m_fixhdr(m0); /* Needed sanity check */ 1190 1191 #ifdef MBUF_STRESS_TEST 1192 if (m_defragrandomfailures) { 1193 int temp = arc4random() & 0xff; 1194 if (temp == 0xba) 1195 goto nospace; 1196 } 1197 #endif 1198 1199 if (m0->m_pkthdr.len > MHLEN) 1200 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1201 else 1202 m_final = m_gethdr(how, MT_DATA); 1203 1204 if (m_final == NULL) 1205 goto nospace; 1206 1207 if (m_dup_pkthdr(m_final, m0, how) == 0) 1208 goto nospace; 1209 1210 m_new = m_final; 1211 1212 while (progress < m0->m_pkthdr.len) { 1213 length = m0->m_pkthdr.len - progress; 1214 if (length > MCLBYTES) 1215 length = MCLBYTES; 1216 1217 if (m_new == NULL) { 1218 if (length > MLEN) 1219 m_new = m_getcl(how, MT_DATA, 0); 1220 else 1221 m_new = m_get(how, MT_DATA); 1222 if (m_new == NULL) 1223 goto nospace; 1224 } 1225 1226 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1227 progress += length; 1228 m_new->m_len = length; 1229 if (m_new != m_final) 1230 m_cat(m_final, m_new); 1231 m_new = NULL; 1232 } 1233 #ifdef MBUF_STRESS_TEST 1234 if (m0->m_next == NULL) 1235 m_defraguseless++; 1236 #endif 1237 m_freem(m0); 1238 m0 = m_final; 1239 #ifdef MBUF_STRESS_TEST 1240 m_defragpackets++; 1241 m_defragbytes += m0->m_pkthdr.len; 1242 #endif 1243 return (m0); 1244 nospace: 1245 #ifdef MBUF_STRESS_TEST 1246 m_defragfailure++; 1247 #endif 1248 if (m_final) 1249 m_freem(m_final); 1250 return (NULL); 1251 } 1252 1253 #ifdef MBUF_STRESS_TEST 1254 1255 /* 1256 * Fragment an mbuf chain. There's no reason you'd ever want to do 1257 * this in normal usage, but it's great for stress testing various 1258 * mbuf consumers. 1259 * 1260 * If fragmentation is not possible, the original chain will be 1261 * returned. 1262 * 1263 * Possible length values: 1264 * 0 no fragmentation will occur 1265 * > 0 each fragment will be of the specified length 1266 * -1 each fragment will be the same random value in length 1267 * -2 each fragment's length will be entirely random 1268 * (Random values range from 1 to 256) 1269 */ 1270 struct mbuf * 1271 m_fragment(struct mbuf *m0, int how, int length) 1272 { 1273 struct mbuf *m_new = NULL, *m_final = NULL; 1274 int progress = 0; 1275 1276 if (!(m0->m_flags & M_PKTHDR)) 1277 return (m0); 1278 1279 if ((length == 0) || (length < -2)) 1280 return (m0); 1281 1282 m_fixhdr(m0); /* Needed sanity check */ 1283 1284 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1285 1286 if (m_final == NULL) 1287 goto nospace; 1288 1289 if (m_dup_pkthdr(m_final, m0, how) == 0) 1290 goto nospace; 1291 1292 m_new = m_final; 1293 1294 if (length == -1) 1295 length = 1 + (arc4random() & 255); 1296 1297 while (progress < m0->m_pkthdr.len) { 1298 int fraglen; 1299 1300 if (length > 0) 1301 fraglen = length; 1302 else 1303 fraglen = 1 + (arc4random() & 255); 1304 if (fraglen > m0->m_pkthdr.len - progress) 1305 fraglen = m0->m_pkthdr.len - progress; 1306 1307 if (fraglen > MCLBYTES) 1308 fraglen = MCLBYTES; 1309 1310 if (m_new == NULL) { 1311 m_new = m_getcl(how, MT_DATA, 0); 1312 if (m_new == NULL) 1313 goto nospace; 1314 } 1315 1316 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1317 progress += fraglen; 1318 m_new->m_len = fraglen; 1319 if (m_new != m_final) 1320 m_cat(m_final, m_new); 1321 m_new = NULL; 1322 } 1323 m_freem(m0); 1324 m0 = m_final; 1325 return (m0); 1326 nospace: 1327 if (m_final) 1328 m_freem(m_final); 1329 /* Return the original chain on failure */ 1330 return (m0); 1331 } 1332 1333 #endif 1334 1335 struct mbuf * 1336 m_uiotombuf(struct uio *uio, int how, int len, int align) 1337 { 1338 struct mbuf *m_new = NULL, *m_final = NULL; 1339 int progress = 0, error = 0, length, total; 1340 1341 if (len > 0) 1342 total = min(uio->uio_resid, len); 1343 else 1344 total = uio->uio_resid; 1345 if (align >= MHLEN) 1346 goto nospace; 1347 if (total + align > MHLEN) 1348 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1349 else 1350 m_final = m_gethdr(how, MT_DATA); 1351 if (m_final == NULL) 1352 goto nospace; 1353 m_final->m_data += align; 1354 m_new = m_final; 1355 while (progress < total) { 1356 length = total - progress; 1357 if (length > MCLBYTES) 1358 length = MCLBYTES; 1359 if (m_new == NULL) { 1360 if (length > MLEN) 1361 m_new = m_getcl(how, MT_DATA, 0); 1362 else 1363 m_new = m_get(how, MT_DATA); 1364 if (m_new == NULL) 1365 goto nospace; 1366 } 1367 error = uiomove(mtod(m_new, void *), length, uio); 1368 if (error) 1369 goto nospace; 1370 progress += length; 1371 m_new->m_len = length; 1372 if (m_new != m_final) 1373 m_cat(m_final, m_new); 1374 m_new = NULL; 1375 } 1376 m_fixhdr(m_final); 1377 return (m_final); 1378 nospace: 1379 if (m_new) 1380 m_free(m_new); 1381 if (m_final) 1382 m_freem(m_final); 1383 return (NULL); 1384 } 1385